ChatGPT shopping visibility: how to get your products recommended

ChatGPT recommends products from two evidence layers: the web content its search retrieves and, for participating merchants, the structured product feed OpenAI ingests through its commerce programme. To be recommended you need extractable product pages, complete Product schema, and a brand AI engines can verify. This guide covers each layer and what the ecommerce visibility data shows.

Product research is moving into the chat window. Someone asks ChatGPT for “the best ergonomic office chair under £400” and receives a shortlist of three or four products, often with prices, trade-offs and links. If your product is not on that shortlist, you were eliminated before the customer ever saw a search results page. This guide explains how those shortlists are built and how to get onto them.

How does ChatGPT decide which products to recommend?

For most queries, ChatGPT builds product recommendations the way it builds any answer: its search capability retrieves candidate pages from a web index, then the model selects sources it can extract clear, trustworthy claims from. Our guide on how ChatGPT cites websites covers the general mechanics; for shopping queries the extractable claims are product-shaped: what it is, what it costs, who it is for, why it beats alternatives.

That has two implications. First, crawler access is a hard gate: if GPTBot and OAI-SearchBot are blocked in your robots.txt, your products cannot be considered. Second, extraction quality decides selection. A product page that states “1.2kg carbon frame, 40-hour battery, £329, in stock” in clean text and schema gives ChatGPT everything it needs. A page that hides those facts in images, tabs rendered only by JavaScript, or vague copy gives it nothing.

There is also a third factor: corroboration. Shopping answers lean heavily on comparison and “best of” content from sources the engine trusts. If independent buying guides, category roundups and reviews never mention you, ChatGPT has no second source to verify your claims against, and unverifiable products rarely make the shortlist.

What are OpenAI’s product feed and the Agentic Commerce Protocol?

Alongside organic retrieval, OpenAI now operates a structured commerce layer. Its developer documentation for the Agentic Commerce Protocol describes how merchants build commerce flows inside ChatGPT, and the accompanying product feed specification is explicit about the mechanism: merchants provide a structured product feed file that OpenAI ingests and indexes, so ChatGPT can display products with accurate, up-to-date price and availability.

The feed spec is worth reading even if you never join the programme, because it reveals what OpenAI considers essential product data: stable unique identifiers, GTINs, precise titles and descriptions, price, availability and seller context. It even includes explicit flags, such as is_eligible_search, that control whether a product can be surfaced in ChatGPT search results at all, and a separate flag for direct checkout inside ChatGPT.

Strategically, this is the pattern we describe in the agentic internet: being in the dataset is the new being in the directory. Merchants on platforms with feed integrations will increasingly get accurate representation by default. Everyone else depends entirely on how well their public pages communicate the same facts.

Does Product schema still matter?

Yes, and it may matter more than the feed for most businesses, because it is the layer every AI engine reads, not just ChatGPT. As our GEO for ecommerce guide sets out, ecommerce has a natural advantage here: products are inherently structured data. The work is making that structure machine-readable:

• Product schema on every product page, complete rather than partial: name, brand, price, currency, availability, identifiers.
• Review and AggregateRating markup where you display genuine reviews. “Is it worth buying?” is a core AI shopping query, and a 4.8 average from 340 reviews is a citable fact.
• FAQPage schema on product and category pages answering sizing, compatibility, shipping and returns questions.
• Category-level buying guides that answer comparison queries honestly, because those are the pages shopping answers cite most.

Note the alignment: the fields OpenAI’s feed spec marks as required are largely the same facts Product schema carries. Whether the consumer is ChatGPT’s commerce index or Perplexity’s retrieval, the winning move is identical: publish complete, consistent, structured product facts.

How visible are ecommerce brands today?

Less than you would expect. SearchScore’s ecommerce AI visibility leaderboard audited 23 leading ecommerce brands: the average score was 31 out of 100, and the top score, Shopify’s 57, only reached the Emerging tier. Amazon scored 30, placing it in the Invisible tier, and Etsy scored 8. These are measurements of each site’s own AI-readiness signals, and they show that scale, traffic and brand recognition do not transfer to AI search visibility.

The wider dataset says the same thing. Across 850,000+ websites in the Q2 2026 SAVI report, the average AI visibility score is 34 out of 100 and structured data is the weakest category at 23.1. For a mid-sized retailer, this is genuinely good news: the giants have not locked up the AI shopping shelf, because almost nobody has done the structural work yet.

Which shopping queries should you optimise for?

Not all product queries behave the same way in ChatGPT, and knowing the difference tells you which pages to build.

Category shortlist queries (“best standing desk for small home offices”) are won by comparison content and third-party corroboration. Your product appears here because a buying guide, roundup or well-structured category page put it in contention. If you only have product pages, you are structurally absent from this query class.

Constraint queries (“waterproof hiking boots under £150 with a wide fit”) are won by attribute completeness. The engine is filtering, and products whose size ranges, materials, prices and use cases are stated as plain facts survive the filter. Vague lifestyle copy fails silently.

Verification queries (“is the X200 worth it”, “X200 vs Y300”) are won by honest review content, visible ratings and consistent specifications across sources. These are the last questions a buyer asks before purchasing, which makes them the highest-value citations on the list.

Map your catalogue against these three query shapes and you have a content plan: buying guides for shortlists, complete structured attributes for constraints, and comparison plus review content for verification.

How does agentic shopping change the funnel?

Shopping is where agentic behaviour bites first, because purchase research is exactly the multi-step filtering work agents automate. As covered in our agentic search guide, agentic web traffic grew 1,300% in eight months, and agents take multiple research steps per query, searching, cross-referencing and eliminating options before a human sees anything.

An agent shortlisting office chairs does not scroll your beautifully designed category page. It checks whether your product attributes are consistent across your site, your reviews and your listings, whether your price and availability are current, and whether independent sources corroborate your claims. Contradictions get you eliminated silently. The funnel no longer starts at your homepage; it starts inside a model’s evaluation loop, and only structured, consistent, verifiable products survive it.

What should you do this quarter?

A practical order of operations for an ecommerce team:

1. Verify crawler access for GPTBot, OAI-SearchBot and the other AI crawlers in robots.txt.
2. Audit your product schema coverage. Every live product needs complete Product markup; partial schema underperforms badly.
3. Fix cross-source contradictions in price, naming and specifications between your site, marketplaces and review platforms.
4. Publish honest category buying guides with FAQPage schema, because comparison content earns shopping citations.
5. Evaluate the OpenAI commerce programme if you run direct checkout at scale, and review the feed specification either way as a checklist of the product data AI systems expect.
6. Measure, then track. Run a free SearchScore audit to see where you stand, and check your sector’s position on the ecommerce leaderboard.

The brands winning ChatGPT shopping answers over the next year will not be the ones with the biggest ad budgets. They will be the ones whose product data an AI system can read, verify and safely repeat.