AI is not buying for consumers yet. It is shaping what they consider

The largest near-term commercial risk from AI may arrive before agentic checkout. It is the possibility that a brand never enters the customer’s consideration set.

A new global study from Boston Consulting Group sharpens that concern. BCG reports that AI introduced consumers to brands they otherwise would not have considered in roughly 63% of AI-assisted purchase journeys. The denominator matters: this is not 63% of consumers or all purchases. It is BCG’s reported share of journeys in which AI was already involved. Even with that boundary, the finding points to a consequential role for AI before the transaction. It can help construct the shortlist itself. [1]

That makes the most urgent question for marketers less futuristic than whether an autonomous agent will complete a purchase. It is whether an AI system understands when the brand belongs in the conversation, how it should be positioned and what evidence makes it recommendable.

<h2>The strongest new evidence is about shortlist construction</h2>

BCG’s August 26 report draws on a survey of more than 13,000 consumers across 12 markets. It separately includes more than 100 qualitative interviews and an analysis of more than 1,000 brands using a proprietary research database covering 2022 to 2026. Those components should not be treated as one sample. BCG also does not disclose the consumer survey’s fieldwork dates, country-level sample sizes, recruitment mode or exact AI-question wording. [1]

Within that survey, 31% of respondents said they used AI at least occasionally in purchase journeys. Nineteen per cent were classified by BCG as “AI loyalists” who relied on it regularly, while 13% said they already purchased whatever AI recommended. Among the AI-loyalist subgroup, 70% ultimately bought products AI recommended. [1]

These are self-reported measures, not transaction logs. They do not prove that AI caused any particular purchase. The more defensible strategic reading is that AI has become involved in discovery, comparison and recommendation for a material share of surveyed consumers. BCG’s own interpretation is that AI is increasingly assembling the consideration set before consumers reach brand-owned channels.

The 63% finding is especially important because it suggests AI can do more than rank familiar options. In AI-assisted journeys, it can introduce alternatives that the consumer had not already named. That creates opportunity for challengers and risk for incumbents, but only within the scope of BCG’s under-disclosed journey measure. It is a strong prompt for further measurement, not a universal market law.

<h2>Assistance and delegation are different behaviours</h2>

Gartner’s May release provides the clearest direct contrast between using AI to narrow choices and allowing it to make the decision. In a January 2026 survey of 322 US consumers, 31% were willing to let AI narrow choices for household-supplies purchases and 28% said the same for personal electronics. Willingness to let AI make purchase decisions topped out at 11% across lower-stakes categories, with personal care and household supplies cited as examples. [2]

The 11% figure is not an autonomous-checkout adoption rate or an average across every category. It is a stated-willingness maximum from a small US survey whose public release does not provide the full questionnaire, sampling method or weighting. Still, the internal comparison is useful. Respondents were more open to AI reducing the option set than to AI deciding for them.

This is the distinction that much of the agentic-commerce debate misses. Helping a shopper discover, research and compare products is not the same behaviour as choosing, ordering and paying on the shopper’s behalf. The first can already influence competition even when the second remains limited.

<h2>Canadian consumers report influence, but not unquestioning trust</h2>

A National Bank-commissioned Léger survey makes the middle of the journey more visible. Léger surveyed 1,518 English- or French-speaking Canadian adults online from June 5 to 7, 2026. Thirty-nine per cent said they had used a generative-AI tool in the previous 12 months to support a purchase decision. [3]

The follow-up results apply only to that user subgroup, for which the public release does not disclose a sample size. Eighty per cent said the tools helped them compare options, 47% said AI had a moderate-to-significant impact on their purchasing decisions and 28% said they had regretted a purchase recommended by the tools. [3]

Those findings describe self-reported experience and perceived influence. They do not show that AI caused the purchase or the regret. Their value is the tension they reveal: consumers can find AI useful enough to compare options and influence a choice while still encountering outcomes they later question. Influence is advancing faster than unquestioning trust.

<h2>Discovery is moving upstream, but the studies do not form a trend line</h2>

NielsenIQ and Kearney add directional evidence from research dated January 2026 and published in March. Their material reports that 74% used AI for some form of discovery in a study with 750 respondents. An accompanying chart labels 54% as “AI researchers” and 20% as “AI shoppers”. However, the primary material does not disclose the geography, respondent definition, sampling design or whether the two labels are mutually exclusive. The finding is useful only with that limitation visible. [4][5]

Capgemini’s <em>What matters to today’s consumer 2026</em> supplies an earlier baseline from an online survey of 12,000 adults across 12 countries. Twenty-five per cent said they had used generative-AI shopping tools in 2025, while 31% planned to use them in the future. Seventy-six per cent wanted to set boundaries or rules before a digital assistant acted on their behalf, and 71% expressed concern about a lack of clarity around consent and transparency in generative AI’s collection and use of personal data. [6][7]

Capgemini’s full report dates its fieldwork to October 2025, while the January 2026 release says October and November 2025. Its 31% intention measure has no stated deadline. Those details rule out treating the figure as a 2026 forecast.

The Capgemini, NIQ/Kearney, Gartner, National Bank/Léger and BCG studies cannot be arranged as a statistical adoption curve. They use different countries, samples, definitions, questions, dates and denominators. The responsible conclusion is qualitative: several studies point towards AI playing a meaningful role before the final choice, while the clearest direct assistance-versus-delegation comparison shows greater willingness to let AI narrow choices than to let it make the purchase decision.

<h2>AI does not need to transact to change the market</h2>

A transaction is only the final visible event in a longer decision process. AI can alter commercial outcomes earlier by interpreting a need, defining which attributes matter, retrieving evidence, selecting candidates and framing the comparison.

Consider a customer asking for a running shoe suitable for knee pain, a CRM for a 50-person professional-services firm, skincare for sensitive skin or a quiet luxury hotel in Paris. The AI is not merely retrieving known brands. It is translating a need into criteria, deciding which alternatives qualify and explaining why each one might fit.

A brand can therefore lose before a click, website visit or agentic purchase occurs. Strong brand awareness and a high-converting site do not guarantee inclusion if the system does not associate the brand with the stated need, cannot verify its claims or repeatedly finds competitors better supported.

This extends the earlier AEO Updates distinction between <a href='/articles/ai-consideration-set-entry-versus-winning'>entering the AI consideration set and winning within it</a>. Eligibility asks whether the brand appears at all. Preference asks why it is placed ahead of alternatives. Consumer research now adds a third reason the distinction matters: AI may be constructing a shortlist that contains brands the shopper had not previously considered.

<h2>Consideration is not one score</h2>

The findings also argue against treating AI visibility as a single mention count. A brand can appear without being recommended. It can surface for one need state but disappear when the prompt adds a budget, audience, risk tolerance or use-case constraint. It can be described as reliable but not innovative, premium but not accessible, suitable for experts but not beginners.

A useful planning measure is <strong>share of AI consideration</strong>: the proportion of a defined set of relevant prompt runs in which the brand enters the candidate set. It is not market share, sales share or a population estimate. Its meaning depends on the prompt universe, model mix, geography, time window and inclusion rule.

That is why the prompt set must represent the market rather than one convenient question. As the AEO Updates analysis <a href='/articles/prompt-not-market-ai-visibility-measurement'>The prompt is not the market</a> explains, robustness across paraphrases and prompt families is different from coverage across customer needs, and both are different from changes in retrieval or system conditions. Those layers should remain separate rather than being compressed into one percentage.

<h2>A practical AEO measurement sequence</h2>

A stronger consideration measurement system separates five stages. <strong>Eligibility</strong> asks whether the brand enters relevant candidate sets. <strong>Positioning</strong> records the audiences, needs, occasions and attributes with which the system associates it. <strong>Preference</strong> measures whether the brand is recommended, how prominently and for what stated reasons. <strong>Robustness</strong> tests whether those outcomes survive prompt variations and different answer engines. <strong>Outcome</strong> connects the answer to validation, referral, transaction and, where evidence allows, incremental commercial value.

This sequence is an editorial measurement architecture, not a validated universal model. Its practical advantage is diagnostic. Low eligibility suggests a coverage, entity or evidence problem. Consistent inclusion but weak preference points towards comparative proof, relevance or reputation. High visibility with unstable positioning indicates that the prompt family or source environment needs closer inspection. Downstream conversion without a credible counterfactual still does not prove the incremental value caused by AI visibility.

The result should not be one blended score. Teams need a dashboard that keeps entry, portrayal, preference, robustness and outcomes visible as separate questions.

<h2>What marketers should do now</h2>

Marketers should begin by defining the needs and decision contexts in which the brand genuinely belongs. Product categories are too broad on their own. A useful consideration map includes audience, occasion, problem, constraint, price position and evidence threshold.

They should then audit whether clear, current and independently supportable evidence exists for the claims that would make the brand relevant in each context. That evidence can live on product pages, documentation, research and case studies, but it also needs corroboration across the third-party sources that AI systems may retrieve and use.

Measurement should test families of prompts across more than one system. It should record not only whether the brand appears, but which competitors appear beside it, which attributes the answer repeats, which sources support the recommendation and what changes when a constraint is added.

For commerce teams, structured product data, accurate availability, current pricing and review evidence still matter. They make products legible and comparable. But technical legibility is not the same as strategic preference. A system also needs credible reasons to believe that the product fits the user’s stated need.

Finally, teams should connect consideration measures to downstream behaviour without skipping the attribution problem. AI-referred visits, conversions and transactions can be observed. The incremental value caused by a change in recommendation requires stronger designs, such as controlled tests or credible comparisons, as explored in <a href='/articles/ai-search-transactions-visibility-value'>AEO Updates’ analysis of the AI visibility attribution gap</a>.

<h2>Limits of the evidence</h2>

All five source families are commercial or commissioned studies. They differ in geography, population, sampling, disclosure and the behaviours they ask respondents to recall or imagine. Several omit fieldwork details or question-level bases. BCG’s brand analysis and qualitative interviews are separate from its consumer survey. NIQ/Kearney does not disclose the geography or respondent definition for its 750-person result. Gartner measures willingness, while National Bank/Léger and Capgemini rely on self-reported behaviour, perceptions and intentions.

None of this makes the findings unimportant. It defines what they can support. They show that AI is becoming relevant to discovery, comparison and decision support. They do not establish one universal rate of AI shopping, a longitudinal trend or a causal estimate of AI’s effect on sales.

<h2>AEO Updates Takeaway</h2>

The immediate AEO risk is not that AI completes the wrong purchase. It is that the brand never enters the consideration set.

The newest evidence suggests that AI can introduce options, narrow choices and shape comparisons before a consumer reaches a brand’s channels. That means the practical job is larger than earning a mention. Brands need to become eligible for the right needs, be positioned with accurate and defensible claims, earn preference against alternatives and remain stable when the prompt or system changes.

Autonomous purchasing may grow. Marketers do not need to wait for it. AI can reallocate consideration long before it controls the checkout.

<h3>References</h3>

[1] Boston Consulting Group, <a href='https://www.bcg.com/publications/2026/five-consumer-shifts-reshaping-growth' target='_blank' rel='noopener noreferrer'>Global Consumers Have Moved On. Has Your Growth Strategy Caught Up?</a>, August 26, 2026.

[2] Gartner, <a href='https://www.gartner.com/en/newsroom/press-releases/2026-05-27-gartner-survey-finds-consumers-want-ai-shopping-help-but-not-ai-purchase-decisions' target='_blank' rel='noopener noreferrer'>Gartner Survey Finds Consumers Want AI Shopping Help, But Not AI Purchase Decisions</a>, May 27, 2026.

[3] National Bank of Canada, <a href='https://www.nbc.ca/about-us/news-media/press-release/2026/20260618-nbc-ai-factor-purchasing-decisions.html' target='_blank' rel='noopener noreferrer'>AI is becoming a factor in purchasing decisions: 39% of Canadians use AI to guide what they buy</a>, June 18, 2026.

[4] NielsenIQ, <a href='https://nielseniq.com/global/en/news-center/2026/ai-is-resetting-the-rules-of-growth-in-cpg/' target='_blank' rel='noopener noreferrer'>AI Is Resetting the Rules of Growth in CPG</a>, March 3, 2026.

[5] NielsenIQ and Kearney, <a href='https://nielseniq.com/global/en/wp-content/uploads/sites/4/2026/03/NIQ-x-Kearney_New-Growth-Frontier_PDF-for-Web-1.pdf' target='_blank' rel='noopener noreferrer'>The New Growth Frontier</a>, March 2026.

[6] Capgemini Research Institute, <a href='https://www.capgemini.com/wp-content/uploads/2026/01/Final-Web-Version-Report-Consumer-Trends-2026.pdf' target='_blank' rel='noopener noreferrer'>What matters to today’s consumer 2026</a>, 2026.

[7] Capgemini, <a href='https://www.capgemini.com/news/press-releases/consumers-are-balancing-spending-on-essentials-with-small-indulgences-that-provide-an-emotional-boost/' target='_blank' rel='noopener noreferrer'>Consumers are balancing spending on essentials with small indulgences that provide an emotional boost</a>, January 6, 2026.