The consumer value proposition for agentic shopping is not “let AI buy for me.” It is “help me understand what is worth buying.” Consumers already have access to more products, prices, reviews, promotions and recommendations than they can reasonably process. What they lack is confidence. They need help deciding where to save, when to spend more and what they are giving up with each choice. In an uncertain economy, that may be the first genuinely useful role for an AI shopping agent.

Consumers are managing two kinds of uncertainty

Economic uncertainty is the most immediate. The University of Michigan’s final August 2026 survey found that US consumer sentiment fell 6.3% from July and was 11.2% lower than a year earlier. Consumers remained concerned that inflation would stay elevated and that the Iran conflict would continue pushing gasoline prices higher. Year-ahead inflation expectations were 4%, compared with 3.4% in February before the Iran conflict began. The decline in sentiment was particularly pronounced among older, lower-income and middle-income consumers, who are generally less able to absorb further increases in the cost of living. This is not simply a story about pessimism. It is a story about consumers being forced to make more deliberate choices.

Read the University of Michigan August findings

At the same time, consumers are managing information uncertainty. Boston Consulting Group’s new Consumer of the Future study surveyed more than 13,000 consumers across 12 markets. It found that 43% feel mentally overwhelmed by the volume of information they encounter. More than half do not fully trust any source of information. The immediate economic pressures vary by market. The Michigan findings describe the current US environment, while BCG provides a broader cross-market view. But the underlying consumer response is similar: people are reassessing value while becoming more selective about whom they trust.

Value is not the same as price

One of the most important findings in the BCG study is that 67% of consumers would reject a product they could afford if they did not perceive sufficient value. BCG makes a useful distinction:

Price determines whether a product enters consideration. Value determines whether the consumer buys it.

Value can include quality, performance, convenience, product lifespan, service, availability, package quantity, reduced effort and trust. Only 14% of the more than 1,000 brands BCG analyzed consistently win on value. That is an extraordinary gap. Brands spend enormous amounts of money communicating prices, promotions and product features. Yet most have not clearly established why their product is worth choosing for a particular consumer in a particular situation.

Read the BCG Consumer of the Future report

What this means for agentic shopping

Most of the behavior measured by BCG today is AI-assisted rather than fully agentic. Consumers are using AI to discover products, compare alternatives and make decisions. They are not yet routinely allowing agents to complete purchases autonomously. But the findings reveal the consumer job that future shopping agents will need to perform before consumers grant them greater authority. That job has three parts.

1. Help me control the total cost

A useful agent would do more than find the lowest advertised price. It could evaluate:

• Unit economics

• Product lifespan

• Expected waste

• Delivery costs

• Loyalty benefits

• Replacement frequency

• Returns and warranties

It could tell the consumer where the cheaper choice is sufficient, where paying more provides a meaningful benefit and when an apparent bargain may become more expensive over time. That is a more compelling proposition than simply pressing the buy button.

2. Help me determine what fits my situation

BCG’s findings on solo living show why context matters. Approximately 30% of households in mature markets consist of one person. Solo households spend two to three times more per person, yet 50% to 70% say products and services are not designed for people like them. These consumers often value appropriate package sizes, reliable availability and reduced waste more than bulk promotions. A conventional shopping algorithm might recommend the largest package because it offers the lowest unit price. An agent that understands the consumer’s situation might reach a different conclusion. It could recognize that the larger package requires a higher immediate outlay, creates storage problems or results in waste. The lowest unit price is not always the best value. Value does not live inside the product. It emerges from the fit between the product, the consumer and the moment.

3. Help me trust the decision

BCG found that AI introduces consumers to brands they would not otherwise have considered in approximately 63% of AI-assisted purchase journeys. That means AI is not simply searching an existing consideration set. It is increasingly helping construct it. But consumers are not surrendering control without conditions. Sixty percent still limit the personal or sensitive information they share with AI. Consumers may delegate the search before they delegate the decision. They may delegate the decision before they delegate the payment. For an agent to earn greater authority, it must show its work. The consumer should be able to understand:

• Why the recommendation was made

• Which criteria influenced the result

• What evidence supports it

• What alternatives were considered

• What tradeoffs are involved

• Whether the result was sponsored

• What personal information was used

A shopping agent cannot ask consumers to surrender control and then hide how it reached the answer. Without transparency, agentic shopping could make the trust problem worse. It would become another opaque intermediary deciding what consumers see and which brands receive consideration.

The assignment for brands

If agents increasingly evaluate value on behalf of consumers, brands must make their value proposition understandable in context. The agent needs to determine:

• Who is this product for?

• What tension does it resolve?

• In which moments is it most relevant?

• What benefit does it deliver?

• Why is that benefit differentiated?

• What evidence supports it?

• When is it worth paying more?

This is bigger than product feeds or search optimization. It is a concept-development assignment. Generic promises such as “high quality,” “convenient” or “great value” will not provide enough information. Brands need to articulate the consumer tension, the specific benefit and the reason to believe. Imagine a company developing a new prepared-meal concept. For a family under financial pressure, value might mean replacing an expensive takeout occasion with a meal everyone will eat. For someone living alone, value might mean the right portion without leftovers, waste or a delivery minimum. For a risk-averse consumer, value might mean dependable quality, familiar ingredients and confidence that the meal will turn out properly. The product category is the same yet the underlying tension and definition of value are different.

Where The Yogi AI fits (www.theyogiai.com)

This is one of the problems we are addressing with Yogi AI. Yogi AI uses consumer context to generate multiple concept routes built around a specific audience, tension, benefit and reason to believe. For the prepared-meal example, it could explore three different strategic routes:

1. Stretch the household budget without sacrificing the family meal.

2. Give one-person households the right amount without waste.

3. Reduce the risk of spending money on a meal that disappoints.

The objective is not simply to create more concepts. It is to develop concepts that clearly express why the product is relevant and worth choosing in a particular context. Yogi AI does not prove that a concept will succeed. Consumer validation still matters. It also cannot guarantee that a shopping agent will recommend the product. But it can help brands move beyond generic promises and systematically develop value propositions that consumers can understand and AI systems may eventually need to interpret.

The first battle will not be checkout

Consumers will not delegate more of the purchase journey simply because an agent can transact. They will delegate when the agent consistently helps them make better decisions. For brands, that raises the standard. Being discoverable will not be enough. The brand’s value must be clear, contextual and supported by evidence that consumers and AI can evaluate. The first battle in agentic shopping will not be over who controls the checkout. It will be over who consumers trust to explain what is worth buying.

There are currently no openly gay male professional footballers playing at the top of the game. Not because people don’t accept them, but because everyone assumes everyone else doesn’t. It’s a dynamic that exists across markets and cultures, but one the industry hasn’t examined closely enough. Bolt Insight used AI moderation across a global study to bring it into focus. In this episode, we unpack the “Honesty Gap”, what it took methodologically to find it, and what 2026 looks like for an insights industry still figuring out how to adapt, upskill, and make AI work in practice not just in theory.

Hakan Yurdakul

CEO and Co-Founder, Bolt Insight

Hakan Yurdakul is the CEO and co-founder of Bolt Insight, an AI-native consumer intelligence company combining AI research platforms with full-service human expertise. With 20+ years on the brand side, including senior roles within Unilever’s marketing function, he built Bolt on the conviction that AI surfaces correlations but human judgment delivers causation. Bolt serves 150+ clients across 100+ markets, has conducted 15 million+ consumer interviews and won Best Online Qual at the MRS Awards.

This week provided some of the clearest evidence yet that agentic commerce is moving beyond product discovery.

Walmart is showing how an owned shopping agent can affect basket economics. Microsoft is turning AI visibility into a measurable commerce discipline. Radisson is preparing its hotels to qualify for AI-generated consideration sets. Healthcare is providing a practical model for multi-agent orchestration. Google’s interest in Spirit Airlines data points to an emerging need for operational trajectories, not simply more consumer profiles.

These developments come from different industries, but they are connected.

AI systems are beginning to combine consumer objectives, product information, prior behavior, availability, business rules and transaction capability. The role of the agent is shifting from helping people find options to helping them make and execute decisions.

That changes what brands need to prepare for.

Walmart is showing us what happens when an agent understands the basket

Walmart reported strong ecommerce growth in its second-quarter FY27 results. Global ecommerce grew 23%. Walmart U.S. ecommerce grew 24%. Store-fulfilled delivery grew 40%, and marketplace net sales grew more than 50%.

The most interesting number, however, came during the earnings call.

Walmart said the number of people using Sparky increased 70% year over year. Customers using Sparky also spend approximately 40% more per order than customers who do not.

That is a substantial difference, but it needs to be treated carefully. Walmart has not shown that Sparky causes customers to spend 40% more. Higher-value and more digitally engaged customers may simply be more inclined to use the tool. The number is commercially important, but it should not be presented as proven incrementality.

The more meaningful signal is what Sparky is learning to do.

Walmart described a consumer asking Sparky to build a high-protein weekly meal plan. Sparky develops the plan, identifies the required ingredients and adds them to the basket with one click. It can also remember what the customer previously purchased, online or in a store, so it does not recommend products the household may already have.

That is no longer a simple product recommendation.

It is basket orchestration built around a consumer objective. The agent combines the desired outcome, behavioral history, product availability and transaction capability within one workflow.

The memory component may ultimately be more important than the reported 40% basket difference. A useful agent should not simply keep adding products. It should know when an item is unnecessary, recognize likely household inventory, make appropriate substitutions and understand which adjacent categories genuinely support the consumer’s objective.

This creates a much more interesting research agenda for retailers and brands. We should be measuring:

• Incremental items added by the agent

• Products removed because the agent remembers prior purchases

• Movement into adjacent categories

• Acceptance and rejection of substitutions

• Changes in average basket value

• Purchase completion

• Repeat use

• The point at which consumers are prepared to delegate the entire basket

These measures tell us whether the agent is improving the shopping decision. Merely measuring whether a brand appears in an AI-generated answer will not.

Read Walmart’s Q2 FY27 earnings announcement

AI visibility is becoming measurable, which also means it is becoming table stakes

Microsoft released an AI-shopping playbook this week that deserves more attention than a typical platform announcement.

Its emerging commerce model connects Merchant Center product feeds, AI-native advertising, Copilot Checkout, website-based Brand Agents and AI Visibility insights within Microsoft Clarity.

The Clarity capability is designed to help businesses understand which AI platforms cite them, the queries and topics that produce those citations, how their visibility compares with competitors and what AI-referred consumers do once they arrive on the company’s website.

This is an important step because it connects AI visibility with downstream consumer behavior.

A year ago, an AI visibility audit could legitimately involve running dozens of prompts across ChatGPT, Gemini and Perplexity, recording which brands appeared and diagnosing the apparent gaps.

That capability is rapidly becoming table stakes. It may also become commoditized.

Microsoft is productizing visibility measurement. Shopify is exposing catalog, cart and checkout capabilities through MCP and UCP infrastructure. Other platforms will inevitably build similar measurement and transaction tools.

The more valuable commercial question is becoming:

Why is the brand being selected or excluded, and what commercial outcome results?

This requires moving beyond AI visibility toward what I would call machine decision readiness.

Can the agent understand the product? Can it interpret the claims correctly? Is there enough evidence to support those claims? Does the agent recognize a meaningful difference from competing products? Can it determine when the product is appropriate for a particular consumer, occasion or constraint? What happens after the recommendation is made?

Microsoft argues that structured and complete product information will become a critical marketing asset because agents make decisions using facts and structured data, not conventional brand storytelling alone.

I agree, with one important qualification. Structured data will help an agent understand what a product is. It will not automatically help the agent understand when the product matters.

That requires context.

A brand needs to connect its product information and claims to the needs, moments and trade-offs that shape the consumer’s decision. This is where AI Shelf Readiness, the Moments Engine and Y2S Claims begin to come together. The goal is not simply to make the brand visible. It is to make the brand understandable, credible and appropriate for the decision the agent is trying to make.

Read Microsoft’s AI-shopping playbook

Travel makes the difference between discoverable and decisionable easier to see

Radisson Hotel Group is explicitly building “agentic readiness” into its commerce strategy.

In an interview published this week, Radisson’s vice president of global ecommerce said hotels will increasingly need to be understandable, trustworthy and bookable by AI systems.

Each of those words matters.

Being understandable requires accurate information about the property, rooms, amenities, location and guest experience. Being trustworthy requires credible content, consistent policies and dependable rates and availability. Being bookable requires the technical connections that allow an agent to check inventory, apply loyalty value and complete the transaction.

For years, hotels have competed to rank in a search result. In agentic commerce, they will increasingly compete to qualify for an agent-generated consideration set.

An AI assistant may need to compare hotels, check availability, incorporate traveler preferences, interpret cancellation policies, calculate loyalty benefits and complete a reservation. The hotel does not qualify merely because its page ranks highly or its content has been optimized around the right search terms.

The agent needs enough structured evidence to determine:

• Whether the property is appropriate for this particular trip

• Whether it is actually available

• Which restrictions or conditions apply

• Whether the price is credible

• What meaningfully differentiates it

• Whether the booking can be completed

• What will happen if the traveler’s plans change

GEO and AEO help a brand become discoverable. Agentic commerce requires the brand to become decisionable.

This distinction applies well beyond travel. A consumer-products brand may be visible to an AI assistant and still be excluded from the final recommendation because its claims are vague, the evidence is weak, the product data is incomplete or the agent cannot connect the product to the consumer’s specific need.

Claims therefore become more than persuasive language. They become inputs into machine decision logic.

Read the Radisson interview

The consumer is interested, but trust has not caught up with the infrastructure

Research presented by Knit at TMRE adds an important consumer dimension to these developments.

In its research among 534 consumers, 67% viewed AI as the future of product discovery. Consumers were particularly interested in AI integration and product recommendations, and 87% considered quick refinement important when using AI.

The findings suggest that consumers already see distinct roles for different discovery channels.

Search continues to provide depth and confidence. Social media provides inspiration, demonstrations and social proof. AI helps consumers refine choices, compare alternatives, obtain tailored responses and move more easily toward a decision.

Consumers do not appear to expect AI to replace every other channel. They expect it to connect and improve the experience across them.

Trust remains the constraint. Google and Amazon continue to command considerably more confidence than AI tools. Consumers raised concerns about accuracy, the source of the information, the degree of control AI may eventually exercise and the potential for commercial manipulation.

The research also identified early signs of AI fatigue. Among Gen Z respondents, reported usage fatigue was 27% for TikTok, 30% for Instagram and 29% for ChatGPT.

This is a useful warning. Consumers may want AI support, but they do not want another channel filled with low-value commercial content. If every AI answer becomes a disguised advertisement, the industry will recreate the trust problems already affecting social discovery.

The opportunity is therefore not to maximize the number of AI recommendations. It is to improve the value and credibility of the decisions those recommendations support.

Healthcare provides a more realistic model for multi-agent systems

Raintree Systems introduced two agentic suites this week for rehabilitation healthcare.

Agentic PX is intended to support front-desk and patient-experience workflows. Agentic RCM focuses on revenue-cycle management.

The architecture is more important than the product announcement.

The billing team remains in control. An orchestration layer determines which agent should act next. Specialized agents then perform bounded tasks, using a shared system of record.

The first eligibility agent, called Lucy, can navigate payer phone systems, access portals, process clearinghouse data, conduct benefits conversations and write verified information back into the electronic medical record. Raintree says agents supporting prior authorization, claims, denials and payments will follow in 2027.

Raintree has also published potential efficiency and revenue improvements. These are vendor targets, not independently validated outcomes, so I would not treat them as proven ROI.

The architecture itself is the signal:

Human governor → orchestrator → specialist agents → shared system of record

This is much closer to the likely enterprise model than one general-purpose chatbot attempting to perform every task.

It also creates a larger testing challenge.

What happens when the eligibility agent receives contradictory payer information? What happens when an authorization agent changes the patient’s status? What if one agent makes a locally correct decision that creates a downstream failure? When should the system continue autonomously, and when must it escalate to a person?

Testing one agent against one synthetic consumer will not answer these questions.

Enterprises will need to stress-test agent handoffs, state changes, exceptions, escalation rules and downstream consequences. That is not simply synthetic research. It is synthetic system testing.

The market opportunity could be substantially larger.

Read the Raintree announcement

Agents need process trajectories, not just personas

A final development from travel may be the most important signal this week for the future of synthetic data.

Google has proposed paying $10 million for rights to data from the defunct Spirit Airlines. The court filing was made on August 14, so the transaction itself is not new this week. New reporting from PhocusWire, however, provides more detail about why the data could matter.

The reported dataset includes operational, financial, communications, revenue, website analytics and loyalty information. Industry experts noted that booking curves, pricing, refunds, vouchers and disruption records could help train models to understand how decisions and consequences unfold across an airline operation.

The data is historical, and Google says it will be deidentified. That does not eliminate the privacy and reidentification questions, which will need careful scrutiny.

The strategic implication extends beyond this particular transaction.

Much of the current synthetic-data conversation is still focused on creating synthetic personas or respondents. Those approaches may be useful for some research applications, but agentic systems need more than static representations of consumers.

They need process trajectories:

Situation → decision → system action → changed state → exception → resolution → outcome

In retail, that trajectory might be:

Item unavailable → substitute selected → consumer preference conflict identified → alternate retailer evaluated → delivery window changed → transaction completed

In travel:

Flight canceled → alternatives priced → loyalty status considered → hotel reservation affected → refund initiated → itinerary rebuilt

In healthcare:

Coverage checked → authorization denied → alternative treatment identified → patient notified → appeal initiated

These sequences contain decisions, dependencies and consequences. They show how an action taken at one stage changes the available options at the next.

This type of longitudinal operational data may become much more valuable for training and validating agentic systems than another thousand static synthetic respondents.

Read the PhocusWire analysis of the Spirit Airlines data

The strategic takeaway

Consumer adoption will not move in a straight line. People are interested in AI-assisted discovery and decision support, but trust, accuracy, transparency and control still determine how much authority they are prepared to delegate.

The commercial infrastructure is not waiting for those issues to be fully resolved.

Retailers are building agents that can orchestrate baskets. Technology platforms are measuring AI visibility and connecting it with checkout. Hotels are preparing their inventory and policies for AI-mediated booking. Healthcare companies are building orchestrated networks of specialist agents. Technology companies are seeking operational data that can teach systems how decisions unfold over time.

The market is moving from:

Can the AI find the brand?

to:

Can the AI understand the need, evaluate the evidence, select the brand and execute the decision?

That is a considerably higher standard.

It also changes the work ahead for marketing, insights and ecommerce teams. Product feeds, claims, consumer moments, behavioral data, inventory, pricing, policies and transaction systems can no longer be treated as separate assets managed by separate functions. Together, they form the information environment in which an agent decides whether a brand belongs in the consideration set. AI does not simply need to know that your brand exists.

It needs enough context and confidence to choose it.

The new RTB House research on consumer trust in agentic AI is one of the more useful studies I’ve seen because it gets beyond the question of whether consumers will use AI for shopping. It starts to show us what consumers are willing to delegate, and under what conditions. RTB House found that:

• 59% of U.S. consumers say AI helps them discover brands they did not previously know.

• 63% have used AI to turn a shopping need into at least a general list of products.

• 42% also say AI increases the time it takes them to make a final decision because it gives them more information, more choices and more brands to consider.

That is an important contradiction. We keep talking about AI compressing the funnel. It may eventually compress the transaction. But right now, AI appears to be expanding consideration. And that creates an interesting opportunity for brands.

AI can introduce consumers to brands they may never have searched for. That is particularly important for challenger brands trying to break into an established consideration set.

And this is not just a Gen Z story. In the U.S., RTB House found that 62% of Baby Boomers and 61% of Gen X say AI introduces them to unfamiliar brands, compared with 52% of Gen Z. But discovery is only the first hurdle. The more interesting part of the research is what happens when AI moves from recommending to buying. Consumers are not rejecting that idea. They are putting rules around it.

• 35% want human approval before checkout.

• 31% want the ability to cancel immediately after purchase.

• 29% want purchases limited to trusted retailers.

• 29% want a strict price limit.

• 28% want free or simple returns.

• 42% of U.S. Millennials say they would give an AI agent up to $250 to make a purchase on their behalf if there were a seven-day return window.

That doesn’t look like blind trust. It looks like bounded delegation. You can recommend for me, you can narrow the options, you can even buy for me but only within the rules I establish. This is where I think the conversation becomes much bigger than SEO, AEO or GEO.

BCG recently argued that companies increasingly have to account for two decision makers: the human consumer and the AI systems influencing that consumer. They also make an important distinction between being visible to AI and being credibly recommended by AI. AI does not simply find brands, it evaluates them - pricing, availability, reviews, fulfillment, service and customer experience, all the evidence behind the brand promise.

Put the RTB House and BCG findings together and the implication becomes much clearer. Agentic commerce becomes a brand management accountability. Brands are going to have to earn more than awareness. They need to earn:

Discovery -> Recommendation -> Trust -> Permission -> Delegation.

And each step raises the bar. A return policy is no longer just an operational policy. Retailer relationships are no longer just distribution. Pricing consistency is no longer just revenue management. Reviews, product information, fulfillment and customer service are no longer separate functional metrics. They increasingly become part of the evidence an AI system and a consumer use to decide:

Is this brand safe enough to choose?

And eventually:

Is this brand safe enough for me to let AI choose on my behalf?

That is why I think brands need to stop looking at agentic commerce only through the lens of visibility. Visibility gets you into the consideration set. It does not earn the purchase. And it certainly does not earn delegation. The strategic question I would be asking now is:

What is the next decision our customer is willing to delegate to AI, what boundaries will they put around that decision, and what does our brand have to prove to earn its place inside those boundaries?

That is a much bigger brand question than whether you show up in an AI answer. Are you seeing something different? If so, what am I missing?

RTB House, Who’s Buying? Consumer Trust in the Age of Agentic AI https://www.rtbhouse.com/resources/whos-buying-consumer-trust-in-the-age-of-agentic-ai

What Astride Health’s experience reveals about accuracy, adaptability and the difference between producing research and changing a business.

You can know every one of your customers and still lack a clear growth strategy.

This happens frequently in young businesses. Leaders speak with customers every week. They hear detailed stories, recognize recurring concerns and develop strong instincts about who values the proposition. Yet when the team has to decide which customers to prioritize, those individual stories begin to compete with one another.

The most recent conversation carries too much weight. An unusual customer is mistaken for a larger opportunity. The target audience gradually expands until it describes almost everyone.

Astride Health faced a version of this challenge.

The company provides a proactive approach to healthcare that combines advanced diagnostics, medical expertise, exercise physiology and ongoing support. With approximately 80 members, the team knew its customers well as individuals. What it did not yet have was a structured view of the wider market or a clear answer to a fundamental growth question:

Which consumers represented the strongest fit for Astride’s model?

I recently discussed that journey with Peter Everett, Astride Health’s Chief Growth Officer and a former General Mills executive. Peter entered the segmentation process with two reasonable concerns.

Could an AI-enabled segmentation be accurate enough to guide the business?

And even if the output was interesting, would it change an actual decision?

Those questions matter because segmentation has traditionally been judged primarily by the quality of its methodology. Researchers examine the sample design, variables, clustering, statistical stability and size of the resulting segments.

Leadership teams eventually apply a more direct test:

Did the segmentation help us decide where to compete, and did we do anything differently because of it?

From customer anecdotes to strategic choices

The AI-enabled process identified five consumer segments that could potentially engage with Astride’s proposition.

The value did not come from producing five polished profiles. It came from creating a structured set of alternatives that the leadership team could examine, challenge and compare.

Two segments could be deprioritized relatively quickly. Three represented plausible growth opportunities, but for different reasons. Each had distinct motivations, expectations and barriers. Each would require Astride to tell a different story about the value of its service.

This transformed the conversation. The team was no longer debating a collection of individual customer stories. It was evaluating three different strategic paths.

Astride’s physicians, scientists and exercise physiologist began recognizing behaviors and language they had encountered among existing members. The segments helped organize patterns that members of the team had observed but had not previously been able to connect.

That recognition should not be confused with validation. People can recognize themselves or their customers in a well-written description even when the underlying model is weak. The segmentation still needed to withstand scrutiny.

But when specialists from across the organization can connect a model’s distinctions to recurring behaviors, and those distinctions lead to a coherent strategic choice, that recognition becomes meaningful supporting evidence.

The team was not simply saying, “These segments sound realistic.”

It was saying, “We have seen these needs and behaviors before, and they have implications for how we should grow.”

The decision was more important than the segments

Astride ultimately selected one priority segment.

The company did not choose it because every existing member belonged to it. Nor did it choose the largest or most attractive-sounding audience.

Three elements aligned.

First, the segment had an important and recognizable need. Second, Astride’s service model was particularly well suited to addressing that need. Third, the company’s experience with existing members suggested that people with this profile responded positively to the proposition.

That combination made the segment strategically credible.

The decision then began to affect how Astride operated. It influenced communications, channel choices, aspects of the member experience and the language the team used to describe the brand.

During the process, one of Astride’s physicians proposed the line, “Your next project is you.” The company subsequently used it because it captured something important about the chosen audience: these were people accustomed to investing time and energy in meaningful projects, but who had not always treated their own health with the same intention.

This is where the segmentation moved from research into the business.

A technically sophisticated segmentation can still fail if it does not change priorities, investment or execution. Conversely, a segmentation becomes valuable when people across the organization can use it to make better choices.

The output is not the segment profile. The output is organizational alignment around a customer and the decisions required to win that customer.

Confidence is earned through challenge

There is a temptation to discuss AI-enabled research as if a model produces an answer that the organization either trusts or rejects.

That is not how confidence was established here.

Confidence accumulated as the work survived a series of challenges.

Did the segments reflect the realities of the Twin Cities market rather than a generic national consumer? Did they make sense in the context of a healthcare proposition for which limited direct public data existed? Could the team connect the distinctions to behaviors it had observed? Were the segments sufficiently different to imply different business choices? Could the leadership team explain why one audience should take priority over the others?

This is a more realistic standard for using AI in strategic work.

The goal is not to eliminate human judgment. It is to give human judgment something structured and explicit to interrogate.

The model proposes patterns. The leadership team challenges them. Domain experts test them against experience. Contradictory evidence is investigated rather than ignored. The resulting decision reflects both computational analysis and informed business judgment.

This is also why the process cannot be reduced to “asking an AI to create some personas.” A list of plausible customer descriptions is not a segmentation strategy. The work has to distinguish meaningful alternatives, connect those alternatives to business implications and support a defensible choice.

Segmentation can become a revisable decision system

One of the most important moments in the Astride project occurred when life stage emerged as more influential than we had initially expected.

In a conventional segmentation, introducing an additional dimension can require revising the questionnaire, returning to the field and repeating substantial portions of the analysis. That makes adaptation expensive. It also encourages organizations to treat the original segmentation as fixed, even when the market or the business has changed.

With the AI-enabled approach, we could reintroduce life stage into the system, rerun the analysis and examine how the segments changed.

This suggests a larger shift in how companies can use segmentation.

Segmentation has traditionally been treated as a map of the market. The organization completes the research, selects its target segments and uses the same framework for several years.

AI-enabled segmentation has the potential to operate more like a revisable decision system. New evidence can be incorporated. Assumptions can be challenged. Alternative variables can be tested. The model can evolve as the market, category and organization change.

That does not mean the answer should be continually altered until leadership receives the result it wants. Adaptability without governance can create a different problem: strategic instability.

Every revision should be tied to a clear question, new evidence or a meaningful change in market conditions. The organization also needs to document what changed, why it changed and whether the resulting decision remains comparable with previous work.

The advantage is not that AI makes segmentation permanently fluid. The advantage is that it makes disciplined revision more feasible.

Where AI-enabled segmentation still needs support

The Astride experience should not be interpreted as evidence that AI can replace primary research in every segmentation.

The reliability of an AI-enabled approach depends partly on the quality, relevance and diversity of the evidence available to the system. The approach becomes more vulnerable when customer behavior is poorly represented in accessible information, when a category is discussed primarily by professionals rather than consumers, or when people rarely talk openly about the need being studied.

Healthcare presents several of these challenges. Public information can be dominated by clinical language, while the emotions, tradeoffs and everyday behaviors that shape consumer decisions remain less visible.

There are also questions that an AI-enabled segmentation cannot answer reliably on its own. If a company needs precise segment incidence, revenue sizing or statistically projectable estimates of market potential, additional primary research is required.

The same is true when the decision carries major financial, regulatory or safety consequences. The greater the consequence of being wrong, the stronger the validation standard should be.

The answer is not to choose between AI and traditional research as competing ideologies. It is to match the evidence to the decision.

AI and agentic workflows can help develop an initial market structure, explore competing hypotheses and refine the segmentation. Client expertise can provide a disciplined challenge to those hypotheses. Targeted qualitative or quantitative research can then address the areas where uncertainty remains. A shorter typing study can be used when reliable segment sizing or customer classification is required.

This hybrid approach focuses research investment where it creates the most decision value.

The real test of a segmentation

The Astride project changed how I think about the standard by which segmentation should be evaluated.

Methodological rigor remains important. But rigor is not the end product.

A leadership team should be able to answer five questions:

1. Which audience are we prioritizing?

2. Why did we select that audience over the alternatives?

3. What will we do differently in communications, experience, innovation or investment?

4. What evidence supports the decision, and where does uncertainty remain?

5. What new evidence or market change would cause us to revisit the segmentation?

If the organization cannot answer those questions, it may have a segmentation study. It does not yet have a segmentation strategy.

AI can accelerate the analysis, widen the set of hypotheses and make the segmentation easier to revisit. But its value is ultimately determined by something far less technical.

Did it help the leadership team make a choice it could understand, defend and execute?

That is the decision AI-enabled segmentation must earn the right to influence.

#CustomerSegmentation #AIinMarketing #MarketResearch #ConsumerInsights #GrowthStrategy #AgenticAI

Over the past two years, the conversation around synthetic data has changed dramatically. The debate is no longer about whether synthetic data belongs in market research. Most organizations have accepted that it does. The questions have shifted toward validation, correlation with human respondents, and identifying where synthetic approaches can accelerate learning.

I attended Quirk’s New York this week and spent most of my time in the synthetic data sessions. While the presentations covered different use cases and came from different organizations, I left with one consistent observation.

The market research industry has become increasingly focused on synthetic personas.

That isn’t necessarily a problem. Synthetic personas have an important role to play. The challenge is that they are starting to represent the entire synthetic data conversation, when in reality they are only one methodology within a much broader field.

I think that distinction matters because it is beginning to shape expectations of what synthetic data can and cannot do.

Today, most discussions seem to follow a familiar pattern. Organizations start with survey data, use that data to generate synthetic personas, ask those personas the same questions they would ask human respondents, and then compare the outputs. We compare purchase intent, concept scores, claims testing, brand perceptions, and correlations with traditional research.

Those are all worthwhile exercises. Validation is essential if organizations are going to trust these approaches. But notice where most of the discussion has been concentrated.

We debate the quality of the source data.

We debate the quality of the outputs.

What we rarely debate is the algorithm sitting between them.

That is where I think the industry is missing an important opportunity.

When people talk about synthetic data, they often describe it as though it were a single technology. It isn’t. Synthetic data is better understood as an ecosystem of different methodologies, each designed to answer different kinds of business questions.

Persona generation is one methodology.

Agent-based modeling is another.

Digital twins represent another.

Generative Adversarial Networks (GANs), probabilistic simulation, hybrid behavioral models and reinforcement learning approaches all represent different ways of constructing synthetic systems.

Each begins with different assumptions.

Each requires different inputs.

Each produces different outputs.

More importantly, each is designed to solve different problems.

I don’t think we’ve been making that distinction clearly enough.

Instead, I think the industry has unintentionally anchored itself on one architecture. We have become so focused on synthetic personas that we’ve started evaluating the entire field through that single lens.

That creates a mismatch between expectations and capability.

If a synthetic persona struggles to predict complex market dynamics, that doesn’t necessarily mean synthetic data has failed. It may simply mean we are asking one methodology to solve a problem better suited to another.

Likewise, if synthetic personas perform well in concept screening, hypothesis generation or early-stage exploration, it doesn’t automatically mean they should be used to answer every research question.

Methodology matters.

The business problem should determine the methodology, not the other way around.

One conversation at Quirk’s reinforced this idea for me. I met Millie Marconi from Australia, who described work using agent-based modeling for Shopify. What caught my attention wasn’t simply the modeling technique. It was the starting point. Rather than beginning with survey responses, the system begins with behavioral data and then supplements it with survey data where appropriate.

Whether that ultimately proves to be a better approach isn’t really the point.

The point is that it represents a completely different architecture.

One model starts by asking what consumers say.

The other starts by observing what consumers do.

Those are fundamentally different ways of building synthetic systems.

That conversation made me realize that we may be spending too much time debating source data and validation metrics while giving comparatively little attention to the algorithms that actually transform one into the other.

Interestingly, this also caused me to reflect on my own Quirk’s presentation. I argued that synthetic consumers have four important jobs: testing scenarios before real-world data exists, simulating interactions, surfacing hypotheses, and helping prioritize where research investment should go. I still believe those are valuable applications.

What I would add today is that not every algorithm is equally suited for every one of those jobs.

A persona engine may be highly effective for generating hypotheses or exploring reactions to a concept.

An agent-based model may be better suited to understanding how consumers, AI shopping agents, retailers and brands interact over time.

Another methodology may be better suited to forecasting market evolution or testing competitive scenarios.

The algorithm matters because it determines what questions the model is capable of answering with confidence.

I suspect the next phase of synthetic data will be less about building increasingly realistic personas and more about matching the right modeling architecture to the right business problem. That feels like a much more productive direction for the industry than asking whether synthetic respondents can replace human respondents.

As market researchers, we’ve always selected different analytical techniques depending on the question we were trying to answer. We don’t use conjoint analysis for every business problem. We don’t use segmentation for every strategic decision. We don’t use ethnography to answer every pricing question.

Synthetic data should be no different.

The future of synthetic research won’t be defined by one methodology becoming dominant.

It will be defined by understanding when to use synthetic personas, when to use agent-based modeling, when to build digital twins, when hybrid approaches make more sense, and how these methods can work together.

I have a feeling we’re only at the beginning of that conversation.

Much of the discussion around AI commerce has focused on SEO, GEO, AEO, structured data, and content optimization. Those capabilities matter, and every brand should be improving them. But I believe they’re solving a downstream problem. AI commerce isn’t fundamentally changing how brands get found. It’s changing how brands get chosen. That’s why I believe AI commerce is ultimately a brand management accountability. Three observations have led me to that conclusion.

First, consumers aren’t using AI as a better search engine.

Throughout this series, I’ve argued that consumers are using AI to help resolve uncertainty. They aren’t typing keywords. They’re describing situations, preferences, constraints, trade-offs, and goals. They want recommendations, not results.

That changes the job of the LLM. Instead of matching keywords to web pages, it has to understand the consumer’s context, determine what matters, evaluate competing brands, and decide which recommendation best fits the situation. The starting point isn’t search. It’s consumer decision-making.

Second, AI doesn’t evaluate marketing assets. It evaluates brands.

When an LLM decides whether to recommend a brand, it doesn’t stop at a product page or a website. It pulls together evidence from wherever it can find it: positioning, product descriptions, retailer content, reviews, pricing, availability, customer service, expert opinions, social conversations, and countless other signals. Consumers experience one brand. AI evaluates one brand. Only organizations separate those signals across marketing, ecommerce, customer service, product, operations, and supply chain. That distinction matters because AI isn’t judging how well you’ve optimized a page. It’s judging whether your brand is the best answer for the consumer standing in front of it.

Third, brand management now extends beyond consumer perception.

Traditionally, brand managers were responsible for shaping how consumers understood the brand. That responsibility isn’t going away, but it’s expanding. Brands now need to be understood by two audiences simultaneously: consumers and AI. That requires a level of consistency that many organizations weren’t designed to deliver. Positioning, claims, product information, reviews, retail execution, customer experience, and commercial content all contribute to how AI understands the brand. When those signals reinforce one another, AI gains confidence. When they conflict, confidence erodes. That’s not an SEO issue. It’s a brand management issue.

The implication

If your organization approaches AI commerce as an extension of search, improving discoverability will almost certainly improve performance. But discoverability alone won’t determine which brands AI recommends. Recommendation depends on confidence. And confidence depends on how coherently your brand is represented across every signal AI evaluates.

That’s why I believe the next chapter of AI commerce belongs with brand management. SEO, GEO, and AEO will remain important capabilities, but they’re enablers of a broader objective. The real challenge is ensuring that every part of the organization contributes to a brand that AI can confidently understand, explain, and recommend.

If you’re thinking about how your brand will compete in AI commerce, you’ve probably started asking questions about SEO, AEO, or GEO. Those are important capabilities because they help AI discover your content and retrieve information about your brand. But as AI increasingly moves from finding products to recommending them, another challenge begins to emerge.

Recommendation requires understanding. Before AI can confidently recommend your brand, it has to answer two very different questions. The first is whether it understands your brand well enough to know who it’s for, what problem it solves, how it’s differentiated, and why someone should believe its claims.

The second question is just as important. Does AI understand the consumer well enough to know what they’re actually trying to accomplish? What moment are they in? What trade-offs are they making? What constraints matter? What does success look like from their perspective? Those are two separate forms of understanding, yet they’re inseparable when it comes to making a recommendation.

Think about how you make decisions as a marketer. You wouldn’t recommend a premium anti-aging moisturizer simply because someone asked for skincare. You’d first want to understand who they are, what concerns they’re trying to address, how they currently shop, what they’ve already tried, how much they’re willing to spend, and what outcome they’re hoping to achieve. Only then would you decide which brand is the best fit.

AI is beginning to make decisions the same way. That has changed how I think about AI commerce. I’ve come to believe that brands need to prepare for two different kinds of intelligence.

The first is making the brand understandable to AI. Can AI consistently infer what your brand stands for, when it should recommend it, and why it deserves to be chosen?

The second is helping AI understand the consumer’s context. Recommendations don’t happen in a vacuum. They happen within moments, where goals, emotions, constraints, and circumstances shape what “best” actually means.

Those two questions have become the foundation for much of the work I’ve been doing over the past year.

One focuses on whether a brand is ready to be understood by AI. I call that AI Shelf Readiness™.

The other focuses on understanding the consumer’s moment before making a recommendation. I call that The Moment Engine™.

Neither framework exists to optimize search. Both exist to improve judgment. Because the future of AI commerce won’t be determined by whether AI can find your brand. It will be determined by whether AI understands both your brand and your consumer well enough to confidently bring them together.

I’ve been thinking about why some AI conversations feel incredibly valuable while others feel almost unnecessary. I don’t think it has much to do with which AI platform we’re using. I think it has everything to do with the moment that brought us there.

Nobody wakes up in the morning and decides, “I’m going to use ChatGPT today.” They wake up with a problem they need to solve.

• A family is planning their first vacation to Europe with young children. A son is trying to help his elderly mother book her first flight since losing her husband.

• A homeowner has a dishwasher break two days before relatives arrive for the holidays.

• A parent is standing in the baby aisle after their infant reacted to a new formula.

The moment comes first. AI is simply invited into the decision. I think that’s an important distinction because it changes how we think about AI commerce.

For years, we’ve asked which channels consumers use during the purchase journey. Increasingly, I think the better question is which moments naturally invite AI into the decision. Those moments aren’t all created equal. Some decisions are simple. Others carry real consequences. Some require almost no context. Others require AI to understand budgets, family dynamics, health concerns, previous purchases, travel preferences, timing, or dozens of other variables before the recommendation feels credible.

The research reflects this. Morning Consult found that while AI adoption continues to grow, nearly two-thirds of Americans still trust AI only a little or not at all. At the same time, The New Consumer found that younger consumers are increasingly willing to give AI access to purchase history, email, financial information and even health data if it leads to better recommendations.

At first glance, those findings seem contradictory. I don’t think they are. Consumers aren’t giving AI permission because they trust AI unconditionally. They’re giving AI more context because they believe better understanding leads to better recommendations.

That becomes even more important as decisions become more complex. It’s probably not a coincidence that many of the first AI agents and connected services are appearing in categories like travel, finance, productivity and real estate. Those aren’t simple purchase decisions. They’re situations where consumers naturally expect to answer questions before receiving advice, where the cost of getting it wrong is high, and where better context can dramatically improve the outcome.

Nobody expects AI to ask twenty questions before recommending paper towels. They absolutely expect it before recommending a financial product, planning a family vacation or helping choose a retirement strategy. The moment determines how much understanding is required before a recommendation feels credible. And the understanding determines how much confidence consumers place in the recommendation.

I think that’s where brands should focus. Instead of asking, “How do we optimize for AI?”, perhaps the better question is:

“At which consumer moments will AI become part of the decision, and what does AI need to understand before our brand becomes the right recommendation?”

That’s a very different conversation. It shifts the focus away from technology and back to consumer behavior, where I believe it belongs. I’d be curious to hear your perspective.

What was the last moment that really mattered to you where you turned to AI for help?

Was it planning travel? A financial decision? A major purchase? Something related to your health? A work decision?

And just as importantly, what made you trust the recommendation enough to act, or what made you stop and verify it first? I suspect those moments will tell us far more about the future of AI commerce than any adoption statistic ever could.

I’ve been thinking about what actually marks the beginning of agentic commerce.

Most of the discussion today is about technology. AI will book our vacations, reorder groceries, renew prescriptions and eventually buy products on our behalf. That’s certainly where the technology is heading, but I don’t think that’s the real story.

The real story is consumer behavior.

The question isn’t whether AI will be capable of doing these things. It’s whether consumers will be comfortable letting it.

I don’t think we’ll wake up one morning and suddenly hand over our shopping decisions to AI. We’ll get there gradually, one permission at a time.

The research is starting to reflect that.

BCG recently found that 30% of consumers say they would be comfortable letting AI make purchases on their behalf. That’s a surprisingly high number considering how early we are in this transition. But I don’t think it means consumers are ready to delegate shopping overnight. I think it tells us they’re beginning to imagine a future where AI plays a much bigger role in their decision-making.

At the same time, Morning Consult paints a more nuanced picture. AI adoption continues to rise, yet trust remains low, with nearly two-thirds of Americans saying they trust AI only a little or not at all. Perhaps even more telling, Americans trust themselves more than any institution to use AI responsibly.

Taken together, those findings suggest something important.

Consumers aren’t ready to delegate judgment.

But they are beginning to delegate something else.

Context.

The New Consumer’s latest research found that 32% of Millennials are very comfortable giving AI access to their purchase history, 25% are very comfortable sharing their email, and 26% are very comfortable sharing health information if it leads to better recommendations. Gen Z shows similar patterns. When you include people who are somewhat comfortable, those numbers become the majority across many of these categories.

I don’t think those numbers are really about data privacy.

They’re about context.

Consumers are beginning to recognize that better recommendations require better understanding. If AI knows what I’ve bought before, what brands I prefer, where I shop and what constraints matter to me, it has a much better chance of recommending something I’d actually choose.

That’s a very different form of delegation than asking AI to complete the purchase.

In other words, consumers may delegate context long before they fully delegate judgment.

We’re already starting to see companies building around this idea. Ario, for example, enables consumers to voluntarily connect their purchase histories across retailers, giving brands and AI systems access to consent-based, SKU-level behavioral data rather than relying solely on surveys or inferred preferences. The goal isn’t simply more data. It’s richer context that helps AI understand real shopping behavior before making a recommendation.

I think this has important implications for brands.

For years, our job has been to help consumers understand our brands. That doesn’t change.

But AI commerce adds two new responsibilities.

First, we need to help AI understand our brands. That means going beyond product features and claims. AI needs to understand who the brand is for, what problems it solves, the situations where it genuinely excels, and just as importantly, when it isn’t the right recommendation.

Second, we need to help AI understand the consumer. That doesn’t necessarily mean collecting more data. It means identifying the context that actually changes the recommendation and making it easy for consumers to share that information with confidence and clear permission.

I think every brand should start asking one new question:

What is the next decision consumers are willing to delegate in our category?

The answer will be very different for paper towels than it will be for infant nutrition, financial services or travel. Categories with higher consequences will require much more understanding before consumers are willing to hand over judgment.

That’s why I don’t think agentic commerce will arrive through one dramatic technological breakthrough.

It will arrive through thousands of small moments where consumers gradually become more comfortable sharing context, accepting recommendations and, eventually, delegating decisions.

The brands that understand that progression won’t simply be the ones that show up in AI recommendations.

They’ll be the ones consumers are willing to let AI choose.

One finding from Morning Consult’s latest AI Trust Report really caught my attention. Nearly two-thirds of Americans say they trust AI only a little or not at all. At the same time, AI usage continues to grow rapidly. Consumers are clearly finding value in these tools even as trust declines.

That shouldn’t be happening.

For decades, marketers have assumed that trust precedes adoption. Build trust first, and behavior follows. But AI appears to be breaking that rule. The more I thought about it, the more I realized we may be asking the wrong question. We’ve been asking whether consumers trust AI. I think consumers are asking something much simpler.

Did the AI actually understand me?

Those aren’t the same question. Think about how we judge recommendations in everyday life.

Imagine asking ChatGPT to recommend a hotel for a trip to Spain. You’re traveling with your American husband who doesn’t speak Spanish and your grandmother who doesn’t speak English. She has limited mobility and you’d like to avoid tourist-heavy areas. If the AI simply returns the ten highest-rated hotels in Barcelona, it has answered your question.

Consumers aren’t evaluating whether AI is intelligent. They’re evaluating whether AI understood the context behind the request. The more complex the situation becomes, the more important that distinction becomes. Nobody expects AI to deeply understand them before recommending paper towels. The stakes are low and the cost of a poor recommendation is minimal.

But the equation changes quickly when the recommendation involves infant nutrition, healthcare, financial products, or travel. These aren’t just product decisions. They’re contextual decisions. They involve trade-offs, constraints, emotions, and consequences that aren’t obvious from a simple prompt. When consumers don’t feel understood, something predictable happens.

• They verify.

• They read reviews.

• They ask friends.

• They visit another website.

• They seek a second opinion.

They’re not necessarily rejecting AI. They’re checking whether AI got them right before trusting that it got the recommendation right. That distinction has important implications for brands. Much of today’s conversation is focused on making brands more visible to AI. That’s important. But visibility alone won’t accelerate adoption if consumers don’t believe the AI has understood the situation they’re in.

The future of AI commerce may depend less on recommendation quality than recommendation relevance. And relevance starts with understanding.

What Should Companies Do?

If this hypothesis is right, then the goal isn’t simply to make AI recommendations more accurate. It’s to make consumers feel understood before the recommendation is delivered. That requires a different design philosophy. Instead of immediately answering the question, AI may first need to demonstrate what it heard. Imagine the difference between these two responses.

“Here are the three best hotels for your trip to Spain.”

Versus...

“You mentioned you’re traveling with your grandmother, who has limited mobility, and your husband, who only speaks English. I prioritized hotels with multilingual staff, elevators, central locations, and easy access to public transportation. Based on those needs, here are my recommendations.”

The second recommendation doesn’t just provide an answer. It demonstrates understanding.

That seemingly small difference may have an outsized impact on trust because it gives the consumer confidence that the recommendation wasn’t generic. It was tailored to the situation they described.

For brands, the implication is equally important. The more context AI has about who your brand is for, the problems it solves, and the situations where it performs best, the easier it becomes for AI to recognize when your brand is the right recommendation. That’s why I ended a recent keynote with this thought:

AI doesn’t recommend the best brand. It recommends the brand it understands.

I think there’s a companion idea emerging on the consumer side.

Consumers don’t trust AI because it’s intelligent. They trust AI when they believe it understood them.

If that’s true, then the future of AI commerce won’t be won solely by better models or better algorithms.

It will be won by better understanding.

A recent McKinsey/Adobe study found that 64% of consumers are already using AI tools to discover products and brands. At the same time, Bain and eMarketer report that roughly 18% of consumers have made purchases influenced by or through AI-driven experiences.

Those numbers caught my attention.

Not because they suggest AI shopping is coming. They suggest it’s already here.

Most of the conversation I’ve heard over the last year has focused on technology. How do brands become more visible in ChatGPT? How do they optimize product content? How do they increase the chances of being recommended by an AI assistant?

Those are reasonable questions. But I think they miss something more important.

I’ve spent most of my career studying how consumers make decisions. Whether I was working on toothpaste, shampoo, hearing aids, or consumer technology, the challenge was rarely about awareness alone. Consumers don’t buy products simply because they know they exist. They buy products because they feel confident enough to make a choice.

That’s why I think the real story behind AI shopping isn’t recommendation. It’s trust.

For decades, consumers have relied on a familiar set of decision-making shortcuts. We trust brands we’ve purchased before. We trust retailers we know. We trust friends, experts, reviews, and sometimes complete strangers on the internet. Shopping has always been a trust exercise disguised as a transaction.

Now AI is entering that equation.

What makes this particularly interesting is that trust won’t develop evenly across categories. If ChatGPT recommends paper towels, most consumers probably won’t think twice. If it recommends a breakfast cereal, many people may accept the suggestion. But what happens when the recommendation involves a product for your child? A health concern? A financial decision? The stakes change. The perceived risk changes. The need for reassurance changes.

I’ve written previously about how brands in complex, high-consideration categories are beginning to integrate directly with ChatGPT through apps and connected experiences. I think this trend reflects an important reality: in categories where consumers need more confidence, brands may not want to rely solely on being recommended by an AI assistant. They may want to participate directly in the conversation, providing additional information, guidance, and reassurance. The more consequential the decision, the more likely consumers are to seek deeper engagement before they act.

That’s why I don’t believe AI shopping adoption will happen in a straight line.

Some categories will move quickly because consumers already see them as low-risk decisions. Others will move much more slowly because consumers want validation, reassurance, or a sense of control before committing. The categories that require confidence rather than convenience may be the most difficult for AI to influence.

This is where I think many organizations are focusing on the wrong problem. Visibility matters, but visibility is only the beginning. A consumer can see an AI recommendation and still ignore it. They can receive a recommendation and immediately seek confirmation from friends, reviews, social media, or a retailer. Recommendation does not automatically lead to trust, and trust does not automatically lead to purchase.

The brands that win in AI commerce may not be the brands that are most visible to AI systems. They may be the brands that understand how confidence is built when consumers are uncertain.

As AI becomes part of the shopping journey, I think we need to spend less time asking whether AI can recommend products and more time understanding when consumers will trust those recommendations enough to act on them.

Because the biggest barrier to AI commerce may not be technology.

It may be human psychology.

Every major shift in marketing creates a new set of metrics.

When television emerged, marketers measured reach and frequency. When digital advertising took off, we became obsessed with clicks and conversions. Search brought us impressions, rankings, and share of search. Social media introduced engagement rates, followers, and sentiment.

Now, as AI becomes part of how consumers discover, evaluate, and purchase products, a new measurement race is underway.

The challenge is that there is no universally accepted scorecard for agentic commerce. As brands experiment with ChatGPT, Gemini, Perplexity, Claude, and AI-powered shopping assistants, a growing number of technology companies, agencies, and consultants are introducing their own metrics. Everyone is trying to answer the same question:

How do we know whether a brand is winning in an AI-mediated world?

The most widely discussed metric today is Share of Model. Think of it as the AI-era equivalent of Share of Voice. The idea is straightforward: when consumers ask AI systems questions about a category, how often does your brand appear in the response? If an AI assistant is asked about the best hearing aids, skincare products, or running shoes, Share of Model measures how frequently a brand is included in those answers.

There is obvious value in this. If a brand is invisible to AI systems, it has little chance of influencing consumers. Yet Share of Model has an important limitation. Being mentioned is not the same as being recommended. A brand can appear frequently without ever becoming the preferred choice. Visibility matters, but visibility alone has never guaranteed growth.

This has led many organizations to focus on AI Citation Share. Rather than measuring mentions, Citation Share measures how often AI systems reference a company’s website, product information, research, reviews, or other content assets. The underlying assumption is that citations reflect authority. If AI systems consistently rely on a brand’s information, that brand is likely viewed as a trusted source.

Citation Share is particularly attractive because it is actionable. Brands can improve content quality, strengthen digital assets, and increase authoritative references. However, citations have the same weakness as visibility metrics. Consumers do not purchase citations. A brand can become highly cited while a competitor receives the actual recommendation.

This brings us to what may be the most commercially relevant metric currently being discussed: Recommendation Share.

Recommendation Share attempts to measure how often a brand is included in the final set of options actively recommended by an AI system. This represents a meaningful shift. Rather than asking whether AI knows about a brand, Recommendation Share asks whether AI believes the brand should be considered.

For many categories, this may become one of the most important indicators of competitive strength. The challenge is that the metric remains difficult to standardize. Recommendations vary by prompt, consumer context, geography, and model updates. Two consumers asking seemingly similar questions may receive entirely different answers.

While most of the industry’s attention is focused on visibility and recommendations, another metric deserves more discussion: Prompt Penetration.

Prompt Penetration measures the percentage of shopping journeys that begin inside an AI environment. In other words, how many consumers are actually using AI to help make purchase decisions?

This may sound less exciting than Share of Model or Recommendation Share, but it answers a critical question. A category where consumers rarely use AI for shopping may experience little disruption in the near term. A category where AI becomes a routine part of the purchase journey could change dramatically over the next few years. Prompt Penetration is not a brand metric. It is a market readiness metric.

Finally, there is Agent Selection Rate, a metric that looks beyond recommendations and into the future of delegated purchasing.

Today, most consumers still want options. They ask AI for advice and then make the final decision themselves. But agentic commerce points toward a different future. Consumers increasingly may ask AI to choose on their behalf. When that happens, the question changes from “Was my brand recommended?” to “Was my brand selected?”

Agent Selection Rate attempts to measure how often an AI agent ultimately chooses a product when consumers delegate the decision. While the infrastructure is still developing, this may become one of the most important metrics in commerce over the next decade.

What is striking about all of these metrics is that each captures only one piece of the puzzle.

Share of Model measures visibility. Citation Share measures authority. Recommendation Share measures consideration. Prompt Penetration measures adoption. Agent Selection Rate measures delegated choice.

None of them, by themselves, explain how brands grow in an AI-mediated marketplace.

That should not be surprising. Every previous era of marketing required multiple metrics to understand consumer behavior. Agentic commerce will be no different.

The current debate around AI metrics reminds me of the early days of digital marketing when everyone focused on clicks because clicks were measurable. It took years before marketers realized that measuring what was easy was not always the same as measuring what mattered.

We may be in a similar moment today.

The race to measure agentic commerce has begun. The winners will not necessarily be the companies with the most metrics. They will be the organizations that understand what each metric is actually telling them, what it is missing, and how to connect those signals to real consumer behavior and business outcomes.

A few days ago, I found myself spending far more time than I expected looking through the growing list of companies that have integrated into ChatGPT. At first, it was simply curiosity. I wanted to see who was showing up. What I expected was a random collection of brands experimenting with AI.

What I found instead was a pattern. Travel companies seemed to be everywhere. Expedia, Booking.com, Priceline, Skyscanner, TripAdvisor, Viator, GetYourGuide, Hyatt, Hilton, Wyndham, and Virgin Atlantic were all represented. Real estate platforms such as Zillow, Redfin, Apartment List, and Zumper were there as well. Then came DoorDash, Uber Eats, Instacart, Uber, Turo, Klarna, Etsy, Target, and a growing list of shopping and transportation companies.

The concentration was impossible to ignore. The more I looked, the less interested I became in the individual companies. What interested me was what they had in common.

At first glance, the answer isn’t obvious. Travel, real estate, transportation, shopping, and food delivery don’t appear to belong in the same conversation. They serve different needs, operate different business models, and compete in completely different markets. But then it hit me. All of these businesses sit at the point where consumers are trying to make a decision.

Think about planning a vacation. The challenge isn’t booking a hotel. The challenge is deciding where to go, when to travel, which airline to fly, which hotel to stay in, what activities to book, and how much to spend. The booking itself takes minutes. The decision can take days or weeks.

The same is true when searching for a home. Consumers aren’t simply looking for a property. They’re evaluating neighborhoods, schools, commute times, affordability, future value, and dozens of other factors before they ever contact an agent.

Even food delivery follows the same pattern. Most of us have opened DoorDash or Uber Eats and spent more time deciding what to order than actually placing the order.

The common denominator isn’t the transaction. It’s the decision.

Historically, consumers handled those decisions themselves. They searched, compared, read reviews, visited multiple websites, and bounced between tabs trying to narrow down their options. Entire industries were built around helping consumers navigate that complexity. What makes the ChatGPT integrations interesting is that they seem to be concentrated in exactly those categories where decision-making is difficult.

• Travel is complex.

• Real estate is complex.

• Transportation is complex.

• Shopping is complex.

• Food delivery is complex.

All involve large numbers of choices, fragmented information, competing alternatives, and significant cognitive effort. And that’s where AI appears to create immediate value. Instead of opening ten tabs, consumers can ask a question. Instead of manually comparing options, they can ask AI to summarize tradeoffs. Instead of sorting through endless possibilities, they can ask for recommendations tailored to their specific situation. Viewed through that lens, the companies appearing in ChatGPT may not be early adopters because they see AI as another marketing channel. They may be there because they recognize something more fundamental.

If AI becomes part of how consumers make decisions, then the point of influence begins moving upstream. The battle is no longer limited to the moment of transaction. It starts much earlier, at the moment consumers begin exploring options and narrowing choices. That may explain why travel companies, real estate platforms, marketplaces, delivery services, and transportation providers seem to be moving first. They’re not connected by industry. They’re connected by decision complexity.

And if that’s true, the growing ChatGPT ecosystem may be giving us an early glimpse of where AI is likely to reshape customer journeys first. Not where products are manufactured. Not where transactions are processed. But where decisions are made. The most interesting thing about the ChatGPT app directory may not be the brands that are there.

It may be the signal those brands are sending about where the next battleground in commerce is emerging.

Why the next battle is not visibility. It’s recommendation.

Read more

A few months ago, most conversations about AI in marketing focused on productivity. How do we create content faster? How do we automate workflows? How do we reduce costs?

Those are important questions. But I think they are distracting many leaders from a much bigger shift that is already underway. The way consumers discover, evaluate, and choose brands is changing. Not because consumers are changing. Because the systems helping them make decisions are changing.

Over the past several months, we’ve seen a steady stream of announcements from companies like Amazon, Walmart, Mastercard, OpenAI, Google, Shopify, Robinhood, and others. At first glance, these announcements seem unrelated. A shopping assistant here. A payment capability there.An AI recommendation engine somewhere else. Viewed individually, they look like product enhancements. Viewed collectively, they tell a very different story. AI is moving closer to the point of decision.

For decades, marketers have competed for visibility. We fought for shelf space. Then we fought for search rankings. Then we fought for social reach. Then we fought for retail media placement. The objective was always the same. Increase the likelihood that a consumer would encounter our brand. The emerging challenge is different. The question is no longer simply: Can consumers find my brand?”

The question is becoming: “Will AI recommend my brand?”

That may sound like a small distinction. I don’t think it is.Imagine a consumer asks:

“What is the best hearing aid for someone who spends a lot of time in restaurants?”

Or:

“What is the best running shoe for marathon training?”

Or:

“What is the safest SUV for a growing family?”

Increasingly, consumers are not receiving a list of links. They are receiving a recommendation. Or a short list of recommendations. That recommendation is shaped by what AI can find, understand, compare, and trust. Which leads to a question that every executive should be asking:

What does AI currently think about my brand?

Most organizations have never investigated. Recently, Ballantine’s explored how AI systems represented their brand. The issue wasn’t visibility. AI knew the brand existed. The issue was understanding. In some cases, AI relied on incomplete, inconsistent, or outdated information when describing the brand. That observation should concern every marketer. Because recommendation begins with understanding.

Before AI can recommend your brand, it must first understand what your brand stands for, who it serves, and why it should be considered. Many organizations are assuming that visibility equals understanding. Those are not the same thing. AI may know your brand exists and still misunderstand your positioning. AI may find your products and still struggle to explain your value proposition. AI may mention your brand and still recommend a competitor.

That is why I believe every organization needs an AI Shelf Readiness strategy. Historically, shelf readiness meant ensuring your products were available, visible, and properly represented where consumers shopped. AI is creating a new version of that challenge.

• Can AI find your brand?

• Can AI explain your brand?

• Can AI accurately represent your claims, proof points, and differentiators?

• Can AI distinguish you from competitors?

• Can AI confidently recommend you?

Those questions are becoming increasingly important because AI is starting to sit between brands and consumers. The implications go far beyond marketing. This affects brand strategy. It affects e-commerce. It affects retail media. It affects product information. It affects consumer insights. And eventually, it will affect how companies compete.

The organizations that gain an advantage over the next several years will not necessarily be the ones spending the most on AI. They will be the ones ensuring that AI understands them better than it understands their competitors. That requires a different conversation.Not:

“How many impressions did we generate?”

Not:

“How many clicks did we receive?”

But:

“What does AI say about our brand?”

“When does AI recommend us?”

“When does AI recommend someone else?”

“And why?”

The shift may sound subtle. It isn’t. For years, marketers optimized for discovery. Increasingly, they will need to optimize for recommendation.And the brands that start building that capability now will have a significant advantage as AI becomes a larger part of the customer journey.

For Paid Subscribers

In this week’s subscriber edition, I introduce a practical Recommendation Readiness framework, share early lessons from ongoing pilots, provide a self-assessment audit you can run on your own brand, and outline five questions every leadership team should be discussing right now.

Because the future challenge isn’t simply whether consumers can find your brand.

It’s whether AI systems choose your brand when consumers ask for help.

What I said in San Juan and what it means for every marketer right now. I want to start with the belief I’ve carried for thirty years. The job of marketing is to create preference in the mind of a human being. It shaped every brief I wrote. Every budget I allocated. Every strategy I stood behind. I still believe it.

I just no longer believe it’s enough.

Because the buying decision is no longer made entirely by a human being. It is made increasingly, structurally, at scale by a decision system that combines human judgment with algorithmic filtering. And most brands are only designed for one half of that system.

I said this out loud on stage in San Juan on May 14th to more than 1,500 marketers, agency leaders, and brand practitioners. I want to say it here, with more room to explain why it matters and what to do about it.

THE SHELF HAS MOVED

Let me put the scale of this plainly. $20.9 billion in AI retail spending this year alone. An 805% surge in AI-referred traffic on Black Friday 2025. AI-mediated purchases converting at four times the rate of traditional channels. ChatGPT is now the second-largest source of e-commerce referral traffic in the United States behind only Google.

This is not a forecast. This is last quarter. And here is the structural fact that changes everything for brand strategy: Only 12% of URLs cited by ChatGPT, Perplexity, and Copilot rank in Google’s top 10.

88% of the AI shelf is invisible to your current SEO strategy.

These are not the same shelf. And if you are optimizing for one while the other grows, you are building equity in a channel that is becoming less decisive.

THE TRUST CLIFF

58% of consumers use AI to research purchases. 14% trust AI to complete the purchase. That gap is the most important number in marketing right now. I call it the trust cliff. It is not a sign that AI commerce is failing. It is a precise description of where the human holds on and where the opportunity lives for brands that understand it.

Anthropics 81,000-person study across 159 countries, the largest qualitative study of AI users ever conducted, found that hope and alarm about AI do not divide people into separate camps. They coexist in the same person. The same consumer who loves AI for research stops cold at the transaction. Not because the technology isn’t capable. Because the trust architecture isn’t there yet.

This is the design problem that no amount of creative solves. Consumers welcome AI that feels assistive. They reject AI that feels like substitution for their own judgment. The moment an agent tries to make the choice for them rather than with them, they opt out. That is not a barrier to agentic commerce. That is its design requirement.

TWO DECISION SYSTEMS. ONE BRAND. ALMOST NONE READY.

Marketing is no longer one job. It is two. Humans respond to emotion, story, identity, meaning, felt trust. AI systems evaluate consistency, evidence, reliability, data completeness, verifiable history.

Most brands are built entirely for the first list. That worked when the only decision-maker was human. But now the machine filters first. If your brand doesn’t perform in that evaluation, you never get to the human.

And the brands that are finding this out the hard way are not small or under-resourced. A senior executive at Pernod Ricard discovered that Ballantine’s, their Scotch whisky brand, was being described by AI as a prestige product. Ballantine’s is not positioned as prestige. It is accessible. Mid-market. Reliable.

AI was building a reputation the brand wasn’t managing, measuring, or even monitoring.

This is happening at scale, across categories, right now. The brands that act are the ones that will own the defaults. The brands that wait will inherit whatever the algorithm decided in their absence.

WHAT THE BIGGEST BRANDS JUST DID

Six weeks before I gave this talk, L’Oréal, Unilever, Mars, Beiersdorf, and Reckitt moved together.

They published machine-native catalogs through a protocol called AMP — built by a company called Azoma. Published once. Surfaces everywhere: ChatGPT, Perplexity, Rufus, Google AI.

The CEO of Azoma explained the move plainly: “They’re not about to hand control of how their products are represented to an AI black box.”

That is infrastructure. Not a pilot. Not an experiment.

And it is six weeks old.

Ruroc — a ski helmet brand — saw a 14x increase in ChatGPT traffic and became the number one AI-recommended ski helmet brand. Pandora achieved 60% case deflection improvement and a 10% NPS increase through AI agents.

Williams Sonoma deployed an AI shopping agent. Brands with agents: 6.2% growth. Brands without: 3.9%. That is 59% higher for the brands that moved first.

The infrastructure is being built. The defaults are being set. The question is not whether this happens. The question is whether your brand is in the room when it does.

WHY PUERTO RICO IS WHERE THIS GETS LEARNED FASTEST

Anthropics research measured AI sentiment across 12 regions. Latin America and the Caribbean scored as the most AI-optimistic region on Earth. Not cautious. Not resistant. Leaning in.

And Puerto Rico specifically sits at the intersection of three structural advantages:

Diaspora commerce. $1.8 billion in cross-border buying power, already active inside AI commerce systems. These are real consumers, already transacting, already inside the infrastructure.

Spanish-first optimization. Seventy percent of AI training data is English. Neutral Spanish fails in Puerto Rico — it doesn’t capture the cadence, the Spanglish, the Taíno influences that define how this market actually communicates. Build for Puerto Rican Spanish specifically, and you unlock 725 million Spanish speakers globally with precision no mainland brand can replicate. That is a moat.

Closed-loop feedback. 3.2 million people. 93% internet penetration. Signal is clear. Noise is low. What works here, works fast — and it scales.

Closed ecosystems are not limitations. They are where new operating models are built and pressure-tested before they go global.

THREE DECISIONS. ONE WINDOW.

McKinsey estimates 12 to 18 months to build AI commerce readiness. The window opened in early 2026.

Decision 1: Find your position. Open ChatGPT or Perplexity today. Search your category. Search your brand name. What comes back is your AI shelf. Most brand leaders have never looked at it.

Decision 2: Close the gap. Is your product catalog machine-readable? Are your claims verifiable? Are you protocol-compliant with the systems that are already surfacing your competitors? If you don’t know the answers, that is the gap.

Decision 3: Move now. The brands that move now will set the defaults that latecomers inherit. The window will not stay open indefinitely.

I ended the talk in San Juan with one sentence. I’ve been saying it in briefings for a year.

AI doesn’t recommend the best brand. It recommends the one it understands.

Make sure yours is understood.

If you want to go deeper — the full diagnostic, the trust architecture framework, or what AI commerce readiness looks like in practice — reply here or reach out directly.

One of the more interesting things about concept development is how quickly organizations become attached to the first viable direction. An idea enters the room and almost immediately the process begins moving toward definition. Teams start asking the practical questions. Who is this for. What is the benefit. What should the claim be. How do we position it. What does the concept sound like.

All of that is understandable. Organizations are designed to create momentum. Once an idea feels promising, there is pressure to turn it into something tangible as quickly as possible. But over time, I’ve come to think that one of the most important distinctions in concept development is this:

The idea is not the concept. The concept is an interpretation of the idea. And the way that interpretation takes shape changes everything that follows. This becomes easier to see when you look at how differently the same product can be framed depending on which tension the organization decides to resolve. A hydration product can become a performance story. Or a social identity story. Or a sustainability story. Or a sensory novelty story. Or a wellness story. Same product, different concept. And more importantly, different future. That last part is often underestimated. Because when organizations choose a concept direction, they are not simply choosing language. They are choosing:

• which audience matters most

• which tension deserves focus

• which meaning the brand will begin to own

• what consumers will start associating with the product over time

That is a much bigger decision than most concept discussions acknowledge. What makes this difficult is that the first viable direction often feels correct simply because it arrived first. The organization starts building around it. Slides get created. Research gets commissioned. Creative discussions begin. Internal alignment forms around the framing. The direction gains momentum, and momentum itself starts becoming evidence. But early momentum is not always the same thing as strategic strength. Sometimes the chosen direction is simply the most familiar one, the safest one, the one closest to the category convention. The one easiest to explain internally. This happens often in CPG innovation. A new ingredient enters the pipeline and the organization immediately frames the product around efficacy because that is where the technology naturally points. A sustainability improvement gets translated into responsible consumption because that is where the category conversation already exists. A sensory innovation becomes a flavor story because that feels easiest to communicate. None of these directions are necessarily wrong. But they are still choices. And the moment one direction starts hardening inside the organization, alternative interpretations begin disappearing from consideration.

That is where things get interesting. Because sometimes the strongest concept is not sitting in the most obvious framing of the product. Sometimes it sits in a completely different tension. A hydration product might become less about health and more about social belonging. A beauty product might become less about efficacy and more about emotional reassurance. A household product might become less about cleaning performance and more about removing friction from daily life. The product itself has not changed. The meaning of the product has. And consumers respond to meaning far more than organizations sometimes realize. This is one of the reasons concept development becomes more strategic than it appears from the outside. The challenge is not simply generating a concept.

The challenge is understanding the range of concepts that could emerge from the same idea before the organization commits to one. That requires a different kind of thinking. Not just executional thinking. Interpretive thinking. The teams that consistently build stronger concepts tend to spend more time exploring those alternative interpretations before locking into a direction.

They ask different questions. What changes if we focus on a different audience. What happens if we frame the benefit differently. What tension creates the strongest pressure. Which version feels most culturally relevant. Which direction creates the most distinctiveness over time.

Those conversations often surface possibilities that were invisible in the first round of development. And importantly, they create a stronger understanding of why one direction deserves to move forward over another. That clarity matters because once a concept direction is chosen, the organization begins constructing an entire system around it.

• Packaging evolves around the promise.

• Claims reinforce the chosen meaning.

• Creative starts amplifying the selected tension.

• Consumer expectations begin forming around the framing.

At that point, changing direction becomes increasingly difficult. Not because the organization lacks flexibility, but because systems naturally resist reorientation once momentum has formed. Which is why early concept choices matter so much. Not because they determine everything permanently. But because they quietly shape the trajectory of everything that follows.

Over time, I’ve come to think that one of the hardest disciplines in concept development is resisting the urge to confuse the first compelling direction with the strongest possible one. Those are rarely the same thing. And sometimes the most important work happens before the organization decides which version of the idea it wants to bring into the world.

In this episode, Ana shares her perspective on how AI is transforming creativity, ownership, and entrepreneurship. After downloading ChatGPT the day it launched, Ana used AI to 10x her income, micro-retire, and launch multiple businesses — including AnaGPT and LawRobo. We explore what founders and creators need to understand about intellectual property in the AI era, who really owns AI-generated content, and how to build brands that remain valuable as technology reshapes the future of creativity.