How to get your products recommended by ChatGPT: what OpenAI publishes, what you control, how to measure it

ChatGPT recommends products it judges relevant to the buyer’s request, using structured product data, public web information and the conversation’s context. You cannot pay for a recommendation: OpenAI says product results are not ads. What you control is the quality, freshness and reach of your product information, and you measure the result by tracking buyer questions over time.

What OpenAI actually publishes about product selection

Most advice on this topic is guesswork dressed up as method. Start from what OpenAI states in its own documentation, then separate it clearly from what is reasonable inference.

According to the OpenAI Help Center article on shopping with ChatGPT Search, a product appears in the carousel when ChatGPT perceives it as relevant to the user’s intent, taking into account the query and context such as Memory or Custom Instructions. The same article lists what the selection draws on: structured metadata from first-party and third-party providers (price, product description), other third-party content, and OpenAI’s safety standards and policies. It also says product results are selected independently and are not ads.

The shopping research help article describes a longer, guided mode: ChatGPT asks clarifying questions, then looks across the web for up-to-date price, availability, reviews, specs and images. It may use merchant product data provided through the Agentic Commerce Protocol, publicly available product information and other retail sources. Review summaries shown in that mode are generated from public reviews, and OpenAI notes that reviews and ratings are not verified by OpenAI.

What OpenAI does not publish is a ranking formula. Nobody outside OpenAI knows the weights. Any article that promises “the 7 ranking factors of ChatGPT” is presenting a hypothesis as a fact.

Organic recommendations and ads are two separate systems

This is the single most important point for budget decisions, because it changes what money can and cannot buy.

The Ads in ChatGPT help article states that ads do not influence ChatGPT’s answers, run on separate systems from the chat model, and that advertisers cannot shape, rank or alter responses. Ads appear below a response, labeled as sponsored and visually separated from it. OpenAI’s ads documentation describes ad selection as based on the context of the conversation, the ad’s landing page, title and copy, and context hints the advertiser provides at the ad group level.

The practical consequence: running ads does not make ChatGPT recommend you in its answer, and stopping ads does not remove you from it. Treat the two as separate channels with separate measurement. The split is covered in detail in ChatGPT organic vs ChatGPT Ads attribution.

The levers you actually control

1. Send a direct product feed

The OpenAI product feed specification lists the fields that make a useful discovery feed: item_id, title, description, url, brand, seller_name, image_url, availability and price. Two flags govern where an item can go:

• is_eligible_search: “true enables search eligibility; false disables it and checkout eligibility. Omitted or empty: true.”
• is_ads_eligible: “Set true for products Ads should process.” When omitted, ads processing is disabled unless a feed-level default applies.

A feed is not the only way ChatGPT learns about products, since it also reads public web information. The value of a feed is control: the price and stock OpenAI holds are yours, at the time you send them, rather than whatever a third party scraped last week. The specification is explicit about freshness: “Submit the current price; update the feed when a sale starts or ends” and “Update availability when the item becomes available or sells out.”

The full field-by-field walkthrough is in The OpenAI product feed spec, explained, and the reasons an item silently disappears are in Products not showing in ChatGPT shopping.

2. Make every product record complete and factual

The feed best practices ask for “concise, factual copy that helps users understand products”, valid and encoded URLs, a consistent seller name, and variant-specific title, description, media, availability and price when variants differ. They also recommend putting user-facing option dimensions such as color or size in variant_options.

Why this matters for recommendations, as a reasonable inference: a buyer’s request is specific (“a warm desk lamp under $60 for a small room”). A product can only be matched to the attributes it actually states. A record that says “Lamp LD-02” with a two-line description gives the model nothing to match against; one that states the light temperature, dimmability, size and price can be matched to many more requests.

A concrete checklist per product:

• A title that names the product the way a buyer would, with the distinguishing attribute (material, size, use).
• A description of a few factual sentences: what it is, who it is for, key specs, what is in the box.
• A brand and a GTIN or manufacturer part number where one exists, so the exact product can be matched across sources.
• Several images, and variant options named in words (not “Option 1”).
• Your return policy and seller name, consistent across the catalog.

3. Keep price and availability true

When several merchants sell the same item, OpenAI’s help content mentions price and availability among the factors weighed. A product listed as in stock that is sold out, or a price that changed yesterday, is a poor answer to a buyer and a reason to prefer another source. Daily full deliveries plus prompt updates on price and stock changes are the baseline.

4. Earn reviews, and expose them

The feed specification includes review_count (“Nonnegative whole number of product reviews”) and star_rating (“Average product rating on a 0 to 5 scale”). Shopping research summarizes public reviews. Reviews are a buyer signal you cannot fake and should not try to: OpenAI generates summaries from public review content, and an inconsistent or manipulated review profile is a risk, not a shortcut. Collect genuine reviews, show them on your product pages, and send the counts in your feed.

5. Let OpenAI’s search crawler reach your site

ChatGPT search uses a dedicated crawler. The overview of OpenAI crawlers states that “OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features”, while GPTBot is used to crawl content that may be used in training. The two are controlled separately in robots.txt. OpenAI recommends allowing OAI-SearchBot to help ensure your site appears in ChatGPT search, and notes it “can take ~24 hours from a site’s robots.txt update” for the change to apply.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Disallowing GPTBot is a legitimate choice about training. Disallowing OAI-SearchBot, often done by accident with a blanket bot rule or a firewall setting, removes your pages from search answers.

6. Be described consistently across the web

Shopping research gathers information from publicly available sites. A reasonable inference: when your product name, price and key attributes are consistent across your own site, your feed and the retailers or review sites that list you, there is less contradictory information for a model to reconcile. Inconsistent names across channels (a different title on a marketplace, an outdated price on a comparison site) are worth fixing for buyers anyway.

What does not work

• Buying ads to rank in the answer. OpenAI states ads do not influence answers.
• Keyword stuffing titles. The best practices ask for concise, factual copy. A title written for a model rather than a buyer reads as spam to both.
• Fake reviews or fabricated ratings. They break consumer protection rules in most markets and contradict the public review content a model can read.
• Scraping ChatGPT to “check rankings”. OpenAI’s terms of use forbid extracting data from its services by web scraping, other than as permitted through the API. And a single answer is not a stable ranking anyway: answers depend on the conversation’s context.

How to measure your visibility without scraping

You cannot see a ChatGPT “position report” the way you see a search console. What you can do is measure, repeatedly and honestly, whether your shop is named for the questions your buyers ask.

The method:

1. Write real buying questions, the way a buyer phrases them: “Which reading lamp works best in a small bedroom?” rather than “reading lamp”. Cover your main product types and price points.
2. Set the buyer’s country. Answers depend on where the buyer is and which shops sell there.
3. Ask each question on a schedule, weekly is enough, and record the shops named and their order.
4. Track two numbers per question: how often your shop is named across checks (citation rate) and your best position when named.
5. Record who is named instead of you. That list is your real competitive set in ChatGPT, often different from your search competitors.
6. Change one thing, then wait. Improve the product data for a category, then watch the citation rate over the following weeks rather than after one check.

Do it with an API call to an OpenAI model with web search, not by automating the ChatGPT interface. A model with web search is not the consumer app with a user’s Memory and history, so the absolute numbers are an approximation; the trend over time for the same questions is what carries the signal.

How Convrail helps

Convrail covers both sides of this work. On the data side, it validates every product against the OpenAI specification before delivery, shows a completeness score with the exact fix for each product, and can draft clearer titles and descriptions with AI for the feed only, never touching your store pages. See feed optimization and the product feed.

On the measurement side, the visibility module tracks the buying questions you write, per country, every week. For each question it shows how often your shop is named, your best position, and which shops are named instead. It runs on an OpenAI model with web search, with no scraping of ChatGPT, and is included in the Scale plan; details on the pricing page.

Summary

You cannot buy a ChatGPT recommendation, and nobody outside OpenAI knows its ranking formula. What OpenAI does publish points in one direction: relevance to the buyer’s request, judged on the product information available. Send a direct, fresh and complete feed, keep price and stock true, earn real reviews, let OAI-SearchBot crawl your site, and measure the result with the same buying questions over weeks rather than with one-off screenshots.