Now that most marketing teams are onboarding Claude this can finally be solved.

Everyone is running in singleplayer mode though. All their skill files, brand guardrails and historical campaign results are floating everywhere.

Marketing teams must go from singleplayer AI → multiplayer AI to solve this properly.

A few months ago I wrote about how future marketing teams will be living inside of GitHub. This is the concept that all your important marketing files (brand files, guidelines, historical campaign test log) live in a shared repository. It ensures that everyone is running on the same path.

Lately I’ve been thinking more about how to solve this in a lighter weight approach if GitHub seems too technical or advanced for your needs.

If I was a lean marketing team of 3-5 and didn’t want to go through a technical GitHub setup, I’d use Notion. I’m going to start by giving a quick primer on the GitHub shared repo, what it solves, and how to stand up a similar solution on Notion.

If you prefer to see me demo a Notion multiplayer team on video, I recently joined The Quest podcast to chat about it.

Okay, let’s get into it.

–

Tools I actually use and recommend

If I wouldn’t recommend it to you without getting paid, it doesn’t go in this section. Want to sponsor The AI Marketer? Email me at jon@jon4growth.com

Riverside is what I use to record, caption and edit all my startup’s case studies. Their platform saves me so much time in not needing multiple tools to do all of this. Get 1 month of Riverside Pro for free with my code: CLAUDEMARKETERS

Scrunch is the only AEO tool I’ve tested that doesn’t just measure your AI visibility, it fixes it by serving AI-readable pages straight to the bots.

–

A primer on the GitHub solution

The idea with the GitHub solution is to create a repository that the entire marketing team loads inside of their Claude instance. This solves for teams that are struggling with scattered Claude skills stored on local devices, campaign learnings spread across sheets, etc. It houses it all in one place that an entire team can access.

The GitHub repo looks a little something like this when set up:

The marketing exec on the team creates this repo, loads in all relevant brand/ skill files, and then distributes across the team. This keeps the team on the same page, and allows for them to continuously upload new files and campaign learnings to it.

Some honest limitations with GitHub include 1) longer onboarding for those that aren’t technical, 2) everyone needs to load new repo into Claude when files get updated and 3) must have the repo folder selected in Claude Code when using it.

This is where the lightweight Notion approach comes to the rescue.

Share

Why Notion for multiplayer AI

It’s very similar to the GitHub solution to going multiplayer with your marketing team, without the technical headache and much less setup time. Some of the advantages include:

1. Houses folders & files in a very structured way

2. Claude has a native Notion MCP connector

3. Minimal onboarding time for entire team

4. Many teams already using Notion

5. Add tables into pages for campaign tracking

6. Claude always uses latest Notion page info

If these advantages have convinced you to run your team out of Notion, keep reading to learn how easy it is to build.

How to build your Notion marketing OS

Step 1: Create a Teamspace

In Notion, create a new Teamspace and name it “Marketing Team OS.” This is what gets shared with the entire team once it’s done.

The main Teamspace page should briefly explain how everything works, what lives in each sub-page, and who owns what. Assign a DRI (directly responsible individual) per page. Without ownership, pages go stale.

Step 2: Build these core pages

I won’t dive into each page (there are 12) so I don’t clutter this essay with dozens of screenshots, but if you’re interested, I’ve made my demo OS Notion Teamspace available here.

I highly suggest starting your OS with the following 3 pages I’ll be showing: Marketing Context, Experiments & Learnings and Skill Library. Let’s begin with the Marketing Context page.

Step 3: Populate Marketing Context first

Imagine being in a leadership meeting where it’s decided that a new ICP will be added to the GTM strategy. Instead of having everyone update their Claude instance, the Marketing Context page gets updated. Now, when everyone loads into Claude to run any new marketing campaigns, it’ll be using the latest info about the new ICP.

This is the highest-leverage page. Every marketer’s Claude instance automatically has the latest context the next time they run a prompt.

Now, let’s jump to how Experiments & Learning page looks:

Step 4: Backfill Experiments & Learnings

This is the page that solves the Postmates problem; it sucked because we’d sometimes run the same exact campaign test twice. Maybe a marketer that had left the company had already run the test last year. Or maybe we didn’t sync about THAT exact campaign.

With this tracker, everything gets logged and instantly gives anyone access into what worked or failed in the past. You can then ask Claude questions like:

• What are the best performing headlines in the last year?

• How much have we spent on influencer marketing tests?

• Who on the team has had the most test wins?

• How are we pacing to reach our goals?

• Which channel is being least tested this quarter?

I asked the first question to Claude and here’s what I got:

Now you can simply ask Claude about historical campaign results rather than hunting down dozens of doc links or waiting for a team to reply on Slack.

Since Claude can write directly into Notion, marketers can also prompt it to log new campaign results from the chat window. There’s no excuse not to log.

And since I gave the limitations for the GitHub approach, I also need to provide the other side of the coin. For Notion setup, you will have 1) no version history with changes like GitHub has and 2) no review step before pushing changes to files/ folders.

Step 5: Set up the Skill Library

The one other important item, outside of constant campaign logging, is ensuring that everyone on the team is using the latest .md files in their Claude instance.

As a quick primer: Skills are .md files that your local Claude instance uses to do tasks with the right context each time. This is incredibly useful so that you’re not having to repeat yourself or give Claude context on repetitive tasks.

Skills live in Notion in the Skills page (the editable source, one page per skill, human-readable) and each person’s local machine (the actual .md files their Claude reads to do the work). To bridge the two, you need to create a plugin which is a packaged bundle of the skill files plus a reference to the team’s Notion connector.

To set this system up:

1. Create a plugin .zip file (contains all skill files with reference to Notion connector)

2. Everyone on the team drags this zip to Claude and prompts “Upload this plugin and copy CLAUDE.md from the marketing-os-skills plugin into the root of this project folder.”

3. Everyone on team prompts “Update my local Skill files from every page in the Notion Skill Library” whenever updates are made

To edit the Skills, you can either edit it directly inside Notion or ask Claude to do it and update Notion for you. When anyone on the team updates their Skill files, they’ll have the latest edited version.

Note: If your team is on the Team/ Enterprise Claude account, you can actually have the plugin file in the org-wide catalog.

Instead of everyone on the team running with outdated Skill files, this now keeps everyone on the same page using the latest rules and context.

Step 6: Connect Claude to the Teamspace

Once you’ve got your Teamspace created, use Claude’s Connectors and hook it up to Notion. Make sure you give it access to the Teamspace created. Same goes for everyone in the marketing org.

It shouldn’t take longer than a few minutes to get everyone hooked up.

That’s the beauty of this Notion setup. Stupid simple for your team to get started.

The part most teams get wrong

What’s important after onboarding everyone, is that a culture of constantly logging campaigns is instilled. If everyone connects to this OS and it remains stale, the entire system breaks.

I highly suggest having an internal champion that maintains a weekly cadence of ensuring that the OS is up to date. This is typically the marketing exec or marketing lead.

A few early marketing teams I’ve seen running this in the wild will announce in their team meetings to keep everything updated. If you keep pushing it and everyone has that “aha moment” of value and time saved in the long run, you’ve won. The entire team will continuously update.

I’m hoping I’ve convinced you of all the benefits you’ll reap from going multiplayer AI.

To get started, do this today (half an hour):

1. Create a Teamspace called Marketing Team OS

2. Build three pages: Marketing Context, Experiments & Learnings, Skill Library

3. Add your brand’s ICP and voice into Marketing Context

4. Backfill your last 10 campaigns into Experiments & Learnings

5. Connect Claude → Notion, grant Teamspace access. Repeat for your team

6. Ask Claude a question based on what’s inside your Teamspace

This’ll get you to an “aha moment” immediately.

From there, designating individuals on the team to build the rest of the pages, backfill campaign data and maintaining the pages will put you on the right track.

Again, for a shortcut of the demo Teamspace, it’s linked here. You can duplicate it and start from there. Happy building, my fellow AI marketers.

–Jonathan

Subscribe now

I’ve run into a few instances where a SaaS product or site I use doesn’t connect to Claude. There isn’t an MCP and I’m stuck back to doing things manually. It’s a quiet killer of productivity on tasks that I still have to do every week.

That’s now changed. Yesss. Claude just released an ability to create skills simply by recording. I tried it on a Substack import I’ve been doing by hand for months. It took a few minutes to record and now it runs on a schedule without me touching it.

Here’s why I’m writing this: most marketers I talk to don’t know this feature exists yet. And the use cases for Marketing Ops are fairly significant, especially if your stack has gaps where MCPs don’t exist yet.

Note: I also filmed a “how to” lesson on Claude Marketers if you prefer to learn visually.

Alrighty, let’s dive in!

Share

–

Tools I actually use and recommend

If I wouldn’t recommend it to you without getting paid, it doesn’t go in this section. Want to sponsor The AI Marketer? Email me at jon@jon4growth.com

Scrunch is the only AEO tool I’ve tested that doesn’t just measure your AI visibility, it fixes it by serving AI-readable pages straight to the bots.

GoDaddy’s Airo for WordPress helps keep your site growing after launch. Publish posts, refresh pages and update products with a simple prompt, right inside WordPress. Explore at no cost with a 7-day trial.

–

The 3 Step Recording Workflow

1. Identify the manual task

Start with something you do at least weekly that follows the same steps every time. This can be something like an export, an import, a routing action, a QA check. Literally anything that’s repetitive. If you’re doing it the same way twice, Claude can probably learn it.

2. Hit “Record a skill” and walk Claude through it

In the Claude desktop app, open the Skills panel and click “Record a skill.” From there, just perform the task the way you normally would. Claude watches what you click, the file handling, the copy-paste steps, etc and builds the skill from your actual workflow. You just do the manual task while Claude records.

3. Schedule it

Once Claude confirms the skill, you can set it to run automatically on a daily, weekly, or custom schedule. This is where the massive leverage is. The few minutes you spent recording becomes zero minutes going forward.

A few marketing use cases where this works well:

- Sending ad platform invoices to finance on Slack
- Scheduling posts on LinkedIn or other platforms without a native MCP
- Importing users as newsletter subscribers (more on this below)
- Lead routing from a form export
- Weekly competitor teardowns with screenshots
- New client onboarding steps
- Email render checks
- Creative QA before launch

This feature works especially well for ops tasks your team is currently doing by hand. Any task that would typically fall through the cracks because they’re too small to build a full integration for, but too frequent to keep ignoring.

How I used it this morning (real example)

Every week, I was going into the Claude Marketers admin portal, exporting a CSV of users who’d recently signed up, and importing them into Substack. Manual, repetitive, 2-5 minutes every time and I’d been putting off fixing it because Substack doesn’t have an official MCP with Claude. Bummer. Until today.

This morning I hit “Record a skill” and just walked Claude through the process I’d been doing manually. Export the file, open Substack, upload the CSV, confirm the import. Claude watched the whole sequence. And while I recorded, I narrated the steps to provide even more context on how I was doing everything.

It’s really as if demonstrating something to someone sitting right next to you.

After a couple minutes, Claude asked a couple of clarifying questions and then I scheduled it to run daily. A task I’d done by hand every day for months automated. An honest limitation is that this won’t work well on sites that have anti-bot protections or when the steps change often.

If you’re not sure what to automate first, create a running note this week and log every manual task as you hit it. By Friday, I promise you’ll have a list. Pick the ones you do most often and record them on Claude.

Last month I bought a last-minute flight from LAX to NYC to attend CommerceNext, days before it started. This was a recipe for disaster as it gave me zero time to prepare. In the spirit of being an AI marketer, I thought: how can I still set up quality meetings with GrowthPair’s ICP?

That conference sprint is one of three B2B enrichment workflows I’ve now run with Claude. I’ll walk through all three: what I did, how I set it up, and the exact prompt I used on the conference list.

All you need is a list of emails, Claude, and some iced cold brew (personal preference) to run these starting today.

–

Tools I actually use and recommend

If I wouldn’t recommend it to you without getting paid, it doesn’t go in this section. Want to sponsor The AI Marketer? Email me at jon@jon4growth.com

Riverside beats all other platforms I’ve used to record case studies with clients. I record, caption, and edit clips all with their easy-to-use UI. Get 1 month of Riverside Pro for free with my code: CLAUDEMARKETERS

Scrunch is the only AEO tool I’ve tested that doesn’t just measure your AI visibility, it fixes it by serving AI-readable pages straight to the bots.

–

Workflow 1: Conference attendee scoring

What you need: A conference attendee CSV, Claude with your ICP loaded in a claude.md file.

Alright, let’s dive into the conference blitz process I went through.

Most B2B conferences provide an attendee list prior to the first day, but it’s often thousands of rows and very barebones (name, position, email).

Instead of going through each row manually, I downloaded the CommerceNext list and fed it into Claude to do the following:

1. Score each contact on how well they fit my ICP

2. Search contact on LinkedIn and provide direct URL

3. Create a copy/ paste message I can send when connecting

Since my Claude instance already has a claude.md file with GrowthPair’s ICP, I didn’t need to repeat it. All I did was upload the list and write a prompt like this:

“Given the following conference list of attendees, score each attendee across my ICP and filter out those that don’t fit. Then find each attendee on LinkedIn and craft a personal quick DM that I can copy/ paste when connecting with them. The goal is to set up coffee with them while I’m at the conference for a quick chat.”

Within a few minutes, I had a target list of 200 out of thousands initially on the list. I took that and went to work on LinkedIn. It’s important to note that I don’t automate anything on LinkedIn as it can risk account suspension (not worth it imo).

Share

Workflow 2: Newsletter subscriber enrichment

What you need: Your subscriber list exported as a CSV, Claude, Apollo connector configured in your Claude environment.

As a newsletter creator, I don’t have the best visibility into who my readers are. Aka you, yes you reading this.

And as a founder of a B2B startup, having this info is valuable to understand whether I’m reaching my ICP and even allows me to spin up warm convos with the most engaged. Don’t worry though, I’m not going to infiltrate your DMs. My goal with this newsletter is to teach AI marketing to thousands, first and foremost.

I realize that not every company has a newsletter, but I can’t recommend it enough. You own the audience. You’re not beholden to algorithm changes on social feeds. It’s the best form for reaching your ideal customer.

Similar to the conference list, I fed my entire subscriber list into Claude and asked it to:

1. Match each subscriber to a professional profile using Apollo’s connector

2. Enrich subscriber with info on their company, company size and role

3. Score each enriched subscriber to my startup’s ICP (1-100)

4. Provide a list with top matches and ICP match percentage

It’s important to note that not all emails are able to be enriched, especially if subscribers are using personal email addresses. It’s much harder for Apollo or similar enrichment providers to match without a company email.

The cherry on top about this exercise was that I got a great sample on all my readers to be able to share with prospective partners. I’m now able to provide insight into where my readers work and include slides like this in the partnership deck:

Look at that list of top-tier companies that are reading. You’re in good company (pun intended) ;)

Workflow 3: Website visitor triage in Slack

What you need: RB2B ($49/mo) feeding a Slack channel, Claude connected to that Slack channel.

There are dozens of tools that can de-anonymize website visitors. I personally use RB2B on my startup’s site for $49/mo and it feeds visitor info directly into a Slack channel. It’s amazing.

The big issue I found was that my team wasn’t really doing much with it, especially on days where we had huge website traffic. It became real difficult to keep up with all the Slack pings and sift through the noise.

Instead of manual sifting, I set Claude up to take all the website visitors, score them to our ICP and then feed us only with profiles we care about.

The through-line across all three is that most B2B companies are sitting on data they never act on because the signal-to-noise ratio makes it impossible to work manually. Claude doesn’t get tired, lucky for us.

If your team is getting Slack pings, CSV exports, or subscriber lists and doing nothing with them, start here.

Subscribe now

It was the same problem as paid ads: the data lived in one place, the copy in another, the reporting in a third. Tons of back and forth between tools, team and dashboards, along with tedious work to create reports.

A few months ago I wrote about The AI-Native Paid Ads System. It became one of my most-read essays, and the most common question I got was: “can you do this for email?”

The answer is yes. And frankly, this system can apply to pretty much every marketing pillar.

Before I explain it, here’s an overview graphic:

We’ll start with the first layer, and work our way down.

Layer 1: Connectors & data sources

Here’s where email differs from paid ads. Meta built an official connector, so that essay was plug-and-play. Most email platforms haven’t done that yet, and this is the part people assume blocks them.

It doesn’t. If your tool has an API, you can connect it to Claude. I use Loops (loops.so) for email, and I connected it directly via their API. They also have a dedicated page that most Email Service Providers (ESP) will likely create describing their setup with tools like Claude.

Here’s an example of an ESP that does have a native Connector already inside of Claude:

The point isn’t Loops. The same move works for Klaviyo, Customer.io, Beehiiv, whatever you’re on. The pattern is: grab your API key, send it to Claude, and ask it to wire itself up.

Once connected, the system continuously gets stronger with:

• Campaign and flow performance data from your email platform

• Subscriber events and segment data

• Scheduled tasks running competitive email research

That last one is my favorite. For paid ads I track competitor websites weekly for offer and positioning adjustments. For email, you can do that as well, along with two other really neat things to gather competitive intel:

1. Create a seed inbox and sign up for your competitors’ emails

2. ReallyGoodEmails scrape of your competitors’ emails

You can use Claude’s Gmail connector to log all competitor subject lines, offers and send cadence into a table inside Notion. And for ReallyGoodEmails, Claude can use the Chrome extension to do a regular scrape on their database.

Every competitor email now becomes a data point that will feed into subsequent tests.

If you’re using Claude Cowork, you can set this up through scheduled tasks or on Claude Code using routines. This is such an underrated ability within Claude.

Layer 2: Claude skills and claude.md file

Your claude.md file is crucial not only for this system, but also for anything else you’re running within marketing. It gives Claude a pre-read of your brand’s ICP, tone of voice and guardrails. When you launch up Claude to do anything, it starts with this file as the seed which makes your usage much more efficient.

Then build a skill specifically for email best practices. Whenever you mention anything about email campaigns, it’ll call this skill up to use as a basis for the task you’re looking to accomplish. There isn’t a limit on the amount of skills you add to Claude, but I’d suggest consolidating to one skill for email marketing. The best part is that Claude can build it for you with a prompt similar to this:

I’d like to build a skill for email marketing: campaigns, flows, deliverability, list
 management, copy and design. Research the best minds in email marketing and add all
 their best practices into this skill. This skill will be used whenever I’m writing
 emails, analyzing campaign performance, building flows, or auditing my email program.

Running the email audit

Once you have your connection, claude.md file and skill loaded, run the audit:

You’re auditing my email marketing program. Use the [Loops] API connection to pull live 
data directly. Don’t ask me to paste numbers or screenshots.

Step 1: Confirm the account.

Confirm which workspace/account you’re connected to, then move on.

Step 2: Pull the data. For the last 30 and 90 days, get:

- List health: total contacts, growth rate, churn/unsubscribe rate, sunset candidates 
(no opens in 90+ days)

- Campaign performance: open rate, click rate, unsubscribe rate, spam complaints per 
send and trending

- Flow/automation performance: every active flow, conversion per step, where people 
drop off

- Deliverability signals: bounce rates, spam complaint rate, anything that looks 
like inbox placement trouble

- Segment performance: which segments actually engage vs. which I’m wasting sends on

- Send cadence: how often each segment hears from me, gaps and pileups

Step 3: Write the audit. Structure it like this:

1. Verdict: One paragraph. What’s actually going on in this email program. Be blunt.

2. List health: Growth vs. decay, dead weight, sunset policy gaps.

3. Deliverability: Bounce, complaint, and engagement signals. Am I headed for the 
spam folder?

4. Campaigns: What’s winning, what’s flopping. Subject line and content patterns. 
Cite the numbers.

5. Flows: Which automations print money, which leak. Where people drop off and why.

6. Segmentation & cadence: Who I’m over-mailing, who I’m ignoring, segments I should build.

7. Top 5 actions this week: Specific, prioritized by impact. “Improve your subject 
lines” doesn’t count. Tell me which email, which segment, what change, and why.

Rules:

- Cite real numbers from the data. No vague claims.

- If data is missing, broken, or looks off, call it out.

- Skip corporate hedging. If something is broken, say so.

- End with the 5 actions. That’s what I’ll actually do on Monday.

I ran an audit on the Claude Marketers academy emails I’ve been sending through Loops using this exact prompt and it produced a thorough analysis:

This analysis should be added in as a routine with Claude Code so you have a recurring deep-dive on your latest campaigns. It provides immediate feedback for the team and also feeds right back into a performance database for Claude to draw from.

Subscribe now

Claude Design for bespoke email

With the launch of Claude Design, we now have the ability to create beautiful, on-brand HTML templates using our words. I tested this a bit to spruce up the GrowthPair email templates and then imported the HTML designs into beehiv. They turned out much nicer than our generic templates.

What’s nice about creating designs in Claude Design is that you can either copy the HTML, or share it straight to Claude Code by hitting the “share” button in the top right corner. Makes it super easy to port designs over.

I’ve noticed that it’s best if you have very specific requirements on designs, similar to how important prompting Claude is on anything else. In my case, I specifically ask for a few template variations to choose from that are easily editable.

Layer 3: Claude’s live artifacts and widgets

With your email platform connected, build a live dashboard with artifacts: campaign performance, flow health and list growth that pulls fresh data every time you open it. I wrote a more in-depth essay on building these dashboards in minutes, and the exact same approach works here.

Since Loops doesn’t have an active MCP and limited API access, I had to default to building my dashboard using Claude Code’s widgets. This’ll get better as more ESPs launch native connectors into Claude. In my mind, if you’re a marketing tool and don’t have an MCP within the next 6 months, you’re basically dead. Maybe a spicy take, but I think very true.

Layer 4: Learning loop with GitHub

This is the most important layer to ensure your system saves all campaign data and continues to improve over time. Every subject line test, send time experiment, and segment learning goes into a shared GitHub repo so your entire team and Claude have context on historical performance.

This becomes your source of truth for everyone on the team so that everyone is on the same page and has the same context to work from. It never forgets the emoji subject line test, or Black Friday offer you ran 18 months ago.

The setup steps are identical to the paid ads essay (create folder → Claude Code → private GitHub repo → add teammates), so I won’t repeat them. If you built the paid ads repo already, don’t make a separate one. Instead add an /email folder. This is the whole thesis: marketing functions converging into one repo your team and Claude both work from.

This is the future of email marketing

I’m a strong believer that the future of running email campaigns will happen inside a chat window like Claude Code. What’s nice is that once you have Claude connected to your ESP, you have the ability to not only gather data, but also create new campaigns and tests. It’s as simple as asking Claude to “run 5 headline tests on our onboarding series every week.”

Because Claude has all of the data on campaign performance, it can run an endless loop of iterations that are constantly improving itself.

I don’t want to sugarcoat that this will be an easy setup. It will require time, refining and someone on top of the system to ensure it continues running smoothly. Will it be worth the effort though? Yes, and I’d bet my entire growth career on it.

Subscribe now

TLDR – how to action on this

1. Connect your email platform: official connector if it exists, API if it doesn’t. Loops took me 15 minutes to set up without an official connector.

2. Build claude.md and an email skill: every prompt gets smarter when these are loaded.

3. Run your first email audit: use my exact prompt above. See what it catches that you missed.

4. Ship one test from the audit: take its #1 action and run it this week.

5. Build a simple dashboard: live artifact or widget pulling your campaign and flow data.

6. Schedule competitor inbox research: one automated task tracking your top 3 competitors’ sends.

7. Add email learnings to your GitHub repo: weekly cadence, same repo as your ads learnings.

Until the next one,

Jonathan

About a month ago I got off the waitlist for the beta of ChatGPT ads, so you know I zoomed straight into my first campaign creation.

I’m not claiming to be an expert since this channel is bonkers new, but I’d like to share initial thoughts, the strategies I’m testing and where I think the future of LLM ads are going.

Share

Step 1: Campaign Foundation

The actual campaign creation flow is stupid simple. I expected this as there isn’t going to be all the fancy nuts and bolts for a new channel at the start. You have the option to bid on Reach or Clicks (Conversion bidding coming soon). With the Reach objective, you’re bidding on a CPM basis, and with Clicks it’s on a CPC basis.

For location, you can set this at a country level, or use actual cities. I notice that most cities default to a DMA, meaning they’re targeting a cluster of cities. For example, the San Francisco DMA also includes San Jose, Oakland and a few other major cities in the Bay Area. Here’s a rough look at what the DMA includes (ft. my scribbles on Google Maps):

There’s an option to add a Conversion Event, but again, this isn’t something you can bid on just yet as it hasn’t launched.

Step 2: Ad Group Settings

Within the ad group settings, you can set your CPC bid (Clicks bid type), website URL, along with “Context hints” which are essentially your keyword themes.

I want to take a minute to give my thoughts on this as it’s so important to how search is changing. We don’t have the ability to match for specific keywords, and that’s by design because of how user behavior is adjusting.

Google themselves have gotten ahead of this years ago by loosening what the exact match keyword type would trigger on campaigns. On Google’s website, they show that someone searching for “furniture store” or “home furniture shop” would bring up your ads with the same exact match keyword.

People are no longer searching “running shoes” and instead searching in sentences like “what are the best running shoes for hiking in rough terrain in the winter months” on ChatGPT.

This is why ChatGPT needs keyword themes known as “context hints” to display your ads.

From a strategy standpoint, you should be thinking about segmenting your ad groups by keyword clusters. These clusters can either be centered around intent, product type, service line, ICP you’re targeting, etc. I feel like there’ll be many variations to do this successfully until OpenAI eventually automates and gets amazing at targeting, similar to Meta nowadays.

Here’s an example of a poor “context hint” in your ad group:

“AI notetaker tool”

Compared to a strong “context hint” in your ad group:

“Executives that are evaluating AI notetakers for their growing organization and need a central resource for team meetings, sales calls, and customer calls.”

The strong version does a much better job of describing your ICP, their problem and the solution they’re looking for.

An easy way to think about writing out these “context hints” is to use the following structure:

ICP/ persona + problem + solution they’re looking for

When thinking about the type of phrase to write, also keep in mind the buyer’s journey when they’re searching on ChatGPT. They’re likely past the discovery stage, understand their problem and are actively looking for a solution. This is why LLMs will be great for middle of funnel and bottom of funnel traffic.

This should all be tested and validated with data, but it’s always good to keep in mind where your buyers are in their journey when on a specific platform.

Step 3: Creative + Conversion Tracking

For the final step before launching your first LLM ad, add your URL, headline, description and ad image.

It’s important to note that copywriting for you ads should be much different on ChatGPT versus traditional Google search.

Google: AI Notetaker Tool | Only $9.99/ Month

ChatGPT: Haven’t found the perfect AI notetaker tool yet?

On ChatGPT, the description is capped at 100 characters and should read like a casual conversation. Remember: users are having a conversation with ChatGPT and you want your ad to feel as organic as possible in their chat window.

As I was writing this, OpenAI launched conversion tracking, so I jumped right into it. It’s a little manual of a setup, but here’s how I ran it:

1. Created a Data Source and pasted setup code in header section of website

2. Created Conversion Event to track booked discovery calls

3. Added “event call” code on the “thank you” page so that it fires on each load

It’s not as simple as tagging events on Meta or Google, but it gets the job done and isn’t the most difficult. I believe this’ll get easier as time goes on. If you’re wondering where to pull the event call code, click “View Code” for your main website tag, and you’ll see it in there. Super important to note that you’ll need to edit the event call code with your actual event name.

One big current limitation is that you can’t actually bid on Conversions yet. You solely have the ability to track them, which is still a huge win.

Where LLM ads fit in your stack

It’s still the first inning with LLM ads but OpenAI is giving us the first taste with ChatGPT ads. My main takeaways after running these ads for a month on what a sound strategy could look like (FYI, not an expert.. yet):

1. ChatGPT ads for middle of funnel and bottom of funnel. Consumers will likely be comparing options and making purchasing decisions. This is in contrast with a channel like Meta that can be used for the top of funnel (discovery) stage.

2. Keywords will be broad by nature → will lead to impression spread. Ensuring that you’re using the right keyword themes is more important than ever.

3. LLM conversation → dedicated landing page will crush it. Imagine comparing hiking shoes on ChatGPT, and you then land on a page with the comparison breakdown. The right ad leading to the right landing page matters more than ever.

4. Full-funnel strategy is still crucial and LLM ads are just one more piece to the puzzle, in combination with your other channels, organic content, retargeting, etc

Real performance data from month one

I’m being told by colleagues that are live with ChatGPT ads that we’re being throttled by OpenAI. I’ve increased my bids from $2 → $5 → $10 and still only getting dozens of impressions daily.

But we do have an idea on the CPC for my keyword themes centered around “hiring marketing talent” of around $6.65. To compare, this is about ¼ the cost we pay on Google for the same keywords, so the CPC is dirt cheap. Of course, we’ll need to keep an eye on conversion rates to see how they differ from other channels.

What to do this week

1. Get on the ChatGPT ads waitlist if you haven’t already

2. Map your current Google keyword clusters to ChatGPT “context hints”

3. Add conversion pixel to your site and map important events

4. Launch your first ads!

This is early days, but the CPCs are too good to ignore. Will report back with conversion data as I get more volume.

Until the next one,

Jonathan

Subscribe now

It’s called Content Machine 2000. I built it in a weekend with Claude Code and zero engineering background. It runs the entire content side of my business, from ideation to analytics to outreach, and it replaced four SaaS subscriptions in the process.

This is a walkthrough of what it actually does and how I use it day-to-day. If you’re a creator or founder publishing regularly, you’ll walk away with a blueprint you can copy, modify, or steal outright.

Why I built it

The short version: I tried Taplio. I tried the alternatives. None of them fit how I actually research, draft, and iterate. So I went manual, spending 45 to 90 minutes a day scrolling LinkedIn and X, bookmarking posts, and pattern-matching by hand. When you’re publishing 5 times a week, that math gets ugly fast.

The bigger problem wasn’t the time. It was that staring at a feed hoping for inspiration is a terrible ideation process. I needed something that surfaced patterns from real engagement data, not whatever the algorithm wanted to show me that morning.

Eight hours of focused work later, four for the MVP, four more for polish across a week and I had a tool that does more than any SaaS I’d ever paid for.

What Content Machine 2000 actually does

It’s seven features, all running locally on my machine, talking to LinkedIn through Apify’s API (which runs scrapers on their servers).

1. Morning briefing

The first screen I see every day pulls my engagement stats from the last 24 hours and the trailing week, surfaces my top three performing posts, and flags content pillars I haven’t covered recently.

Think of it as a daily standup with myself, except the data is already on the screen before I sit down with coffee.

2. Research scraper

Finds the top-performing AI and marketing posts on LinkedIn from the past 24 to 72 hours. I see what’s resonating in my niche before I open the app which means I’m not making decisions based on whatever the algo decides to feed me.

The key word here is signal. The scraper sorts by real engagement, not virality theater. I’d rather see a 200-comment thread from a peer I respect than a 50,000-impression post from a guru clipping someone else’s idea.

Share

3. Trend tracker

This clusters topics getting the most engagement in my niche and surfaces the themes that are picking up momentum. Recently it flagged “job market disruption and the junior marketing crisis” which was a topic that wasn’t on my radar at all.

I wrote a post about junior marketers off the back of that signal. It beat my typical impression count (I usually land 2,500 to 3,500 per post). I wouldn’t have come up with the angle on my own.

That same trend sparked a second post (“Can we stop AI bro marketing?”) which was a take on people claiming they’ve replaced their entire marketing team with one Claude subscription. The trend tracker connected something the data was showing with something I’d already been noticing.

4. Lookalike content generator

This takes my top-performing posts and generates new angles on them.

If a post about hiring globally hits big, the tool doesn’t spit out a rewrite. It surfaces five adjacent angles: the operations side, the cost side, the talent quality side, the timezone side, the cultural side. I pick the one I have a real opinion on and write that one myself.

5. Narrative arcs

Probably the most useful feature in the whole tool, and the one I’ve never seen in any off-the-shelf product.

It tracks my content pillars (the recurring themes I write about) and shows how long it’s been since I last posted about each one. Right now my dashboard reads something like:

• Claude Marketers Academy: 8 days

• GrowthPair: 33 days

• AI tools for marketers: 4 days

• Personal/founder lessons: 12 days

• Hiring & talent: 19 days

I rotate between 3 to 5 pillars a week based on that data. It prevents fatigue on any single topic and keeps me from accidentally ghosting a pillar for a month. The repost feature ties into this: I used the narrative arc to time a repost about Claude Marketers that hit 101,730 impressions because the data showed it had been long enough since I’d touched the topic.

This is the kind of feature that only exists when you build for yourself.

6. Content recycler

Identifies past top performers that are worth reposting. Reposts typically pull about 80% of the original engagement on my account, which means my best thinking compounds instead of evaporating from the feed after 48 hours.

For a creator publishing daily, this is gold. Half the battle is realizing the post you wrote four months ago is better than anything you’d write today on the same topic.

7. Engagement scraper

Finds people engaging with content in GrowthPair’s ICP: VPs of marketing, founders, growth leads at venture-backed companies. I get a list of warm signals every morning that I can then DM.

This feature is still a bit experimental but going to be testing some outbound with those that engage with my content. It’s definitely much more warm once they’ve seen my content, rather than a cold pitch.

A morning with Content Machine 2000

Here’s what a typical Monday looks like.

6:30 AM. I sit down with coffee, open the dashboard. Morning briefing tells me yesterday’s post landed at 4,200 impressions, my GrowthPair pillar is at 33 days (overdue), and the trend tracker flagged two new conversations picking up steam in the AI marketing space.

6:40 AM. I pull from the lookalike generator. My top post from last week was about AI hiring, and the tool surfaced an angle about how junior marketers are getting squeezed. I write a draft, and publish it on LinkedIn.

7:00 AM. Post goes live. I respond to comments for 20 minutes.

9:00 AM. I check the engagement scraper. Three names from real target companies showed up. I send two short, specific DMs about their comments on the post.

Total time spent on content for the day: ~20 minutes. Before the tool, it was closer to 3 hours, most of it scrolling.

What I deliberately didn’t build

This matters as much as what I did build.

Post generation: Content Machine 2000 doesn’t write a single word that goes on my LinkedIn. I prototyped a “draft the post for me” feature early and killed it inside a week. The value of thought leadership collapses the moment the thinking isn’t yours. The tool makes me faster at finding what to say but doesn’t decide what I say.

Auto-posting and auto-scheduling: Programmatic posting carries real risk of LinkedIn flagging the account. Manual posting takes 30 seconds and isn’t worth the tradeoff.

A “send me notifications” layer: I want this to be a place I go rather than a thing that pings me (I get enough notifications already).

If you take one thing from this section: build the parts of your workflow that are hard to do well manually.

What this has changed

The first is that writer’s block has basically stopped happening. Not because the tool writes the posts, but because every morning I sit down already knowing three to five things worth saying. The trend tracker, narrative arcs, and lookalike generator stack into a continuous queue of starting points. The gap between “what do I post today?” and “here’s the thing I’m posting about” closed from 45 minutes to about 90 seconds.

The second is harder to quantify. Building this thing made me a better marketer because I had to make every workflow assumption explicit. Every time I added a feature, I had to define what I actually cared about, what counted as a “top post,” what counted as a “pillar,” what counted as ICP engagement. Forcing myself to define it sharpened the whole operation.

And finally, this tool has definitely added much more pipeline for GrowthPair through the posts I’ve written because of the features. Hard to quantify this one as well since organic isn’t the easiest to attribute, but I do know business has picked up since having more data fuel what I write on social.

Share

What it costs and what it replaced

The whole stack costs me about $15 a month across Apify API calls and a small Claude subscription overhead. The tools it replaced ran me roughly $50 to $150 a month each, and they did less.

There’s no maintenance burden in the way you’d expect from custom software. It’s a tool I built for one user, hosted on localhost. When something breaks I describe it to Claude and it gets fixed in the same session.

How to think about building your own

A few principles, since I get asked this constantly:

• Document the thing you do every Monday morning, then ask Claude to build it.

• Before any build session, say “ask me clarifying questions before you start writing code.” This single instruction surfaces edge cases you hadn’t considered and saves hours of rework.

• Feed it everything you have including your past posts, your analytics exports, your style guide, your ICP definitions, etc.

• Build the MVP in one afternoon, then live with it for a week before adding features.

• Block out 2 to 3 hours for your first session. There’s a learning curve, but the “aha moment” almost always lands inside of the first feature working live.

• Using GStack (Garry Tan’s Claude skills) feels like I added a jetpack to Claude Code. I highly recommend everyone check it out if you’re serious about building.

Why this matters for creators

For the past decade, the rule was: pick the SaaS that’s closest to your workflow and live with the compromises. There’s never been a single SaaS that was perfect for my use case.

Off-the-shelf tools are built for the median user. The median creator doesn’t have your audience, your pillars, your posting cadence, your ICP, or your analytics priorities. And for the first time, the cost of building exactly what you need is low enough that settling for “close enough” is merely a choice.

Content Machine 2000 isn’t special. It took eight hours and runs on a localhost server. Any creator with a clear sense of how they work can now have software shaped to that sense. Just a few focused afternoons, some cold brew and a willingness to spell out what you actually want is all you need.

If you’ve been waiting for permission, this is it. Get after it and build your one-of-one SaaS.

–

The AI Marketer

I'm an Uber & Coinbase marketer turned founder. Join me as I share AI learnings while scaling my startup to $10M ARR.

By Jonathan Martinez

.

It was still a nightmare to track everything while optimizing and creating reports. Tons of back and forth between tools and dashboards.

Over the last couple of months, I’ve been experimenting heavily with Claude (hence Claude Marketers video academy) and I’ve created what I call “The AI-Native Paid Ads System.”

Before I explain it, here’s an overview graphic:

Let’s start with the first layer, and work our way down.

Share

Layer 1: Connectors & data sources

As many have seen firsthand, Anthropic became a powerhouse because of “connectors” which allow you to hook many of your current tools up to Claude. Meta’s recent announcement is what really shows us what’s possible on running paid ads inside of a chat window.

Where this gets powerful is with Higgsfield, which is an image/video generator that uses all the best models (e.g., Nano Banana) to create high quality assets. You don’t need this connector if you’d rather keep the design in human hands. I do think AI creative is useful to validate copy and concepts at a much faster clip. Still not 100%, and I trust a designer much more for top tier creative assets but it’s a nice preview into the future.

Along with Meta and Higgsfield connectors, this system will continuously get stronger with:

• Actual campaign data from Meta

• Scheduled tasks competitive research

The campaign data will flow from the Meta MCP, so you don’t really need to do much there other than run regular analyses.

If you’d like to automate your campaign analyses along with competitive research, use Claude’s scheduled tasks to build this out.

I currently have a scheduled task that analyzes all of my competitor websites weekly and updates a database inside Notion with any changes (e.g., pricing, copy). This could feed creative angles and action items for new paid tests.

Layer 2: Claude skills and claude.md file

Having a claude.md file loaded into Claude will provide context on your brand ICP, design preferences and overall vibe. Every prompt will automatically get a “pre-read” of your brand.

Then you’ll want to add a specific Claude skill for paid ad management. This’ll get called anytime you ask Claude to work on paid ads, and will essentially feed it the brain of a Meta mastermind. To build the skill, you can prompt Claude with something like:

I’d like to build a skill for optimizing and managing paid social ads on Meta. Research the best 
minds on paid social and add all their best practices into this skill. This skill will be used
whenever I’m building ad dashboards, creating ads, optimizing campaigns, etc.

Running the account audit

Once you have your connectors, claude.md file and Claude skills loaded, you can run the audit on your Meta account with this prompt:

You’re auditing my Meta ad account. Use the Meta Ads MCP to pull live data directly. Don’t ask me to paste numbers or screenshots.

**Step 1: Confirm the account.**

List my ad accounts and confirm which one you’re auditing. If only one is connected, use it and move on.

**Step 2: Pull the data.** For the last 30 days and 90 days, get:

- Account-level performance trends (spend, CPM, CTR, CPC, CPA, ROAS, frequency)

- Campaign and ad set breakdowns, top and bottom performers by spend

- Top 10 ads by spend with creative-level metrics, plus any fatigue signals (rising CPM, falling CTR, frequency >3)

- Dataset / pixel quality, event match quality, and tracking errors

- Industry benchmarks for my vertical and objective

- Meta’s opportunity score and any diagnostic flags or auction warnings

**Step 3: Write the audit.** Structure it like this:

1. **Verdict** One paragraph. What’s actually going on in this account. Be blunt.

2. **Account health** Pixel, CAPI, event match quality, dataset issues, tracking gaps.

3. **Performance** How spend is pacing. What’s working, what’s bleeding money. Cite the numbers.

4. **Creative** What’s winning. What’s fatigued. Patterns I should double down on or kill.

5. **Structure & targeting** Campaign architecture, audience overlap, CBO/ABO setup, budget allocation problems.

6. **Benchmarks** How I compare to industry on the metrics that matter. Where I’m ahead. Where I’m behind.

7. **Top 5 actions this week** Specific, prioritized by impact. “Test more creative” doesn’t count. Tell me exactly what to test, in which ad set, and why.

**Rules:**

- Cite real numbers from the data. No vague claims.

- If data is missing, broken, or looks off, call it out.

- Skip corporate hedging. If something is broken, say so.

- End with the 5 actions. That’s what I’ll actually do on Monday.

Higgsfield in the loop

I still think we’re in the first inning of AI creative, and I wouldn’t trust Higgsfield for creative autopilot but it’s very cool to see. Here are some static assets I created for GrowthPair:

It's crazy to think the account audit and these static assets all happened inside of Claude. I was feeling a bit fancy so I asked Claude to animate a static asset into a short video:

You might be thinking these assets are super generic, and you're right. These were quick demos. The next set used a more detailed prompt for the GrowthPair brand and our target customer.

Create an ad using the growthpair.com look (colors, typography etc) and add some imagery 
so it’s not solely plain text on a background. Take inspo from ad libraries from our 
competitors on Meta like MarketerHire and best practices around Meta ad creative.

Much better. The fix took 15 extra seconds.

Layer 3: Claude’s live artifacts and tasks

Since Meta is connected to Claude, you can use all of the data to create beautiful dashboards using live artifacts. I wrote a more in-depth essay about how to build these dashboards in minutes so I won't go into details. Here's an example of a dashboard I built for the thought leader ads we run on LinkedIn.

Layer 4: Learning loop with GitHub

The layer that pulls this all together is feeding your campaign learnings into a shared GitHub repo so your entire team and Claude have context on historical performance.

Source of truth — it never forgets how one of your tests performed.

To build out this layer, you’ll need to be using Claude Code so you can upload all this information into GitHub. Here’s how:

1. Create a folder on your desktop and connect it to Claude Code

2. Prompt Claude Code with something like this: I want to start running my paid ads on GitHub/ Claude Code. The idea is to house our campaign learnings here so that the entire team and Claude know what’s worked and what hasn’t. The entire team will update the folders with their tests. Can you ask any questions necessary and help me set this up so that we have the right foundation to get this kicked off?

3. Claude Code will ask questions and guide the setup for your folder structure

4. Create a repo on GitHub (important: make it private)

5. Give Claude Code the GitHub repo link and it’ll create your initial commit

6. Add your teammates to repo in GitHub (Repo Settings > Collaborators > add their GitHub usernames)

7. Using GitHub Desktop, click “Add Local Repository” and navigate to your marketing team folder that Claude Code created

8. It’ll detect the existing repo, and then you can hit “Push origin”

I believe that even outside of "The AI-Native Paid Ads System," all marketing functions will run their teams out of a GitHub repo. If you start with this, you can always add the files from this GitHub repo into a larger marketing org repo.

TLDR — how to action on this

1. Set up your connectors: Install Meta and Higgsfield connectors in Claude. It takes 2 minutes.

2. Build claude.md and skills: Having these files loaded into Claude will make it much stronger every single time you prompt it.

3. Run your first AI audit: Use my exact prompt above on your biggest ad account. See what it catches that you missed.

4. Test one creative angle: Take the audit’s creative recommendations and spin up 3 new ad variants.

5. Build a simple dashboard: Use Claude’s artifacts to create a live performance dashboard for your campaigns.

6. Schedule weekly competitor research: Set up one automated task to track your top 3 competitors’ ad changes.

7. GitHub repo: Create a repo on Github and ask Claude to share learnings on a weekly cadence.

Until the next one,

Jonathan

Share

I thought to myself, “how cool would it be to have my own JARVIS?”

You know.. the one from Iron Man.

This is Wikipedia’s definition of what JARVIS is:

J.A.R.V.I.S. (Just A Rather Very Intelligent System) is a sophisticated AI acting as Tony Stark’s primary assistant, butler, and user interface for the Iron Man armor. He manages Stark’s infrastructure, provides combat data, and ultimately evolves into the Avenger known as Vision

That’s how my personal project to build a second brain began, which I’ve dubbed Gizmo.

Share

The vision behind Gizmo

I wanted a personal command center that housed all the information about my startup and personal life– something I could open at 7AM daily to get a snapshot on what’s important.

It was going to need information from sources like my email accounts, Slack, iMessage, Notion and bank accounts to be able to surface it all.

There were a few other features that were real important to me as well:

• I wanted to access this anywhere without a computer open 24/7

• Act as an MCP that I can hook into any Claude instance

• Ability to easily query all of the information in my life

With that vision, I poured myself some Equator cold brew at home and got to work on building this inside of Claude Code.

Here’s a preview (note: using dummy data) of how it looks before we get into how you can build this yourself:

As you can see, it’s hosted on the GrowthPair domain as a subdomain that I can access anywhere across devices.

How I built Gizmo

I’m not an engineer, so I simply prompted Claude Code with my vision that looked a little something like:

“I want to build a personal command center with information about my life and startup that I can access anywhere, anytime, and will generate daily briefings for me. Ask me any questions necessary to build out a robust system.”

From there, Claude started asking me questions on what exactly I wanted and I had to decide on the tech stack for the project. To be completely honest, I usually go with what Claude recommends for backend/ frontend systems since I’m not the most qualified to make those picks myself. Going with what Claude recommends has worked out thus far across all my projects.

Here’s where we landed with the tech stack:

• Backend: FastAPI (Python), PostgreSQL with pgvector extension, deployed on Railway

• Frontend: React SPA with Vite, Tailwind CSS

Then came the infrastructure, and how everything was going to work together:

- Railway for hosting (auto-deploy from GitHub main branch)

- Resend for transactional email (briefing delivery)

- Plaid for bank account connections and transaction sync

- Google OAuth for Gmail and Google Calendar access

- IMAP for iCloud email sync

- Apple Contacts database for iMessage name resolution

- macOS launchd for local iMessage sync agent

As you can see, there are many connections and sources that were needed to aggregate all of the information and host it online for me to access anywhere.

But then, I needed AI to actually make use of the entire database I was creating. Here’s what that looks like:

- Claude Sonnet 4 (Anthropic API) — daily briefing generation, chat/RAG, weekly digests,

travel extraction

- Claude Haiku 4.5 — meeting prep briefs

- Google Gemini gemini-embedding-001 — 768-dimensional vector embeddings for semantic

search

- pgvector with HNSW index — vector storage and retrieval

All of this is just a fancy way of saying that Claude API is being used to create daily briefings (two models for separate use cases) and Gemini/ pgvector to create the embeddings needed for search. The vector embeddings are really important because it helps draw similarities with my searches. Here’s a quick explainer on how that works:

• If I search “basketball” without semantic search, the results will look something like

  • basketball game

  • basketball score

• If I search “basketball” with semantic search, I’ll get results like

  • basketball game

  • basketball score

  • NBA playoffs

  • Lakers game

It’s a smarter way to query. Gizmo can now draw connections between “basketball” and “NBA” in this scenario.

Share

How I’m using Gizmo daily

I wake up to an email every morning which preps me for my meetings that day and surfaces any high priority items that I should immediately tackle. It’s doing this by going through all my connected data sources and running it through Claude’s API for parsing/ intelligence on all the data.

Here’s what one of those emails looks like:

I can reply to the email and task Gizmo to draft emails for me, or provide me even more information on anything I need.

Throughout the day, I always have Gizmo in its separate tab that I can view for my daily briefing, search anything and ask it questions like:

“When’s the last time I spoke with Sean?”

“What did Sean and I discuss last week?”

“How many discovery calls did I take over the last month?”

“What are the top restaurant recommendations friends have sent me?”

I also get a snapshot on my upcoming reservations (restaurants, travel, etc) and how my finances are looking across investment accounts. I prefer this over looking through emails or logging into multiple investment accounts.

As I continue using Gizmo, it ingests more data and creates vector embeddings making it more useful. If for whatever reason, I need to jump to a new Claude account, the Gizmo MCP enables Claude to query my context lake within any instance.

Still need to incorporate health data

One big item that I still need to add are health syncs across my Function bloodwork results, Ezra MRI results, Apple Health and hospital records. I want to be able to query anything about my health and get instant information. Things like how my bloodwork has fluctuated over time, my allergies, vaccine records and why I’m potentially feeling a certain way in a given week.

With all of this new healthtech coming out, we have more data than ever to be able to take action on in our daily lives. I also think it’ll be easier to incorporate this data via Claude Connectors/ APIs that these companies release.

How to get started building this

I’ll be completely transparent– this wasn’t something I built over the course of 1 hour, or even 10 hours. There were many bumps in the road, including Claude’s API spending $10+ per day until I optimized it to use cheaper models (Sonnet) for tasks where it didn’t need the latest and greatest (Opus 4.7).

I’ve probably invested ~30 hours over the course of the last few months building this out, adding features, optimizing and ideating on what else I wanted.

If you’re okay with the lift and find something like this useful, my biggest recommendation would be to copy this entire essay and paste it into Claude with a prompt right after that goes something like:

“Build me my own Gizmo using the guidelines from this essay.”

With all of the information packed in here, you’ll be able to avoid many of the mistakes I made and likely build it in ⅓ of the time.

This has been such a fun project that’s taught me so much about API connections, vector embeddings and building a project with various interconnected tools. I highly recommend building this if you want to learn more about technical infrastructure.

If you don’t have the time for this, I do think that there’ll be startups which build a SaaS platform personal sidekick for like $19/ mo. This’ll help you avoid the upfront time investment and ongoing maintenance of the platform.

Looping this into your marketing org

While this isn’t a tool for marketing orgs specifically, I think there are many parallels that can be drawn here. If I was leading a marketing team, I’d take Gizmo and build it to be marketing specific. Some of the features would include:

- Dashboard that pulls in all growth metrics, channel data, experiments, etc

- Paid ad spend budgets, metrics and overall pacing

- Daily briefing on experiment status across channels

- Self-improvement database of all the wins/ losses from our experiments

- Master skills repo for the team to pull from

This would be a dashboard that the entire team could access daily to get a pulse on how everything is going. Again, I feel like there’ll be a SaaS platform that gets built to do exactly this for various orgs within a company.

Any additional questions

For those still reading to this point, you’re a real one and I appreciate you reading through my coffee-buzzing thoughts. If you want to learn more in video format, check out the Claude Marketers video academy that I’ve built as a labor of love.

Claude on,

Jonathan - your AI whisperer

Subscribe now

The latest time sink?

And.. I say that in a good way..

Is Claude’s “Live artifacts” which allow you to create dashboards for your startup.

In this short essay, I’m going to showcase how easy it is to create bespoke marketing dashboards, what I’ve built so far, and how to use tools like Windsor.ai to hook up ad account data.

Share

How to create marketing dashboards on Claude

This will be short and sweet: just go to “Live artifacts” within Claude Cowork and then click “New artifact.” From there, Claude will guide you through some questions that’ll ask the type of dashboard, data and style you’d like. That’s all.

You can also prompt Claude Cowork to build a dashboard for you and ask if you’d like to build a Live artifact.

What I’ve created so far

I’ve built a dashboard to help me visualize creative performance on LinkedIn thought leader ads I’m running for GrowthPair. Since this type of ad format doesn’t really lend itself well to tracking/ attribution, I typically default to measuring CTR on the ads. As a marketer, this sort of pains me, but this ad format does so well since it looks organic. I’ve written about those ads here.

This is what Claude came up with:

It has charts and visualization for metrics like:

1. CTR over the last 90 days vs my target of 20%

2. Budget pacing for the month

3. Action items to pause/ ramp up creative based on performance

4. CTR across all creative in campaigns

Since LinkedIn doesn’t have an official Connector on Claude, I had to use Windsor.ai to connect my ad data into Claude. You can search connectors like “Windsor” or “Supermetrics” on Claude and add those to ingest metrics from channels that don’t have official Connectors.

Then I built a dashboard that married all of my Brex (bank account) and Stripe data in one easy view. Here’s what Claude came up with:

In this dashboard (note: numbers blurred), I’ve got the ability to view:

1. Expenses by category

2. Monthly spend trend lines

3. Expenses vs revenue trend lines

4. Outstanding AR

5. Revenue trends

Instead of viewing two websites, I’ve got it pulling all of the data into one. I think this gets real interesting for marketing teams looking to consolidate all of their ad spend data or trends into one unified dashboard.

Share

Some other random ideas

If I was a marketing consultant today, I’d be eating this up and creating live dashboards for every single client that I managed. It would make management much easier and wow my clients all at the same time. Claude also gives you the ability to download the dash as a snapshot PDF so that you can share it with others.

If I was on a marketing team, I’d be building these dashboards out to make my life easier with management of campaigns and KPIs. I’d create competitor tracking dashboards that monitored new ad creative or website changes in an easy-to-digest UI. Etc etc.

It used to take weeks to build these dashboards on tools like Looker or Tableau, often required knowing some level of SQL and required maintenance. That’s not to say that these Live artifacts won’t require maintenance, but it’s as easy as prompting Claude for edits and improvements.

Happy dashboarding!

P.S. If you’d like a video lesson on this, please reply letting me know. If I get enough folks interested, I’ll film this for the Claude Marketers video academy.

Note: I filmed an entire video lesson track on Claude Marketers that goes over everything I’m about to cover in this essay in the case that you prefer to learn visually.

So, what exactly am I covering? After reading this, you’ll learn:

→ Why it’s important to setup a brand design system

→ How to build beautiful landing page designs

→ How to create full bespoke HTML designs

→ A rapid way to create new social posts/ grids

Let’s get artsy, ya’ll.

Setting up brand design system

The most important step before you get started in Claude Design is setting up your brand design system. There are a couple of ways to do this:

1. Upload your brand elements (icon, colors, fonts, etc)

2. Upload an HTML version of your website

3. Link your Github repo

Share

For the purpose of this tutorial, I’ll walk you through how to load in a few brand assets so that Claude can create your design system. It’s as simple as a few button clicks. First click “Design systems” and then “Create” to get started.

After you click “Create” you’ll be prompted to start loading in assets from your brand, like a logo and typography. The more assets you upload, the stronger your design system will be.

Here’s what the GrowthPair (my startup) design system looks like:

This now becomes the basis for all the designs you create for your brand going forward. When you go to create a new presentation, landing page, etc, you select the design system you created:

It’s important to build this as detailed as possible because everything you create with Claude will use this system as the instructions.

Landing page design

I wanted to see how Claude could do against the GrowthPair site that cost about $15,000 and 2-3 months to build. So I gave it this prompt:

I’d like to create a website designed for GrowthPair. Use all of the brand files that you have to make this a beautiful website that converts. Our ICP are marketing executives looking to scale their team with high-quality global marketing talent. Make it amazing.

This is what it came back with after about seven minutes:

Pretty dang good.

What surprised me the most with Claude Design and Opus 4.7 is how witty and natural the web copy felt. As an example, instead of using “See what our clients are saying” above the testimonial slider, it used the copy “The proof is in the pairing.” I genuinely let out a “wow” when I read that copy– how clever. (p.s. it’s a play on the phrase “the proof is in the pudding”).

I’m hoping that Claude Design doesn’t create the same design for everyone over the next 6-12 months and it becomes the new “AI slop.” An element I’ve seen it really like to use is highlighting of text like it did with “global” in the screenshot above.

Pro Tip: Ask Claude Design to create a file for you that you can load into Claude Code. This’ll allow you to port the new design over to your actual website very easily.

Email and social design

Now, let’s see how Claude will do with email and social designs. Here’s the prompt that I used to create new email designs:

I’m looking to create a couple of new emails for GrowthPair. These emails go out to prospective clients and should have a very nice design. I’ll be using these email designs within Beehiv, which is where we send out emails. Make the designs fun and make sure that they are able to be ported into Beehiv.

For email design, a couple of tips that’ll save you tons of time:

1. Ask Claude to provide HTML copy/ paste code so that you can paste into your preferred ESP (i.e., Beehiv)

2. Ask Claude to ensure the designs are mobile-responsive (super important)

3. When creating your newsletter, make a copy of the design template so you always have that as the basis for new emails with that design

On the social front, here’s what Claude designed:

Something I noticed when designing these social posts, along with everything else is that Claude Design will often get the font style wrong even though I’ve loaded font (Jakarta Sans) into the design system. If this happens to you, you’re not alone. I’m sure this’ll be fixed, but it’s an issue I keep running into. The quick fix is to just ask Claude to fix the font to your brand’s font.

Share

Referencing inspiration files

For marketers that don’t have all the design ideas in their head (like me lol), but are now armed with Claude Design, I have the biggest hack for you.

Simply head over to the following websites:

• Dribbble

• Mobbin

• ReallyGoodEmails

• MotionSites

Within these sites you’ll get Pinterest boards with all sorts of marketing collateral like email and web design inspiration. All you have to do is take a reference inspiration, screenshot it and ask Claude to design something similar. You’ll be surprised at how close it gets to the reference file you’ve attached. It’s almost 1:1 close.

To showcase this, I took a reference website that had a “liquid glass” hero section styling from MotionSites.ai and asked Claude to create social posts with a similar design.

The options are really endless when you have a sea of inspiration living on websites like Dribbble to build from as a starting point.

After testing Claude Design for about 20 hours, I’ve realized that there are clearly still some limitations, but it’s a real powerful tool to get marketing designs off the ground. I’d be totally comfortable shipping some of the designs I’ve created with Claude Design, or at the least, sending over to designers so they have reference points on what I’d like to create.

As I say over at ClaudeMarketers.. Claude on and until next time,

Jonathan - your AI whisperer

As a startup founder, I’m running marketing, ops, sales, finance, recruitment, and the list goes on. But naturally as a career-marketer, I still spend a huge chunk of my time running paid ads, writing content and experimenting with our positioning. In parallel, I’ve been heavily obsessed with Claude and how it can empower my team.

This leads me to how I’ve been acting as more of a Growth Engineer these days, versus a typical Growth Marketer (my job when I was at Uber).

I thought it’d help to share what my daily workflows look like and how it differs widely from my time in growth at various tech companies.

Share

Thinking in systems

The largest shift is in how I’m thinking about the structure of our marketing team, context sharing and experimentation.

Thought process pre- Growth Engineer era:

1. I have new messaging angles I want to test on Meta

2. Go to competitor ad libraries to research what they’re running

3. Write down all the angles we should test and send to design team

4. Launch campaign on Meta

5. Review results in 1-2 weeks and report to the team on wins/ losses

6. Rinse and repeat manually

The thought process post- Growth Engineer era becomes “How can I create a system that enables rapid experimentation on Meta?” So, the steps are instead:

1. Build an agent on Claude that automatically scrapes competitor ads weekly

2. Compiles a list of winning angles that we should test in PDF format

3. Design agent builds 50 variations of the same ad with copy tweaks

4. Another agent launches these ads automatically in Meta

5. That same agent reviews the ads daily and pauses immediate losers

6. Receive a report every Friday with messaging angle test results

In this setup, there’s a system in place, along with agents that are running the show with the messaging tests. This allows for much faster experimentation, that’s more dependable and runs 24/7 (I need my 8hrs of sleep lol).

As a former growth marketer, I was always thinking about how I could gain alpha on competition by using tools like Zapier to automate repetitive tasks. The issue with this is that those tools never really had a mind of their own. This changes with AI. We can now have our brains embedded within systems that are constantly thinking like us.

Context lakes will be the norm

I’ve been on marketing teams of two, all the way to 150 and the problem has always centered around context sharing– both within the marketing team and cross-functionally.

“Where’s the latest brand guidelines?!”

“What are the sales team hearing on their calls this week!!?”

“How many tests did we run last quarter and what were the results?!”

These questions always led to an endless pursuit of trying to locate items from people and scattered Google Drive links.

Growth Engineers now think in systems for their workflows, but also with how their teams share information. Instead of organizing a Google Drive with the latest marketing wins and ad collateral, this’ll all live inside of a system like Github.

Typical software engineers have been far ahead of marketers in how they operate within systems and shared repos within Github. This allows for version control, context sharing across machines and a source of truth where all software code lives.

In the same way that engineers use Github to organize code, marketing teams will use Github to house all their marketing collateral, experimentation data, brand guidelines, etc.

This allows marketing teams to:

1. Share a system of files and folders that constantly update with latest info

2. Work with the latest skills/ agents the team is running on, ICP data, etc

3. View a running log of all changes being made to the shared system (repo)

This style of team will outperform a scattered Google Drive or hundreds of Slack messages to get to an answer. Speed will win in this new era. Growth Engineers will own the build and maintenance of this foundational system.

Share

Building and orchestrating agents

Regardless of the model or platform (Claude, GPT, Gemini, etc), it’ll be table stakes to build agents within these systems. They will run with schedules first, and then autonomously in the future based on its knowledge and parameters. For example, a Growth Engineer may set up a competitor research agent within a Research system that runs weekly at 8am. Eventually these agents will learn to run when they need and optimize themselves.

Competitor research agent → Research system → Marketing OS system

The flow is simple if you look at the multi-step workflow with the research agent feeding data into a research system (folder within Github), that then feeds to the entire OS system.

Now marketing teams will be able to all share information and can prompt their LLM of choice with questions like:

“What’s the latest research on company X?”

“Have there been any recent pricing changes for our competitors?”

No more searching, reading through files or asking others these questions. Instead these agents will arm teams with more information than we’ve ever had, just a question away.

And just like a Growth Marketer would empower their direct reports with knowledge, train them up and make them smarter—Growth Engineers do the same for these agents. That’s the orchestration portion of the job duty. It’ll become critical to understand how to properly build these agents, make them as efficient as possible (srry, no unlimited token glitch yet), and have them output the most reliable and important information.

Just because AI is heavily automated, the agents will only be as good as your last training session and tweak that you made. Rather than 1:1s with direct reports or a team, Growth Engineers will spend that time training the agents they’ve created, connecting them within the systems and analyzing their outputs. This’ll prevent AI slop and hallucinations.

On top of agent builds, there’s also the skillset needed to create bespoke software to support marketing that wasn’t available before (eng resources get prioritized to product). This role will require marketers to think through building versus buying software for their team. I don’t believe teams will build out their own CRMs and ESPs anytime soon, but they’ll build out custom ad hook creators, SEO content recyclers, influencer coordination pipeline dashboards, etc.

Putting it all together

We still have a Growth Marketer at the core, but they’re now thinking differently and operating within a new world of AI possibilities. Digital Marketers evolved into Growth Marketers with the rise of mobile, testing product virality loops, creating referral mechanisms, optimizing CAC/ LTV and working cross-functionally. It’s another radical shift, but will make marketing teams stronger than before.

When you put it all together, this is how a Growth Engineer looks in a modern marketing team:

1. Growth Engineer operator/ orchestrator

2. Foundational system (marketing OS)

3. Bespoke systems (experimentation, research, analysis)

4. Agents working inside these systems

This is in stark contrast with how the old way a Growth Marketer operated in a marketing team:

1. Growth Marketer executing on tests

2. Disparate file systems across Notion, Drive, etc

3. Some automations across Zapier, CRMs, ESPs

4. More junior marketers running research and analyses

One thing to note is that there can be multiple Growth Engineers on a team, similar to how there are multiple Growth Marketers on teams.

My message to all marketers is to embrace the change. It’ll lead to huge tailwinds on your marketing team. And finally, I predict that Growth Engineers will be the hottest new role in tech and a huge force for startups in the years to come.

Subscribe now

I keep coming back to that same feeling I had in 2008 when I was in awe with the App Store on my iPhone. The doors that the early apps opened, from Super Monkey Ball, to iBeer *so nostalgic* were like nothing we had seen before.

So I thought I’d write about what I’ve been building in hopes to give other marketers inspo on what’s possible now. To be clear, I’m not an engineer, but being able to build like one has been so liberating. I feel like my words have become much more powerful.

Let’s dive into three of my favorite Claude projects from the last couple of weeks.

Video academy on Claude

Since I’ve self-proclaimed myself as an animal on Claude, I decided to live up to that and build a video education academy specifically for marketers: Claude Marketers. I built this entire platform within a weekend. Yes, in two days… let that sink in.

I built a Udemy-type website by prompting Claude and filming about 10 video tutorials. The idea was to build the entire platform on Claude Code, while keeping the video lessons all human-generated.

Here’s a look at the first prompt I used to kick things off:

I’m looking to create a website that houses a ton of training material for marketers that are looking to learn Claude cowork and Claude code. The website will be called claudemarketers.com.

The idea is to have video lessons that I film along with a paid community and skills that they can copy and paste into Claude, and really just an overall educational website for marketers looking to be the top 1% of those that are using Claude.

These lessons will actually be repurposed and reused from the ones that I’m building currently for GrowthPair University in Cowork.

The entire lessons, videos, and program should have the ability to log in, pay for a membership, and get access to everything. It should have connectors to things like Google Auth and Stripe.

Ask me any questions necessary before you start building out this platform. And then we can get to building the actual website and platform.

Claude then asked me a series of questions before it got started writing code:

Pro tip: Prompt Claude directly to ask you questions before getting started. This’ll save you so much time and help you refine what you’re trying to build.

Recruitment resume scanner

A big part of my job as a founder is to unblock my team from issues preventing amazing work, so this app was only natural. As a global staffing firm, my team goes through thousands of applications weekly to find the very best candidates. As you can imagine, this is very time consuming and is all done one resume at a time. If my team is reading this… thank you.

Share

I built an automatic scanner for all applicants we receive that scores resumes based on criteria we’ve defined. It’s important to note that this app does not automatically accept or reject candidates, but instead makes it much easier for my team to visually scan dozens of profiles at once. We’ve defined criteria, such as keywords and sample resumes we’re looking for that the app looks for and then gives us simple PASS/ FAIL for the sections.

To build this app, I used two important APIs:

• Claude API which does the actual screening

• Workable API which feeds resumes directly from our ATS (applicant tracking system)

If you asked me to use Git, Github, commits, repos, and hook APIs into custom software just a few months ago, I’d stare at you with a blank face. I explain what all these terms are in the first lesson for Claude Code at Claude Marketers.

Pro tip: Ask Claude to explain what/ how it’s building as it codes and this’ll give you a really nice primer on the fundamentals of software engineering. You can even ask it to explain like you’re a 5 year old (I did!) and it explains in much easier terms.

Content Machine 2000

I spend ½ my days on LinkedIn, whether engaging with content, responding to DMs and most importantly, publishing my own learnings. I’ve been able to build GrowthPair into a seven-figure business entirely with partners like Demand Curve, and thought leadership on LinkedIn.

This past weekend, I thought “why can’t I take my content a step further?” so I created a dashboard even MrBeast would be proud of.

The purpose of this dashboard is to arm me with all the necessary intel needed to optimize how, when and what I post on LinkedIn. It does many things, but here are my favorites:

1. Pulls daily posts going viral within my niche (AI marketing)

2. Watches creators I select and surfaces their top performing posts

3. Creates lookalike content based on my top performing posts

4. Recycles my all-time performing posts so that I can repost them

5. Monthly retros on how I did MoM which self-improves the machine

I now feel like I’m going into posting content that my audience wants, rather than just going off vibes. I’ve become a scientist with this app, and it’s completely bespoke to my needs as a creator, rather than buying SaaS off the shelves with features I don’t care about.

Wrapping up

If anything, I hope this post inspires the next generation of marketers (you!) to build apps that their startups need. After building the last few months in Claude, I’m a believer that we’re entering an era of unlimited bespoke software that startups can leverage for their unique use cases.

Share

When I was on the growth team at Uber, we had fancy bespoke software (i.e., UTM creators, budget allocation automation dashboards) but I’ve never seen that level of customization at a startup. That changes now.

Enjoy building, and as I say, Claude on.

This setup was the norm for a growth team. Looking for docs for hours, or hunting someone down to understand what things changed seemed par for the course. Oh, and don’t get me started on the file naming chaos: Female_Photoshoot_SF_032018_v3_FINAL.png.

Marketers have been dreaming of a better way (not just me?), and it’s been sitting in plain sight in front of us on the product/ engineering teams. Github.

I believe that housing all files inside of Github will be the future state of how a marketing team operates. I’m already doing this at my startup and it’s been a game changer with how fast I move and the amount of context-sharing I can do with everything sitting in the same place.

I’m here to show you just how easy it is to build this setup for your team. Let’s Git.

First, what’s Github?

For those that haven’t started coding up projects on Claude Code, let me give you a brief introduction along with key terms. Github has been around since 2008 and is a platform for developers (and now, marketers!) to create, manage and share their code.

A few key terms to know before we dive in:

Git: A version control system that tracks changes to code over time. Claude Code integrates deeply with Git.

Commit: A saved snapshot of your code at a specific point in time, with a message describing what changed.

Repo: A project folder tracked by Git, containing all your code and its full change history.

It’s basically a much more advanced Notion or Google Docs, at least in the way that we’re going to be using it for marketing.

Use cases for marketing team on Github

Before demonstrating how to do all this, I want to show how my current setup looks like for my startup. Here’s a snapshot of my Github repo:

You may be asking, “So what’s the big deal?”

With this style of setup, I’m able to do 3 very important things:

1. House all marketing files within folders in one organized place

2. Track any changes that my team is making across files (aka, their commits)

3. Ensure my team has real time information on everything on the business

Those are the logistical pieces on maintaining a well organized team, but the fun is really when you’re using Claude Code with this repo.

Share

Now, I can prompt Claude Code with things like:

1. Analyze all sales calls from last 7 days, along with our website messaging to find gaps in our positioning

2. Using our best performing email headlines, create 10 new test variations for our new product launch

3. What did I miss from my team meetings after being out the last 2 weeks

Here’s an example of a prompt I ran directly through Claude Code:

The beauty of this prompt example is that it takes details from various sources all living across folders within the repo. This would have taken me hours to find the right files, put it all together and come up with campaign ideas previously.

In a hypothetical world where I’m working at a company, I now don’t have to go running to the sales team for their latest sales call transcripts, or to the CEO for the company goals.

Context sharing. Real time view of the business. Analyzing information between various folders/ files. It’s all possible now with this setup.

Okay cool, how can I set this up?

I’ve filmed an entire video lesson on Claude Marketers on how to do this, but I’ll demonstrate how easy it is to setup here:

1. Create a folder on your desktop and connect it to Claude Code

2. Prompt Claude Code with something like this: I want to start running my marketing team on Github/ Claude Code. The idea is to organize our brand guidelines, sales calls, etc in different folders. The entire team can access this and update the folders with access to the repo on Github. Can you ask any questions necessary and help me set this up so that we have the right foundation to get this kicked off

3. Claude Code will ask questions and guide the setup for your folder structure

4. Create a repo on Github (important: make it private)

5. Give Claude Code the Github repo link and it’ll create your initial commit

6. Add your teammates to repo in Github (Repo Settings > Collaborators > add their GitHub usernames)

7. Using Github Desktop, click “Add Local Repository” and navigate to your marketing team folder that Claude Code created

8. It’ll detect the existing repo, and then you can hit “Push origin”

As I’m writing this out, I realize that the video lesson I’ve filmed will likely give much more color on how to do this. It may seem technical, but I promise as you get going it’ll be very doable.

The video lesson for setting up a Github repo is in “Claude Code for Marketers” module and titled “Deploying Your App: GitHub, Railway, Apify & Google Auth.” And for the video lesson on this exact essay, you can go to the “Building Your Marketing Team on Claude” module.

Why I’m so hyped

Now that you have an understanding on why I built this, how to do it yourself and the advantages, here’s why I’m truly hyped…

I see a world in the not-so-distant future where orgs like marketing can operate insanely efficient with shared context that’s easily accessed.

Everything from human talent, to AI agents accessing and building out campaigns and more through a Github repo and Claude Code.

It used to take ages to find docs, connect dots between data and ship with speed. You’d need tons of resources and an insane dedication to documenting.

Now, teams with <5 people and those with 100+ can move much faster and truly 10x their outputs. I hate 10x term because it’s so cliche, but it’s true. It all starts with a durable foundation though, and I believe this is the one for the future.

Share

Within the last month, here’s a few things I’ve done with Cowork/ Code:

1. Built a resume screening web app for my recruitment team

2. Built an X social listening tool for keywords in my niche

3. Performed ~5 competitor research analyses on ad libraries

4. Scheduled competitor website monitoring tasks with Cowork

5. Optimized my Google Ads account after Cowork’s audit

For the sake of not making you scroll a list of 25-50, I added in a few of my favorites above.

So I couldn’t hide these superpowers any longer, and wanted to teach what I had learned through immersing myself in Claude. When I say immersing, I really do think I live inside Claude’s chat window more than Chrome nowadays.

Fun fact: I’ve always found myself coming back to teaching through content. I was a Youtuber back in 2008 creating videos on the App Store when it launched. I produced 500 videos for that channel across 7 years.

First, what is Claude Marketers?

It’s a video academy where I walk marketers through setting up Cowork, specific use cases with Cowork, building out sites and apps with Code and more. It basically takes you through the beginning stages of Cowork, and ends with advanced use cases of Code.

In total, there are 11 videos that are self-paced and keep track of your progress as you go through the academy. There are also some copy/ paste skill prompts, a community and an actual certification that you can download after you complete everything.

It’s become a nice little resource for AI-curious marketers who are looking to dive much further into being AI-native.

I’ll keep building this resource with more videos as long as there’s interest.

How I built Claude Marketers

My fiancé was studying over the weekend for her new physical therapy license, so I had an abundance of time to lock in on the build.

The apps I used to create the academy:

• Claude Code

• Wispr Flow

• Screen Studio

• Github Desktop

• Railway

• Google Auth

• GoDaddy

And a whole ton of cold brew.

This was my actual prompt into Claude Code:

I’m looking to create a website that houses a ton of training material for marketers that are looking to learn Claude cowork and Claude code. The website will be called claudmarketers.com.

The idea is to have video lessons that I film along with a paid community and skills that they can copy and paste into Claude, and really just an overall educational website for marketers looking to be the top 1% of those that are using Claude.

These lessons will actually be repurposed and reused from the ones that I’m building currently for GrowthPair University in Cowork.

The entire lessons, videos, and program should have the ability to log in, pay for a membership, and get access to everything. It should have connectors to things like Google Auth and Stripe.

Ask me any questions necessary before you start building out this platform. And then we can get to building the actual website and platform.

I used Wispr Flow to dictate this which is why there may be a misspelling or two.

Note: I later decided to make most of the video academy free, so Stripe became less relevant, but I wanted to show that the first prompt doesn’t need to be perfect.

Claude then asked me a series of questions before it got started writing code:

One of my favorite things about Claude is that it’ll ask you questions that you may have not even thought about when building something. This got me thinking about the structure of the academy and things like which service I’d use to host the videos.

Share

While Claude coded, I filmed

I already have an ideal setup (home office, Shure mic, 4k Anker webcam) and experience recording from my YouTube days so this wasn’t difficult. I filmed all 11 video guides within 2-3 hours with Screen Studio, and then got to editing them.

Once I was happy with the outputs, I exported from Screen Studio and started uploading the videos to a video hosting provider: Bunny.net

Claude Code had finished building the academy website at this point, but I went back and forth on various edits. Going back to what I mentioned above with ditching Stripe, here was my exact prompt used to remove it from the academy:

What’s nice about Claude Code is that for someone non-technical, if you ever get stuck, you can ask it to guide you on whatever you need help with. Here’s an example of me doing that:

I’m showing all of this to make things appear less scary. If you’re curious and okay with asking questions when stuck, you’re going to do just fine building on Code.

Wrapping up

I hope this helps inspire you all to start building on Claude Code. Believe me, it’s not scary. It actually gets super addicting as you see things come to life with your words.

I’ll likely continue to build internal web apps that I wouldn’t have been able to purchase before since they’re so unique to my team’s needs. I predict that over the next 12 months, many companies are going to build their own bespoke tooling rather than shopping around for solutions that aren’t quite perfect.

Enjoy building, and Claude on.

Share

I predict we’re going to see a massive shift going forward because of the ability to “vibe code” really cool and useful tools with tools like Lovable or Cursor. These tools have a major advantage over a PDF: personalized outputs.

Subscribe now

In fact, Demand Gen Report’s research found that 91% of B2B buyers prefer interactive content over static content.

Let me give you a real-world example.

I just built a lead magnet for my startup GrowthPair with a cup of cold brew, Lovable and an idea that I felt would be timely for companies. Here’s a look at an AEO website analyzer that took me about ~52 minutes to build and publish:

Check it out at https://aeoscoring.lovable.app/

Some cool additions I can add to this web app:

• Share with team button to unlock advanced insights

• Video guides on each scored criteria– how to improve each

• Chat module that connects visitors to our sales team live

The options are limitless when creating content in this format. The share button is a viral loop that turns one lead into a buying committee. The video guides increase time-on-site and move the prospect further down funnel. The chat module catches high-intent leads at peak curiosity.

My gut feeling is that an actionable tool like this is more fun than reading through a 19 page PDF guide for most. Personalized web app tools were primarily only afforded to companies that had the engineering and product resources to build them. Not any longer.

How I built my AEO scoring web app

I wish I could say that this took me hours and that I’m some genius, but all it took were my prompt instructions and about an hour of time. Here’s the prompt I used on Lovable:

can you help create an AEO analyzer tool that checks a website for AEO best practices?

this will be used as a lead magnet on LinkedIn that I post. for people to access the website, they’ll need to comment on a post I make

the website should have “by GrowthPair” on it and have a link + CTA on what GrowthPair does and can do for their companies. i’m the founder of GrowthPair for context and GrowthPair helps companies find the best global marketing talent while saving them up to 84% on payroll

This prompt got me ½ way there, and the rest were doing things like:

1. Connecting Firecrawl API so that the tool could actually analyze live website

2. Changing the styling a bit to match GrowthPair branding

3. Adding a “how-to” feature after tool runs website analyses

4. Selecting a domain name to publish the tool on

The playbook is the same

Instead of thinking about this as an entirely new playbook for B2B, not much actually changes:

1. Collect unique data or information relevant to your ICP

2. Compile and package into a PDF vibe coded tool

3. Provide this information in exchange for email address

4. Use the captured data to personalize follow-up

Why tools beat PDFs

With the AEO analysis tool I built, I’m able to capture an email address, domain name and specific gaps in a brand’s strategy. If I had just shared a PDF guide on how to self-score AEO readiness, I’d only know that someone was curious about this for their startup.

With the personalized data, I’m now able to:

• Segment companies in my CRM by their scores

• Enrich my CRM data with company industry and size

• Prioritize companies with the lowest scores

• Provide actionable insights for specific industries

This data asymmetry now creates the ability to increase follow-up communication quality tenfold. Say hello to very useful and personalized email communication.

In this new era, you can create five tool concepts in the time that it’d historically take to produce a PDF guide. The companies that ship fast and iterate will outperform the ones still debating which ebook topic to greenlight next quarter.

Share

I wrote this essay with two folks I admire for their career work in growth & startups:

Josh Grant (VP Growth, Webflow)

Sean Ellis (“Growth hacking” pioneer)

-Jonathan Martinez aka jon4growth

Sean here!

In the early 2000s, I poured millions into paid search because it gave me predictable ROI. We were not just buying traffic. We were optimizing funnels, strengthening referral loops, and scaling paid acquisition against improving lifetime value. It was clean, attributable, and immediate. I could model it. I could control it.

SEO was different. It had more upside, but it also had real downside. It felt messy and slow. It was hard to measure. And the space was full of hucksters promising results they could not possibly deliver. I chose the channel I could instrument and optimize.

Looking back, that avoidance is one of my biggest regrets in my growth career. The most successful and valuable growth leaders I know built deep expertise in SEO during that experimental phase. They captured compounding advantage while the rest of us stayed in the comfort of paid acquisition.

AEO feels similar today. It is early. It is imperfect. It is harder to attribute than paid channels. But it may be even more important than early SEO. Search engines helped you capture intent. Large language models increasingly shape how intent is formed in the first place. If your company is included in the answers, you influence perception before a prospect ever visits your site.

That is a powerful position to hold.

I do not want to miss this wave. And you should not either. That is why I (Sean Ellis) teamed up with Josh Grant and Jonathan Martinez to break down the highest leverage ways to get started now, before advantage compounds for someone else.

The AEO Diagnostic

Jonathan here!

To get started, a quick directional test that anyone can run to see how they’re ranking on LLMs is to search your phrases on ChatGPT or your favorite LLM. For GrowthPair, I may search something like:

“What are the top agencies for hiring global marketing talent?”

This is simply to get your feet wet and see where you’re at. If you have multiple phrases and are doing this daily to see where you stack up, it’ll get very time consuming.

Before diving into tools, it’s worth understanding how people actually search on LLMs. AppSamurai’s marketing manager, Metehan Yeşilyurt, studied 1,827 real ChatGPT conversations and discovered the query median length was 11 words. This contrasts with the average Google search of just 3.4 words.

This change in consumer behavior requires a mindset shift for researching the phrases you’d like to go after.

To get a bit more advanced with your analyses, I suggest using a tool like Peec, AirOps or Ahrefs which will query LLMs daily for your phrases to see where you’re ranking.

To illustrate where most companies are today: at GrowthPair, a global marketing talent startup I founded, I’d rate us a 4/10 on the AEO scale. We have the basics in place (llms.txt, FAQ sections, etc) but we still have tons of work to do on researching and optimizing further.

If you’re early in your SEO journey, temper your AEO expectations accordingly since there’s a strong correlation between ranking on Google and LLMs. While AEO introduces new tactics and metrics, it builds on the same content authority principles as traditional SEO rather than replacing them entirely.

The 80/20 AEO Playbook

Josh here!

SEO captured explicit intent. AEO shapes synthesized understanding.

Search engines helped companies win when buyers typed a clear query. They retrieved relevant sources and ranked them. Large language models go further. They do not just retrieve. They synthesize. They summarize categories, compare options, and often frame the problem before a buyer has visited a single website.

In search, you competed for ranking. In AI systems, you compete for inclusion in the explanation.

If your company is not part of that explanation layer, you do not just lose traffic. You lose framing power. And framing power compounds over time.

So what should you actually do right now?

The answer is not to chase prompt hacks or overhaul your entire marketing strategy. It is to build AEO the same way you would build any durable growth engine. Intentionally. In layers. With discipline.

I think of this as the Inclusion Stack. It has four components:

1. Question Ownership

2. Onsite Encoding

3. Offsite Consensus

4. Citation Engineering

You do not need to perfect all four immediately. But you do need to begin strengthening each one deliberately. Here is what that looks like in practice.

1. Question Ownership: Start With Question Mining

If you do nothing else this quarter, start here.

The foundation of AEO is not content production. It is question discovery. I recently wrote an entire guide on how to mine questions at scale.

AI systems reflect real human language. If you optimize for what you hope buyers are asking instead of what they are actually asking, you are building on assumptions.

Instead of brainstorming topics, extract buyer language from places where intent already exists:

Sales call transcripts
Support tickets
Demo objections
Reddit threads and community Slack groups
Product reviews
Competitor comparison pages

Cluster those questions. Prioritize them by revenue proximity, authority fit, and competitive weakness. Search volume matters less than whether you can credibly own the answer.

This defines your Answer Surface Area. It is the set of questions your company should naturally appear in when AI systems explain your category.

A focused two week sprint that surfaces and prioritizes one hundred real buyer questions will do more for your AEO roadmap than months of unstructured publishing.

Start there.

2. Onsite Encoding: Make Your Expertise Easy to Understand

Once you know which questions you should own, encode them clearly on your site. LLMs reward clarity. Ambiguity reduces inclusion.

Begin with fundamentals.

Clean crawlability.
Remove thin or duplicative pages.
Add structured data where it reinforces meaning.

Then rewrite your homepage and core product pages so anyone can answer, in one sentence:

Who is this for?
What problem does it solve?
How is it different?

If your positioning requires explanation from your sales team, it will not translate well into AI generated summaries.

Then apply FAQ enrichment to your most commercially important pages. Add real buyer questions. Answer them directly in plain language. Define terms clearly. Surface tradeoffs honestly.

At Webflow, aligning core product and category pages with real buyer phrasing improved clarity first. Visibility followed. We were not adding content for volume. We were removing friction so models could extract meaning more easily.

This is work you can complete in weeks.

3. Offsite Consensus: Authority Is Broader Than Your Domain

AEO cannot be won solely on your own website.

Large language models synthesize from across the web. Reddit threads. Review platforms. Industry publications. Partner blogs. Podcast transcripts. Analyst commentary. Community conversations.

If your expertise only lives on your domain, your influence will eventually plateau.

This is not about chasing press mentions. It is about showing up where your category is already being discussed and contributing something meaningful.

Participate in high signal community threads with substance, not promotion. Publish structured insights in outlets that shape professional opinion. Ensure your positioning is accurate and differentiated on comparison platforms. Share original ideas that partners and peers reference.

Authority in the AI era is distributed. Models recognize patterns across independent surfaces. When your perspective appears consistently, credibility strengthens.

Your website makes you legible.
Your broader presence makes you credible.

Inclusion is rarely the result of a single page. It is often the result of visible consensus forming around your expertise.

4. Citation Engineering: Structure Ideas to Be Referenced

In SEO, ranking was often enough. In AEO, citation matters.

Models favor content that is clear, structured, and distinctive.

Name your frameworks.
Define terms precisely.
Use comparisons and decision trees.
Share data, even modest internal insight.

Instead of offering general advice, articulate defined systems.

A simple test helps. If an AI system had to summarize this page in one paragraph, what would it pull?

If the answer is unclear, tighten the structure.

LLMs do not reward volume. They reward clarity and extractable insight.

5. Measure Like an Investor, Not a Traffic Analyst

This is the mindset shift that ties it together.

If you treat AEO like SEO, you will default to clicks. But clicks are downstream.

Instead, measure like an investor managing exposure across an influence portfolio.

Track visibility across your core questions.
Monitor share of voice relative to competitors.
Assess citation frequency.
Review context and sentiment.

Review monthly. Allocate effort the way an investor allocates capital. Strengthen weak clusters. Expand into adjacent territory. Consolidate thin content. Double down where inclusion increases.

You are not optimizing a page. You are increasing exposure to upstream influence. That is portfolio management.

The Compounding Advantage

Twenty years ago, the biggest growth advantage went to teams who invested in SEO before it was clean and measurable.

A similar moment exists today.

The market does not just search anymore. It asks. And the companies that consistently answer shape how the category is understood.

SEO captured demand. AEO shapes demand. Traffic is downstream. Inclusion is upstream.

If you want to act now, the path is clear.

Mine the real questions.
Encode clarity onsite.
Build credibility offsite.
Structure ideas for citation.
Measure like an investor.

Do that consistently and you will not just appear in AI generated answers. You will help define them.

Subscribe now

Measuring Early AEO Efforts

Hey, Jonathan back to tackle metrics!

As Josh mentioned, measuring clicks is now very inefficient since consumer behavior is changing. AEO introduces several new metrics worth tracking, and the best way to understand them is to compare old versus new. SEO focuses on driving traffic via clicks (rankings, CTR), whereas AEO emphasizes influence and trust by becoming the source of truth for LLMs.

• SEO: Organic traffic, keyword rankings, backlinks.

• AEO: Brand mentions, answer position, citation source, sentiment

The most important metric is Visibility: how often you’re mentioned in LLM responses across your tracked queries.

Share of Voice (SOV) measures how often a brand is mentioned in LLM responses relative to competitors for a specific set of queries, calculated as (Your Brand Mentions ÷ Total Brand Mentions) × 100.

The last metric I’d put in my top three is citation frequency, which is how often LLMs are citing your brand. Citation frequency is particularly powerful because citations (where LLMs link to your content as a source) signal authority. When LLMs cite your site, they’re essentially validating your expertise to users.

Here’s a quick example of how I’d track all metrics for the search phrase “What are the top agencies for hiring global marketing talent?”:

Visibility: GrowthPair appeared in 37 out of 100 tracked prompts = 37% visibility

Share of voice: GrowthPair appeared 30 times, competitors appeared 70 times = 30% SOV

Citation frequency: GrowthPair was cited 8 times out of 100 tracked prompts = 8% citation rate

Again, tools like Peec, AirOps, and Ahrefs can track this across multiple LLMs daily.

Conclusion: The Compounding Advantage

Sean back to wrap things up!

Twenty years ago, I chose the channel I could measure over the one that could compound. It felt disciplined at the time. In hindsight, it was cautious.

Today, large language models and AI are changing how people make buying decisions. Prospects are not just searching for links. They are asking for synthesized answers, recommendations, and summaries. And those answers increasingly shape perception before a company ever gets the chance to pitch itself.

That creates a new layer of leverage. The brands included in those answers influence how problems are framed, which options are considered, and what “best” even means. Over time, that influence compounds.

You do not need perfect attribution to start. You do not need a dedicated team. As Josh laid out, begin by mining the real buyer questions, encode your expertise clearly onsite, build credibility offsite, and structure your ideas so they can be cited. Measure progress directionally and iterate with discipline.

In growth, the biggest advantages rarely come from doing something perfectly. They come from doing something early. AEO is one of those moments. Your compounding advantage begins when you decide to participate.

Subscribe now

Read until the end for access to the full 23-page PDF guide I created.

So naturally I thought, what else can I do with Claude’s best-kept secret: Skills? Can I teach Claude to run paid ads?

Note: this isn’t meant to replace a paid ads team, but instead augment their work and make them 10x marketers.

Subscribe now

What are Claude Skills

If you didn’t read my article on turning Claude into a marketing team, let me give a quick refresher on what Claude Skills are.

They’re essentially mini agents that are called up in your chats depending on what you’re working on. For example, if you’re building a campaign brief and you have a campaign brief skill, it’ll pull those instructions up and execute.

Where this gets real powerful is when you have 5-10 various skills loaded into Claude, and it starts chaining them together in its responses.

Instead of getting generic responses (goodbye felicia!), you get thorough and on-the-money responses based on your Skills’ instructions.

P.S. If you comment on my LinkedIn post, I’ll DM you a 23-page PDF guide which includes all the .md files to set this up.

How I taught Claude to run ads

I wanted Claude to do a few core tasks very well that’d typically take hours for a trained paid ads specialist to do:

1. Ad scoring

2. Competitive scanning

3. Creative generation

I’ve spent hundreds of millions in paid ads while working at Postmates, Uber and Coinbase– and equally, hundreds of hours running these tasks. Yes, I even have pictures to prove it:

Here are the example instructions for the ad scoring skill.

## Mode 1: Ad Scorer

When the user shares ad copy, creative descriptions, screenshots, or asks you to evaluate an ad.

### The Scoring Framework

Score every ad across six dimensions on a 1-10 scale. The composite score is a weighted average.

**1. Hook Strength (20% weight)**

Based on Schwartz’s awareness levels and modern scroll-stopping principles.

- Does the first line / first 3 seconds stop the scroll?

- Is the hook calibrated to the audience’s awareness level?

- Unaware: Lead with the problem or a pattern interrupt, never the product

- Problem Aware: Agitate the pain using the audience’s own language

- Solution Aware: Differentiate, show why THIS solution vs. alternatives

- Product Aware: Handle objections, stack proof, make the offer irresistible

- Most Aware: Just make the offer, keep it simple

- Schwartz’s rule: “The more aware your market, the less you need to say.”

**2. Message Clarity (15% weight)**

Rooted in Ogilvy’s “Big Idea” principle — every great ad is built on one simple, memorable idea.

- Can someone understand the core offer in under 5 seconds?

- Is there one clear message, or is it trying to say three things at once?

- Ogilvy’s test: “Is this idea so simple it could be explained in a single sentence?”

**3. Persuasion Architecture (20% weight)**

Based on AIDA (Attention → Interest → Desire → Action) and Hopkins’ scientific advertising principles.

- Does the ad follow a logical progression from hook to CTA?

- Interest: Does it hold attention after the hook with proof, story, or curiosity?

- Desire: Does it build want through benefits (not just features), social proof, or transformation?

- Action: Is there a clear, single CTA that tells the viewer exactly what to do?

- Hopkins’ rule: Every element must earn its place.

**4. Proof & Credibility (15% weight)**

Informed by Hopkins’ insistence on specificity and Ogilvy’s research on what makes claims believable.

- Are claims backed by specific evidence? (”Saved $47,000 in Q3” beats “Save money”)

- Types of proof: statistics, testimonials, case studies, authority figures, demonstrations, before/after

- Hopkins: “Specificity is the soul of salesmanship.” Vague claims create zero conviction.

- Ogilvy’s finding: Ads with explicit, specific promises are 48% more likely to improve brand perception.

**5. Platform Optimization (15% weight)**

Based on Motion’s creative analytics framework and current platform best practices.

- Meta: Does format match placement? (Reels vs. Feed vs. Stories have different specs and attention patterns)

- TikTok: Does it feel native? (Overly polished = skip. UGC-style, creator-led, trend-aware = stop)

- Character limits respected? (Meta primary: 125 chars above fold. TikTok: text overlay readability)

- Thumb-stop potential in the first frame/second

- For video: Is the hook in the first 3 seconds? (Motion benchmark: 30-40% hook rate is strong)

**6. CTA & Conversion Path (15% weight)**

Grounded in Bird’s direct response principles and Hopkins’ measurability obsession.

- Is there ONE clear call to action? (Multiple CTAs = no CTA)

- Does the CTA match the awareness level? (Don’t ask Unaware audiences to “Buy Now”)

- Is there a reason to act now? (Urgency, scarcity, or time-sensitivity)

- Bird’s rule: Make responding as easy as possible. Every click, every form field, every moment of confusion costs conversions.

- Is the action trackable? Can you measure whether this ad drove results?

### Scoring Output Format

For each ad, produce:

```

AD SCORE: [Composite Score]/10

1. Hook Strength: [X]/10 — [one-line rationale]

2. Message Clarity: [X]/10 — [one-line rationale]

3. Persuasion Arc: [X]/10 — [one-line rationale]

4. Proof & Credibility: [X]/10 — [one-line rationale]

5. Platform Optimization: [X]/10 — [one-line rationale]

6. CTA & Conversion: [X]/10 — [one-line rationale]

TOP STRENGTH: [What’s working best and why]

BIGGEST LEAK: [The #1 thing hurting performance, with a specific fix]

QUICK WIN: [One change that could lift performance with minimal effort]

```

### Scoring Calibration

To keep scores meaningful and consistent:

- 9-10: Top 5% of paid social ads. Exceptional on this dimension. Rarely given.

- 7-8: Strong. Clearly above average, minor room for improvement.

- 5-6: Average. Does the job but nothing special. Most ads land here.

- 3-4: Below average. Clear weakness that’s likely hurting performance.

- 1-2: Broken. This dimension is actively sabotaging the ad.

Don’t grade on a curve. A 5 means average, not bad. Most ads ARE average on most dimensions, and that’s fine to say.

Each of these Skills can be packaged into a separate .md file but for simplicity sake, I’ve built it all as one copy/ paste .md file (paid-social-creative-lab.md). This Skill will activate whenever you need to score ads, analyze competitors or build ad creative.

Okay cool, how do I set this up?

I’m unsure why Claude hides the Skills in their user interface, rather than highlight it as a major product feature. It’s like McDonalds leaving their golden french fries off their drive-thru menu.

Claude Skills are installed using markdown (.md) files, which house all the information, rules and output formats to keep responses high quality. Instead of prompting Claude from scratch every single time you want a campaign brief or report, you install the skill once and it kicks in whenever the task comes up.

For the purpose of this essay, I’m going to teach you how to install using Claude Cowork on your desktop app, but if you’re using Claude Code or the web version, grab my PDF guide for full instructions.

1. Copy the code block above

2. Create a new .md file and paste the contents

3. Name it paid-social-creative-lab.md

4. Add it to your Claude Cowork settings

5. Start a conversation with a relevant request and Claude will use the skill automatically

The biggest hiccup most will experience is in creating the .md file, so below are specific instructions on how to do that on a Mac or Windows.

On Mac

1. Open TextEdit.

2. Press Cmd+Shift+T to switch to plain text mode.

3. Paste the skill code, then File > Save As and name it ending in .md (e.g., paid-social-creative-lab.md). Make sure the format is plain text, not Rich Text.

On Windows

1. Open Notepad.

2. Paste the skill code, then File > Save As.

3. Change “Save as type” to All Files, and name it ending in .md (e.g., paid-social-creative-lab.md).

This is how it looks with your Skills installed on the Claude desktop app:

How I’m using this

A major competitor of GrowthPair (my startup) is MarketerHire, so I had Claude run a competitive analysis. Shh, don’t tell MarketerHire lol.

Within minutes I had an 11-page PDF analysis where Claude analyzed 110 of their ads, scored them, and gave recommendations on how to capitalize on their weak points.

Tip: The analysis initially outputted a Cursor .md file, but I just asked Claude to convert that into a PDF. If this happens to you, you can always ask Claude to reformat/ repackage into another file format.

This is just a small snippet of the PDF guide that it created:

How I plan to continue leveraging this

I’ll likely be building something similar to this but for Google and LinkedIn ads as well. I’m a one-person marketing team at my startup, so being able to leverage Skills for paid social and paid search will be a huge lift.

Other than that, I see a world in which this also runs on a weekly basis to analyze any changes from my competitors’ ad libraries on Meta.

If I wanted to take this a step further, I can have the PDF guide automatically sent through on Slack to myself and the rest of the team. I’m always cautious with something like this because of notification bloat though. But something to think about.

I’m curious how everyone else will use this, and if any other ideas? Always keen to learn more on the AI front as things are ever changing.

Until my next AI marketing essay,

Jonathan

Want the full 23-page PDF guide, including the .md file you can literally just copy/ paste into Claude that I made? Just comment on my LinkedIn post here and I’ll DM you the guide.

It’s called a recursive self improvement loop.

This is one of the biggest unlocks I’ve made with prompting over the last few months, by far. So, what’s a recursive loop, you ask?

To understand, it’s best to explain what 90% of people are doing when prompting today. Most people prompt once and ship whatever comes back.

A recursive loop works differently. It creates, then evaluates its output and continues improving upon itself.

For example, a recursive loop could work like this:

1. Generate ad creative, then grade itself on copy strength and recreate

2. Build email sequence, then grade itself on subject line, body, and improve based on the scoring

The pattern is simple: prompt > builds first output > grades that output > builds again > finalizes when scoring is perfect.

The output becomes infinitely better as it gives Claude the time to think and build a better response by continuously evaluating itself.

Here’s how you build this:

1. Write down how you’d evaluate the output. What would make it great vs good? This becomes your scoring criteria

2. Turn each criteria into a score i.e. subject line strength 8/10

3. Ask Claude to create and then score against your criteria

4. Now build the loop. Tell Claude to create output and score against your criteria, then keep improving until it’s perfect

You’ve just created a teammate that keeps iterating based on your feedback.

Share

Using .md files to advance this further

An .md (markdown) file is the type of file that Claude will accept for Skills which I wrote about in-depth with instructions in my previous essay. It’s essentially just a fancy sounding way to describe a text file which you can upload into Claude.

To automate this recursive self loop, build an .md skill file on Claude so that it has your criteria every time you prompt it to build X thing.

Your .md skill file can have a scoring criteria for the prompts you most frequently use, and will always measure itself on that.

If I’m always prompting Claude to create paid ad copy, I may want to have an .md skill file that gives scoring criteria (example below), and then rebuilds based on the score:

##Ad Scorer
### The Scoring Framework
When prompted to create paid ad copy, create the ad copy, score every ad across six dimensions on a 1-10 scale, and then continue improving the paid ad copy to increase score. 

**1. Hook Strength (20% weight)**
Based on Schwartz's awareness levels and modern scroll-stopping principles.
- Does the first line / first 3 seconds stop the scroll?
- Is the hook calibrated to the audience's awareness level?
- Unaware: Lead with the problem or a pattern interrupt, never the product
- Problem Aware: Agitate the pain using the audience's own language
- Solution Aware: Differentiate, show why THIS solution vs. alternatives
- Product Aware: Handle objections, stack proof, make the offer irresistible
- Most Aware: Just make the offer, keep it simple

**2. Message Clarity (15% weight)**
Rooted in Ogilvy's "Big Idea" principle, every great ad is built on one simple, memorable idea.
- Can someone understand the core offer in under 5 seconds?
- Is there one clear message, or is it trying to say three things at once?

**3. Persuasion Architecture (20% weight)**
Based on AIDA (Attention → Interest → Desire → Action) and Hopkins' scientific advertising principles.
- Does the ad follow a logical progression from hook to CTA?
- Interest: Does it hold attention after the hook with proof, story, or curiosity?
- Desire: Does it build want through benefits (not just features), social proof, or transformation?
- Action: Is there a clear, single CTA that tells the viewer exactly what to do?

**4. Proof & Credibility (15% weight)**
Informed by Hopkins' insistence on specificity and Ogilvy's research on what makes claims believable.
- Are claims backed by specific evidence? ("Saved $47,000 in Q3" beats "Save money")
- Types of proof: statistics, testimonials, case studies, authority figures, demonstrations, before/after
- Hopkins: "Specificity is the soul of salesmanship." Vague claims create zero conviction.
- Ogilvy's finding: Ads with explicit, specific promises are 48% more likely to improve brand perception.

**5. Platform Optimization (15% weight)**
Based on Motion's creative analytics framework and current platform best practices.
- Meta: Does format match placement? (Reels vs. Feed vs. Stories have different specs and attention patterns)
- TikTok: Does it feel native? (Overly polished = skip. UGC-style, creator-led, trend-aware = stop)
- Character limits respected? (Meta primary: 125 chars above fold. TikTok: text overlay readability)
- Thumb-stop potential in the first frame/second
- For video: Is the hook in the first 3 seconds? (Motion benchmark: 30-40% hook rate is strong)
**6. CTA & Conversion Path (15% weight)**
Grounded in Bird's direct response principles and Hopkins' measurability obsession.
- Is there ONE clear call to action? (Multiple CTAs = no CTA)
- Does the CTA match the awareness level? (Don't ask Unaware audiences to "Buy Now")
- Is there a reason to act now? (Urgency, scarcity, or time-sensitivity)

### Scoring Output Format
For each ad, produce:
AD SCORE: [Composite Score]/10
1. Hook Strength: [X]/10 — [one-line rationale]
2. Message Clarity: [X]/10 — [one-line rationale]
3. Persuasion Arc: [X]/10 — [one-line rationale]
4. Proof & Credibility: [X]/10 — [one-line rationale]
5. Platform Optimization: [X]/10 — [one-line rationale]
6. CTA & Conversion: [X]/10 — [one-line rationale]
TOP STRENGTH: [What's working best and why]
BIGGEST LEAK: [The #1 thing hurting performance, with a specific fix]
QUICK WIN: [One change that could lift performance with minimal effort]

### Scoring Calibration
To keep scores meaningful and consistent:
- 9-10: Top 5% of paid social ads. Exceptional on this dimension. Rarely given.
- 7-8: Strong. Clearly above average, minor room for improvement.
- 5-6: Average. Does the job but nothing special. Most ads land here.
- 3-4: Below average. Clear weakness that's likely hurting performance.
- 1-2: Broken. This dimension is actively sabotaging the ad.
Don't grade on a curve. A 5 means average, not bad. Most ads ARE average on most dimensions.

Note: For ChatGPT, you’ll want to copy the scoring criteria you build into your Project or GPT instructions. There’s no .md file needed there.

By building this, your responses from Claude or GPT will be much more thought out and require less back and forth. It’s like having a team working together to improve upon an initial draft, until there’s a polished version ready for going live.

What’ll separate marketers (including myself!) in the next 12 months is how well we can prompt in order to get the best possible outputs from AI.

Subscribe now

Read until the end for access to the full 27-page PDF guide I created.

Now that I know how crazy it can be for productivity, I don’t understand how Claude doesn’t put this front and center, or do a better job onboarding this feature.

First, what are Claude Skills?

They’re essentially mini agents that are called up in your chats depending on what you’re working on. For example, if you’re building a campaign brief and you have a campaign brief skill, it’ll pull those instructions up and execute.

Where this gets real powerful is when you have 5-10 various skills loaded into Claude, and it starts chaining them together in its responses.

Instead of getting generic responses (goodbye felicia!), you get thorough and on-the-money responses based on your Skills’ instructions.

P.S. If you comment on my LinkedIn post, I’ll DM you a 27-page PDF guide which includes all the .md files to set this up.

What did I build using this?

I specifically wanted to augment my lean startup’s marketing team, so I built 10 Skills that a general marketing team would have:

• Campaign brief generator

• Landing page builder

• Launch checklist engine

• Ad copy variant engine

• Email sequence optimizer

• Conversion copy auditor

• SEO content engine

• Content multiplier (my fav)

• Performance report builder

• Competitive intel skill

Here are example instructions for one of the Skills - the campaign brief generator:

---

name: campaign-brief-generator

description: Activates for campaign briefs, campaign plans, and new campaigns

---

# Campaign Brief Generator

You are a senior marketing strategist. When asked

to create a campaign brief, follow this framework:

## Required Inputs

- Campaign objective (awareness/conversion/retention)

- Target audience segment

- Core offer or message

- Budget range and timeline

## Output Structure

1. EXECUTIVE SUMMARY (3 sentences max)

2. MESSAGING HIERARCHY

- Primary message (1 sentence)

- Supporting messages (3 max)

- Proof points

3. CHANNEL PLAN

- Primary channels with rationale

- Budget split recommendation

4. ASSET LIST with dimensions/specs

5. TIMELINE with owner assignments

You’ll notice that the copy in the .md file needs to have a very specific writing style, starting with a name, inputs, structure. As I was building these .md files, there were many nuanced rules like the name not being able to have capitalized letters in it (i.e., Campaign-Brief-Generator returned errors).

The easiest way to build additional Skills is to copy the example above and edit it to your requirements while retaining the format.

Subscribe now

Okay cool, how do I set this up?

Claude Skills are installed using markdown (.md) files, which house all the information, rules and output formats to keep responses high quality. Instead of prompting Claude from scratch every single time you want a campaign brief or report, you install the skill once and it kicks in whenever the task comes up.

For the purpose of this essay, I’m going to teach you how to install using Claude Cowork on your desktop app, but if you’re using Claude Code or the web version, grab my PDF guide for full instructions.

1. Copy the code block above

2. Create a new .md file and paste the contents

3. Name it campaign-brief-generator.md

4. Add it to your Claude Cowork settings

5. Start a conversation with a relevant request and Claude will use the skill automatically

The biggest hiccup most will experience is in creating the .md file, so below are specific instructions on how to do that on a Mac or Windows.

On Mac

1. Open TextEdit.

2. Press Cmd+Shift+T to switch to plain text mode.

3. Paste the skill code, then File > Save As and name it ending in .md (e.g., campaign-brief.md). Make sure the format is plain text, not Rich Text.

On Windows

1. Open Notepad.

2. Paste the skill code, then File > Save As.

3. Change “Save as type” to All Files, and name it ending in .md (e.g., campaign-brief.md).

This is how it looks with your Skills installed on the Claude desktop app:

Where this gets wild

You can think of the Skills as an independent agent that’s sole job is to do that one function (like campaign brief generation) very well.

And what does an amazing marketing team do? They work together to build amazing campaigns.

Now when I go to prompt Claude, I can ask to create a campaign brief based on competitive intel on what my competitors are doing within the last few months. And to spice it up a bit, I’ll ask Claude to create a landing page for the campaign.

Here’s Claude pulling the Skills up:

Output:

How I plan to continue building

I’m looking to build on top of this with more Skills so that they can continue chaining and working together. Lifecycle, ad management, etc. As the founder of a lean team, I’m also very curious on how my other orgs can use these Skills across recruitment and operations. Is anyone interested in me writing about that? I know I mostly have folks reading because they want to get an inside look at my marketing mind, but let me know.

Want the full 27-page PDF guide, including the .md files you can literally just copy/ paste into Claude that I made? Just comment on my LinkedIn post here and I’ll DM you the guide.

Every marketer will need to become a Growth Engineer very soon. The role is becoming inherently more technical. Growth Engineers are a hybrid of growth marketer and engineer; they think less about running experiments and more about building the infrastructure that enables those experiments.

Several factors have driven this shift, but the most obvious is the proliferation of AI tools inside growth teams in 2025. I’m not just talking about ChatGPT. Tools like Lovable, Gumloop, Gamma, and Replit now allow marketers to build entire product features and landing pages through prompting alone.

This also disrupts how we think about cross-functional collaboration inside startups. Instead of relying on design, engineering, and product to launch new tests, growth teams will increasingly be expected to execute on their own. Rather than a marketer going to data science to create a Q1 performance dashboard, they’ll be expected to build it themselves. Rather than pitching a feature to product, they’ll be expected to ship it.

Share

As cost to build drops, growth engineers need to ship fast

It’s no secret that the cost to build software is collapsing. Possibly to zero. As a result, competition will increase dramatically. Growth teams will need to become leaner, faster, and far more technical in how they approach acquisition and retention. A growth engineer’s skillset will include:

• Creating AI agents for competitive intelligence and campaign optimization

• Building full product features and landing pages

• Creating complex dashboards to unify marketing data

• Connecting data pipelines for attribution and incrementality testing

• Designing high-quality performance ad creative

• Launching workflows to automate content creation

The exact skills will evolve, but the theme remains the same. Ownership from end to end. Building campaigns and systems from scratch, then iterating on them continuously.

A case study from Canva

The ability to build personalization at scale is perfectly illustrated by Canva’s work in programmatic SEO. They’ve quietly created millions of templates optimized for search. Each template (resumes, social posts and presentations) targets a long-tail keyword and pulls users directly into the editor.

They engineered a distribution system where every template is a landing page and every landing page is a product entry point. This is very different from growth marketing/ content marketing as it requires technical chops to pull off.

Instead of a traditional growth flow of search → blog → CTA → signup → dashboard, Canva took the approach of search → editable design → account creation.

Pulling this off requires technical execution across template rendering, preview images, metadata, schema, and localization across languages. I expect work like this to be owned end to end by growth teams in the very near future.

Traditional growth marketers will get replaced

A traditional growth marketer could easily ideate the concept of millions of templates acting as an acquisition and retention engine. That’s where it would stop. Building the infrastructure, maintaining the systems, and running experiments on top of those templates requires engineers who think about growth.

This example still involved close collaboration with an engineering team. But as the friction of requiring engineering hours disappears, growth teams unlock the ability to experiment far more aggressively and independently.

If startups prioritize hiring technical marketers, they gain a massive advantage over teams constantly fighting for resources to ship tests. This role is still early, but hiring managers should look for marketers who instinctively ask, “how can I automate this?” and feel an itch to build things themselves. That mindset will separate the top 1 percent of growth teams from everyone else.

A company already hiring this way

My eyes lit up when I saw a recent growth role from Ramp that clearly reflects the growth engineer archetype. Take a look at some of the requirements below from the job listing:

The posting emphasizes a bias toward shipping over spec’ing. That single line captures the shift perfectly and the next era of marketing will reward people who build, not just plan.

To summarize:

• Growth engineers are a new breed that blend growth marketing and engineering

• As build costs fall, growth teams must ship faster for distribution advantage

• Canva shows how technical growth systems can unlock massive scale

• Since the role is still new, hire marketers with an itch to build infrastructure for testing

Here’s to the next decade of growth engineers.

Subscribe now