AI agents that shop: what changes for brands, and what to do now

Short answer: AI assistants are moving from recommending products to buying them. Today they mostly do the research: they search, compare prices, deals and reviews, and hand the shopper a shortlist. An agent can only shortlist what it can read, so your product data (feeds, structured data, prices, stock, shipping and returns) now matters alongside your content and your reviews. Make offers machine-readable, keep data in sync, and earn mentions in the sources AI trusts. Checkout readiness comes after that.

What an AI shopping agent does

A shopper describes what they want, in their own words and with their own limits: 'running shoes for flat feet under €120, delivered by Friday'. The agent turns that into searches, reads product pages, feeds and reviews, compares options and returns a short list. The newest ones go one step further and complete the purchase, with the shopper approving the total. As Becca Coggins, a senior partner at McKinsey, puts it, autonomous agents 'do the legwork — searching, filtering, comparing and even purchasing on behalf of the customer'.

For a brand, the change is who reads your store first. A person scans a page and forgives a missing detail. An agent reads data, and if the price, size or return policy isn't there in a form it can parse, it is likely to pick a store where it is.

Where it already happens (October 2026)

Where the platforms document how they work, two things repeat. They read structured product data (feeds and catalogues), not only pages. And they draw on other sources before they recommend: reviews, editors' lists, Q&A, comparison pages. Meta hasn't said how Muse picks products yet.

How big it is, and how fast

• Shoppers already buy with AI's help. Roughly 23% of Americans made purchases using AI in the past month. Morgan Stanley estimates AI shopping agents could account for $190 to 385 billion of US ecommerce spending by 2030, 10 to 20% of the market (Morgan Stanley, December 2025).
• The forecasts are large. McKinsey estimates agents could orchestrate as much as $1 trillion of US retail revenue by 2030, and as much as $3 to 5 trillion globally (McKinsey, 2025).
• AI shoppers arrive ready to buy. In July 2026, AI-referred visits to US retail sites were up 62% on a year earlier, converted 60% better than other traffic and brought 53% more revenue per visit (Adobe).
• Many stores aren't readable yet. Adobe scored the homepages of a broader set of US retailers at only 61% for LLM visibility: parts of those pages can't be read by AI models.
• Trust is the brake. In a Vogue Business survey reported by Fortune, 31% would let an AI agent handle their shopping and 72% would not share card details. Expect agents to research far more often than they buy, for now.

How an agent decides what to recommend

1. Can it find you? Where platforms document it, agents work from catalogues. ChatGPT takes product feeds with fields such as price, sale price, availability, ratings, shipping and returns. Google uses Merchant Center. Perplexity syncs catalogues through PayPal. A product that isn't in the feed, or whose title says 'The Classic' instead of what it is, is hard to match to a shopper's request.
2. Does it trust you? Assistants lean on reviews, rating counts, editors' lists and comparison pages. Amazon says its assistant draws on customer reviews, community Q&A and information from across the web, including publishers such as Good Housekeeping and Vogue. This part is classic AI visibility: being named by the sources AI reads for your category.
3. Can it act? To compare and buy, an agent needs the price, the current sale price, stock, delivery cost and return terms, and a checkout that doesn't break without a human. A deal that only exists in a pop-up or a discount code is easy for it to miss.

None of the platforms publish how much each signal weighs in ranking; OpenAI's feed specification says 'Eligibility does not guarantee display.' Treat the list as the conditions for being considered, not a formula.

What to do now

1. Get every product into the feeds agents read. Google Merchant Center first; Shopify's catalogue if you sell on Shopify; PayPal's store sync if you take PayPal. Check that your best sellers actually show up when you ask the assistants for them.
2. Write titles and descriptions in shoppers' words. Product type, key attribute, who it's for, size: 'Fragrance-free goat milk moisturiser for dry skin, 50 ml' beats a collection name.
3. Add structured data that matches the page. Product, Offer (price, availability) and AggregateRating markup, plus shipping and return details, all agreeing with what a visitor sees.
4. Make offers machine-readable. Put the sale price in your feeds and in the page markup, list current promotions on a public page with dates, and avoid deals that only exist behind a pop-up, an email signup or a code. It matters most in sale periods such as Black Friday.
5. Keep price and stock in sync. Update feeds when a sale starts or ends and when something sells out; ChatGPT's feed, for one, doesn't schedule price or stock changes from dates. A wrong price or a false 'in stock' costs the sale.
6. Put answers in text. Sizing, ingredients, materials, compatibility and who the product is for belong on the product page as text, not only in images or PDFs.
7. Grow and expose reviews. More reviews on your hero products, rendered in the page itself (not only in a script-loaded widget) and passed in feeds where supported.
8. State shipping and returns plainly. Agents compare total cost and risk. A clear policy page and the matching feed fields help.
9. Don't block AI by accident. Check that robots.txt and bot protection let AI search crawlers read your product pages (how to check).
10. Then prepare checkout. Guest checkout, Shop Pay or PayPal if you use those platforms, and Google's Buy button if you are eligible. Agent checkout is small today, but Meta, Google, Amazon and Perplexity are all building it.

What doesn't change

Agents still lean on what others say about a brand to decide whether it is good. If editors, comparison sites and reviewers don't mention you for 'best X under €50' or 'best gift for Y', clean product data alone won't put you on the shortlist. Our guides on what makes content get used by AI and fixing what AI says about your brand cover that side. Nobody can promise a place in an agent's shortlist; the steps above improve your chances.

Check where you stand

Ask ChatGPT, Gemini and Perplexity the questions your buyers ask before they buy: a price band ('best vitamin C serum under €30'), a deal ('best running shoe deals this week'), a comparison ('[your brand] vs [competitor]') and a value question ('is [your brand] worth it?'). Ask each several times, because answers vary, and note who is named, which sources are cited and whether your prices and offers are right.

Sources

• Fortune, 'Meta's new AI can shop, choose and check out. Most people don't want it touching their money' (2026), citing a Vogue Business survey.
• PYMNTS, 'Shopify brings Shop Pay checkout to Meta's Muse AI agent' and 'Meta adds PayPal to Muse AI assistant for global checkout' (2026).
• Digital Commerce 360, 'OpenAI shifts checkout plans in its agentic commerce strategy' (2026).
• OpenAI, 'Product feed specification'.
• Google Merchant Center Help, 'About the Universal Commerce Protocol (UCP) and UCP-powered checkout'.
• Amazon, 'Alexa for Shopping (formerly Rufus)'.
• PayPal, 'PayPal and Perplexity launch Instant Buy' (2025).
• Morgan Stanley, 'Here come the shopping bots' (8 December 2025).
• McKinsey, 'The agentic commerce opportunity' (2025), as reported by Digital Commerce 360.
• Adobe Digital Insights, AI traffic data for July 2026, as reported by Digital Commerce 360.