ChatGPT shopping — how to get your products recommended

How ChatGPT's shopping surface actually works

When a query reads as shopping intent, ChatGPT assembles a set of candidate products by searching the web in real time and drawing on structured product metadata it can read from retailer and brand pages. It then filters that set against the constraints in the prompt — price ceiling, features, use case — and presents a shortlist with merchant links. OpenAI has stated these results are organic and not ads: there's no pay-to-rank slot, so eligibility comes from how readable and trustworthy your product data is.

Two practical consequences follow. First, the products that surface are the ones whose key facts — price, availability, specs, ratings — are exposed as clean, machine-readable data rather than locked inside JavaScript or images. Second, because the surface pulls from across the web, your product can be recommended via a marketplace listing or a review roundup even if your own page is weak — which is exactly why your own page should be the strongest, most accurate source so the engine cites you on your terms.

Let the right crawlers in — and check your CDN, not just robots.txt

ChatGPT's shopping results depend on its web-fetching crawlers reaching your product pages: OAI-SearchBot builds its web answers and ChatGPT-User fetches specific pages on demand when a user asks. If either is blocked in robots.txt, your products can't be read for live recommendations. Note these are distinct from GPTBot, OpenAI's training crawler — blocking GPTBot has no effect on whether you appear in ChatGPT's live shopping surface, and confusing the two is a common, costly mistake.

For ecommerce there's a second gatekeeper people miss: the CDN or bot-management layer. Shopify, Cloudflare, Akamai, and similar services often challenge or block unfamiliar bots before robots.txt is even consulted — so the AI crawlers get a CAPTCHA or a 403 regardless of how permissive your robots.txt is. Verify the AI user-agents are allowlisted at the edge, and confirm a real fetch succeeds rather than assuming robots.txt is the whole story.

Product and Offer schema are the backbone of eligibility

Schema.org Product markup, with a nested Offers block, is how you state — independently of how your page renders — exactly what you sell and on what terms. It's the most important structured data for ChatGPT shopping because it hands the engine the precise fields it filters on, in a format it doesn't have to infer. Add a Product block to every product page with these fields, kept in sync with the live page:

• name, brand, and description — the exact product identity, so the engine doesn't conflate your item with a similar SKU.
• Offers with price, priceCurrency, and availability — this is how the engine satisfies 'under $300' and 'in stock' constraints; a stale or missing price gets you filtered out or misrepresented.
• aggregateRating and reviewCount — surfaces your star rating as a machine-readable trust signal instead of leaving it trapped in a review-app widget.
• gtin, mpn, or sku — global identifiers let the engine match your product to the same item on other sources and corroborate price and specs.
• Item attributes (color, size, material, and category-specific specs) via additionalProperty — these are the exact qualifiers shoppers put in natural-language prompts.

Product feeds and where your data lives off-site

On-page schema is one source; structured product feeds are another, and they increasingly feed AI shopping surfaces indirectly. A clean, accurate product feed — the same Google Merchant Center / Shopping-style feed many stores already maintain, with GTINs, prices, availability, and images — keeps your catalog represented consistently across the marketplaces, comparison sites, and aggregators that engines also read. ChatGPT shopping has integrated with merchant data partners, so well-maintained feed data improves the odds your product is both discoverable and described correctly.

The discipline that matters is consistency. The price, title, GTIN, and availability in your feed, your on-page schema, and your marketplace listings should all agree. When sources contradict each other — one says $249, another $279, a third says out of stock — the engine loses confidence and either omits you or surfaces a competitor whose data is internally consistent. Treat your feed, your schema, and your listings as one synchronized record, not three independent ones.

What makes a product get recommended over its rivals

• Constraint match — the product literally satisfies the qualifiers in the prompt (price, size, feature), and those qualifiers are stated as extractable text or structured data the engine can check.
• Review corroboration — a credible rating and review volume, ideally consistent across your site and independent platforms, signal a safe recommendation; a product reviewed only on its own store with no outside signal is riskier to suggest.
• Availability and accurate pricing — in-stock items with a current, correct price are favored; the engine avoids recommending products it might describe wrongly.
• Specification completeness — pages that plainly state the attributes shoppers ask about (battery life, dimensions, compatibility) win narrow, high-intent queries that vaguer listings can't answer.
• Cross-source consistency — the same product identity, specs, and price appearing in agreement across your page, marketplaces, and roundups builds the confidence that tips a recommendation your way.

Reviews and trust — why some stores get suggested and others don't

AI engines recommend products they're confident about, and for shopping that confidence rests heavily on reviews and corroboration. The catch for ecommerce is that reviews are usually rendered by a third-party app — Judge.me, Loox, Yotpo — that injects star ratings and text via JavaScript the AI crawler doesn't execute, so a five-star product looks ratingless to the engine. Exposing aggregateRating and reviewCount in your Product schema fixes this: it makes the rating a fact the engine reads directly, regardless of whether the widget runs.

Off-site corroboration compounds the effect. A product whose rating and specs appear consistently on established marketplaces, in reputable category roundups, and on comparison sites reads as a trustworthy recommendation; one that exists only inside your own store gives the engine little to verify against. The goal is for ChatGPT to find the same favorable, consistent picture of your product wherever it looks — that's what converts eligibility into an actual recommendation.

A ChatGPT shopping readiness checklist

• Confirm OAI-SearchBot and ChatGPT-User are allowed in robots.txt and not blocked by your CDN or bot-management layer (test an actual fetch, don't assume).
• View-source a product page and verify the title, current price, availability, key specs, and review rating are present in the raw HTML, not injected by JavaScript.
• Add valid Product schema with an Offers block (price, priceCurrency, availability), aggregateRating, reviewCount, and a GTIN/MPN to every product page; validate with Google's Rich Results Test.
• Synchronize your product feed, on-page schema, and marketplace listings so price, title, identifiers, and availability all match.
• Build and maintain review corroboration on independent platforms, and keep your specs consistent everywhere your product appears.
• Test it live: ask ChatGPT a constraint-based shopping query in your category and check whether your product surfaces and whether the price and specs it cites are correct.