ChatGPT Shopping for merchants: how products get recommended

Ask ChatGPT a buying question — best carry-on luggage, a gift for a coffee nerd, trail runners under $150 — and instead of a paragraph you increasingly get product cards: names, images, prices, merchants, links. That surface is what merchants informally call ChatGPT Shopping.

Two facts anchor everything else, both from OpenAI's public statements as of 2026: results are organic — there is no ad auction and no way to pay for placement — and product information is drawn from the web: OpenAI's crawling and search infrastructure plus third-party product data. The precise ranking mechanics are not public, and any guide claiming to know them precisely is extrapolating. What follows sticks to the levers with observable influence.

Why it matters despite young volume: a ChatGPT product recommendation is closer to a personal referral than a search result. The user asked a trusted assistant, and the assistant chose you. Merchants report small but unusually high-converting traffic from AI surfaces, which matches the wider 2026 industry finding that AI-referred visitors convert well above average.

Reconstructing from OpenAI's documentation and observed behavior, a product surfaces through three overlapping routes. Route one: the live index. OpenAI's search crawlers (OAI-SearchBot for its search index, ChatGPT-User for on-demand fetches, GPTBot for training) read your product pages like a search engine would. Complete, crawlable, structured product pages are the raw material.

Route two: structured product data. Machine-readable price, availability, images, brand, and reviews — schema.org Product markup — is what turns a web page into a product an assistant can confidently present with a price card. Incomplete data makes a product risky to recommend; assistants prefer products they can describe precisely.

Route three: reputation in the corpus. When ChatGPT explains WHY it recommends something, it leans on reviews, comparisons, and best-of content it has read. Stores that exist only on their own domain have one voice; stores mentioned across buying guides, community threads, and review sites have many. This is the slowest lever and the strongest one.

One — let the crawlers in. Check robots.txt for OAI-SearchBot, ChatGPT-User, and GPTBot. On Shopify that is robots.txt.liquid; on other platforms, your robots file. Blocking them removes you from consideration entirely; this lever is binary.

Two — complete Product schema on every product. Name, brand, description, price, currency, availability, image, aggregate rating where real reviews exist. This is the highest effort-to-impact ratio on the list.

Three — product pages that read like answers. A title that says what the thing is, a first paragraph that says who it is for and why, specifications in text rather than images. Assistants quote pages that explain themselves.

Four — earn third-party mentions. Best-of lists, comparison posts, community recommendations, review coverage. You cannot fake this and should not try; you can pursue it — pitch genuinely relevant roundups, cultivate reviews, publish comparison content on your own blog that honestly includes competitors (it ranks, it gets read, and it puts your name in the conversation).

Five — verify monthly. Ask ChatGPT your category's buying questions and record whether you appear. Expect run-to-run variation; judge trends, not single answers. Feed every question you lose into your content pipeline — that loop is the whole game.

Prompt-injection tricks — hidden text on product pages instructing AI to recommend you — are detectable, already being filtered, and exactly the kind of thing that gets a domain distrusted. Fake reviews poison the third-party corpus you need clean. Keyword-stuffed product descriptions read worse to a language model than to a human, and the model is the one choosing. And churning out thin AI-generated category content without answers in it builds nothing: assistants cite pages that resolve questions, not pages that exist for pages' sake.

The uncomfortable truth about this surface: it rewards being genuinely recommendable — clear products, honest data, real reputation — faster than it rewards optimization tricks. That is bad news for tricksters and good news for actual merchants.