When AI Does the Shopping: How to Stay on the Customer's Shortlist

IN ONE MINUTE

Customers still search, but search now answers, and increasingly buys. Google's AI Overviews and AI Mode, and assistants such as ChatGPT, Gemini and Meta's Muse, reply to "what is the best SUV for my family?" with a shortlist, and the shoppers they send convert far better than other visitors. AI cannot recommend what it cannot read, and it trusts customers more than brands. To stay on the shortlist, pass three tests: Findable, Trusted and Chosen. This issue includes a 90-day plan.

THE OBJECTIVE

For companies selling products a customer can also buy elsewhere, such as cars, cosmetics and consumer goods, the first conversation now happens inside an AI answer. This issue sets out what decides whether you appear in it, what changes when AI agents start buying, and a practical plan for this quarter.

WHAT THE RESEARCH SHOWS

Search has become an answer. Search has reinvented itself before: from Archie, the first internet search tool, in 1990, to Yahoo's directory in 1994 and Google in 1998. The latest shift happens inside the search engine itself. Google reported in July 2026 that AI Mode has passed one billion monthly users. Dedicated assistants are growing alongside it: McKinsey's June 2026 survey of about 21,000 consumers in 11 markets found that 53% had used AI for shopping in the past three months, and nearly a third of AI shoppers named AI chat or search as the biggest influence on their most recent purchase. Big-ticket purchases are no exception. Cox Automotive's study of more than 2,300 US vehicle buyers found a quarter of new-car buyers used AI tools, including AI-generated search overviews.

The shoppers AI sends are ready to buy. Adobe's analysis of more than one trillion visits to US retail sites found that in September 2026, visits referred by AI converted 90% better than other traffic and generated 83% more revenue per visit. A little over a year earlier, they converted worse.

The data feed is the new shelf. Google says its Shopping Graph holds more than 60 billion product listings, built largely from the product data merchants submit. Yet in Adobe's data, the average US retail homepage scored 63% for how much of its content AI can understand and cite. What AI reads best is content that explains: specifications, buying guides and policies.

AI trusts what others say about you. McKinsey's research on AI search found a brand's own websites make up only 5 to 10% of the sources AI draws on, and only 16% of brands systematically track how they perform there. A study of more than 800,000 AI answers, commissioned by Trustpilot and run by Seer Interactive, found brands that actively collect and respond to reviews were cited in 75% of relevant answers, against 1% for brands with no review profile.

THE CASE: FROM THE SHELF TO THE DATA FEED

I have worked in both beauty and automotive, two industries where a customer can buy something comparable from someone else. In both, the contest used to be fought over shelf space: the counter position in a department store, the dealership on the right road, the first page of search results. In Issue 016 I described turning display fixtures into a data service; now the shelf itself is becoming data.

In conversations with senior executives across consumer and automotive businesses, I hear the same priority: companies are trying different ways to establish and strengthen their data strategy, because AI cannot generate a relevant recommendation without data.

Ask Google or an assistant for "best SUV" and the answer is assembled from what AI can read and verify: specifications, prices, availability, independent reviews and what real customers say. Then the question narrows. "Best SUV" becomes "best seven-seat SUV for long motorway journeys with a dog". "Best mascara" becomes "best mascara for sensitive eyes that lasts through a humid day". "Premium" or "long-lasting" gives the AI nothing to work with. A boot capacity with the third row up, or a wear test in hours, does. This is where the value proposition now has to work: not as a slogan, but as a specific, evidenced answer to the customer's question, so that AI can rank the brand as the relevant choice.

THE FRAMEWORK: FINDABLE, TRUSTED, CHOSEN

Test	The question AI is answering	What good looks like	How to measure it
Findable	Can I read and understand these products?	Every product attribute complete and identical across website, Google Merchant Center, marketplaces and retail partners; prices and policies in plain text	Share of key pages AI can read; product data errors found and fixed
Trusted	Do independent sources and real customers confirm the claims?	Reviews requested after every purchase, every substantive review answered, independent tests and expert coverage	Review volume, recency and response rate; sentiment in AI answers
Chosen	Is this the best answer to this customer's need?	One clear, evidenced answer for each priority question and use case, including honest trade-offs	How often you are recommended for priority questions; conversion of AI-referred visits

WHEN THE AGENT BUYS

AI is starting to complete the purchase, not just recommend it. In the US, the share of AI users who completed a purchase with AI rose from 10% to 17% in a year, McKinsey found. OpenAI introduced Instant Checkout in ChatGPT in September 2025, and Google began rolling out agentic checkout in Search that November. On 8 September 2026, Meta launched Muse, a personal agent that compares prices and orders on a user's behalf, asking for approval before it completes a purchase. Shopify announced agentic checkout for its merchants through Muse; Amazon blocked Muse from shopping on its site, saying it had not been told about the agent's access.

For leadership teams, this changes four things.

• The customer may never see your website. The agent reads your data, policies and reviews, then decides. The three tests still apply, with less room for error: an agent acts on a wrong price or delivery date.
• You need an agent policy. Which agents may browse, compare and buy from you, on what terms and through which checkout? Allow, partner or block is now a commercial decision, not a technical setting.
• Accountability needs defining. When an agent buys the wrong size or the wrong specification, who handles the return and who carries the cost? Settle it in terms and service processes before volumes grow.
• Share product data; protect customer data. McKinsey advises sharing product, price and stock data widely, while tightly controlling customer, behavioural and loyalty data, including what third-party agents may keep.

THE PLAYBOOK: THE FIRST 90 DAYS

Before day one: name one owner. In most companies, nobody owns what AI says about the brand. AI reads your data, website and reviews at once, but different teams own each. As with AI agents in Issue 019, give one senior commercial leader the outcome and the authority across those teams. McKinsey's latest research makes the same recommendation: cross-functional governance with one accountable leader.

Weeks 1–2: run the test. Collect the 20 questions customers ask before they buy, from sales teams, service logs and your own site search. Put each one to Google (AI Overviews and AI Mode), ChatGPT, Gemini and Copilot. Record three things: are you named, is what AI says about you correct, and which sources does it cite? That is your baseline.

Weeks 3–6: fix the data. Start with your 50 best-selling products. Check every attribute against one master record and correct the gaps wherever they appear: website, product feeds, marketplaces, dealers and retailers. Add structured product markup to product pages, and move key facts out of images and PDFs into plain text. It pays: Shopify reports, as cited by McKinsey, that AI searches on structured catalogue data convert at roughly twice the rate of searches on scraped or out-of-date data.

Weeks 3–8: build the customer voice. Ask every customer for a review after delivery or handover, and answer every substantive review within days, positive or negative. Fix the causes of recurring complaints.

Weeks 6–12: publish the answers. For each priority question, publish one page that answers it directly: who the product is for, the specifications that matter for that use, the evidence, and who it is not for.

Weeks 8–12: set your agent policy. Agree across commercial, legal and technology teams which agents may buy from you, through which checkout, on what return terms, and what customer data they may see and keep.

Day 90: run the test again. Repeat the 20 questions, compare with the baseline, and report the change to the leadership team with conversion and revenue per visit from AI referrals. Then make it a quarterly review.

THE PART THAT NEEDS CARE

Keep a sense of proportion. Globally, only 14% of AI users have completed a purchase with AI so far, McKinsey found. AI adds a front door; it does not close the others.

Never take a shortcut on trust. Since April 2025, fake reviews and concealed incentivised reviews have been a banned practice under UK consumer law, with fines of up to 10% of global turnover.

WHAT THIS MEANS FOR CEOS AND BOARDS

Ask to see the test. Ask for the 20-question results across Google and the main assistants, and see whether AI's view of the company matches its market position.

Decide the agent policy. Allowing, partnering with or blocking shopping agents is a commercial choice; make sure it is made deliberately, with a named owner.

WHAT THIS MEANS FOR CMOS AND COMMERCIAL LEADERS

Turn the slogan into answers. Give each priority question one evidenced answer by use case, in the customer's words.

Put your own assistant where customers are ready to buy. McKinsey found 56% of consumers used a platform AI tool, against 27% who used a retailer's own assistant; at checkout, the retailer's assistant edged ahead, 15% to 14%. Focus an owned assistant on comparing, finding the right fit and completing the purchase.

WHAT THIS MEANS FOR CFOS AND FINANCE TEAMS

Report AI referrals separately. Track their conversion and revenue per visit; small volumes can carry outsized value.

Fund the fixes as revenue investments. Product data quality and explanatory content compete with paid search for budget; judge both on revenue per visit.

WHAT THIS MEANS FOR TECHNOLOGY AND DATA LEADERS

One master record for every product. Feeds, site, marketplaces and partners should all draw on it, so a price change lands everywhere at once.

Make your data safe to act on. When an agent buys, a wrong price or delivery date becomes an order, not just a wrong answer. Keep prices, stock and policies machine-readable and current, within the guardrails in Issue 019.

THE LAST WORD

AI can only recommend what it can read, and it believes your customers more than your advertising.

If a customer asked Google or an AI assistant for the best product in your category today, would it name you, and could it explain why?

—

Until next week,

Ian Cheng

→ Found this useful? Forward it to a CEO, CMO or head of e-commerce.

→ Subscribe to The AI Executive Briefing, weekly in your LinkedIn inbox.

Sources: OpenAI, "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol", September 2025. Meta Muse launch, 8 September 2026, as reported by Engadget; Shopify and Amazon positions as reported by Adweek and onlinemarketing.de, September 2026. Alphabet, "Q2 2026 earnings call: Remarks from our CEO", 22 July 2026. Google, Shopping Graph figures stated at Google Marketing Live and on Google's shopping blog, May 2026, as reported by PPC Land. Elizabeth Silliman, Jack Trotter and Kelsey Robinson, with Desirae Oppong and Shreya Jaggi, "Who wins when AI does the shopping? New evidence on agentic commerce", McKinsey, 28 October 2026 (about 21,000 consumers in 11 markets, June 2026; Shopify figure as cited there). Cox Automotive, "2025 Car Buyer Journey Study", press release, 13 January 2026 (more than 2,300 recent US vehicle buyers). Adobe, "AI traffic now generates 83% higher revenue per visit for U.S. retailers", 29 October 2026 (more than 1 trillion visits to US retail sites). McKinsey, "New front door to the internet: Winning in the age of AI search", October 2025. Trustpilot, "What AI says about you", research by Seer Interactive, press release 12 May 2026 (more than 800,000 AI responses). Competition and Markets Authority, "Fake reviews" guidance (CMA208), 4 April 2025.