ChatGPT shopping: how to get your products recommended

What shopping in ChatGPT looks like

When someone asks ChatGPT for the best budget running shoes, they can see product cards with an image, price, a short summary of reviews and links to retailers. Those cards are chosen from the web, not from paid placements. The person can then ask follow-up questions to narrow the list, such as asking for a wide fit or a shoe under a certain weight.

The same kind of question in Google AI Mode, Perplexity and Copilot produces similar product lists. The job is the same across all of them: be the product with clear, consistent data on pages the assistants read. If your product page is vague, out of date or missing key facts, the assistant has no reason to include it.

For example, a small coffee roaster in Bristol might find that asking an assistant for the best single-origin beans for a moka pot returns a list of larger brands. The fix is not to chase the assistant directly, but to make the roaster's own product pages answer that exact question with facts an assistant can quote.

Where the product data comes from

Assistants build product answers from three main sources. Each one matters, and they work best together. If one source is weak, the others can still carry you, but consistency across all three is what makes a product easy to recommend.

• Product pages with structured product data: name, brand, price, availability, ratings and identifiers such as GTIN, marked up with schema.org Product markup so the facts are machine-readable.
• Merchant feeds: Google's guidance for its own AI features points shops to Merchant Center; OpenAI has published ways for merchants to supply product data too. A feed gives assistants current prices and stock instead of a crawl of stale pages.
• Reviews and comparisons on other sites: the summaries in product cards draw on reviews across the web, so review profiles and listings on retail and comparison sites matter as much as your own page.

For example, a phone accessory seller in Toronto might have a perfect product page but no GTIN in the markup. An assistant comparing two nearly identical cases may skip the one it cannot identify with confidence. Adding the identifier and keeping the feed in step with the page removes that doubt.

The direction of travel is toward feeds and structured data doing more of the work. Assistants want current prices and stock, not a guess based on a page that was last updated months ago. If you sell on a marketplace as well as your own site, keep the product name and identifiers the same in both places.

Fix the product pages first

Your product pages are the foundation. Before worrying about feeds or agent checkout, make each page answer the questions a buyer would ask. This is the part you control completely, and it feeds every other channel.

1. One product per page, with the name, what it is for, price, variants, stock and shipping in text near the top. State the facts a buyer asks about: materials, sizes, compatibility, warranty.
2. Add Product structured data and keep it in step with the page. Let the AI crawlers in; a blocked crawler means no card. The free robots.txt checker tests them at once; link to /tools/robots-txt-checker.
3. Write the comparison the assistant would otherwise write: this product against the two obvious alternatives, honestly. Pages that compare are the pages that get cited.

For example, a dentist in Leeds might ask which electric toothbrush is best for sensitive gums. If your page for a sensitive-gum brush lists the brush head type, pressure sensor, battery life and what the warranty covers, an assistant has something to quote. If the page only says 'gentle and effective', the assistant has nothing to work with.

The comparison point is often missed. Assistants frequently answer by weighing options. If your page already does that work clearly, it becomes an easy source to cite. If it does not, the assistant will find another page that does, and that page may be a competitor's.

For more on how this differs from traditional search, see GEO vs SEO. The short version is that you are writing for a reader who is an assistant, not only for a person scrolling a results page.

Agents that buy

The next step is checkout inside the assistant. The person says buy it and an agent completes the purchase with a retailer. Google has referred shops to its agent-friendly web guidance and to the Universal Commerce Protocol, and OpenAI has built checkout into ChatGPT with partner platforms.

The Universal Commerce Protocol is an open protocol for AI agents to discover, compare and buy from shops in a standard way. Most small shops will not implement it by hand. Platforms will add support, and your job is to choose a platform that does.

For a small shop today this means: use a platform that supports these protocols as they arrive, keep product data accurate, and make sure your checkout works without tricks that an agent cannot follow. For example, a checkout that depends on a hover menu or a custom captcha may block an agent. A plain, standard checkout flow is more likely to work.

This is still early. The practical move is to keep your product data clean and your checkout simple, so that when agent checkout reaches your platform, you are ready rather than rebuilding under pressure.

Measure whether you are the product named

You cannot improve what you do not measure. Ask the assistants the buying questions your customers ask, in their country, and read which products and shops they name. Do this on a schedule, because answers change as pages and feeds change.

Vasukai asks ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Perplexity, Claude, DeepSeek and Sarvam on a schedule and shows who is named instead and which sites are cited. Plans start at $19 a month, and there is a free check with no card that shows what ChatGPT says about your business.

For example, a bicycle shop in Melbourne might check monthly for the question 'best commuter bike under 1000 dollars'. If the assistant names three other shops and cites two review sites, that tells the shop where to focus: its own product page, its review profile, or both.

The point is not to chase a single answer. It is to see the pattern over time: which assistants name you, which questions you appear for, and which pages get cited when you do not. That pattern tells you what to fix next.