How to get your products recommended by ChatGPT

When a shopper asks ChatGPT what to buy, your product can show up in two ways. The written answer can name your brand, drawing on the web pages ChatGPT's search reads. And ChatGPT can show product results: your product itself, picked from product data. This guide covers what OpenAI says about both, and what a store can change, step by step. The same steps carry over to Perplexity, Google's AI Mode, Copilot and Claude, and the last sections cover what's different there.

How ChatGPT picks products, in OpenAI's words

OpenAI describes how it chooses the product results in ChatGPT search (OpenAI Help: Shopping with ChatGPT Search):

• They aren't ads, and OpenAI's partnerships don't influence them.
• They're built from structured product data, such as price and description, from merchants and third-party providers, along with other content from the web.
• The review summaries are written by the model from reviews on public websites. OpenAI doesn't verify them.
• When several sellers carry a product, ChatGPT weighs availability, price, quality, and whether the seller is the maker or the main seller.
• Which products appear depends on the shopper's question and on context such as their saved memories and custom instructions, and OpenAI says it expects to personalize merchant rankings further.

So four things are in a store's hands: whether ChatGPT can read your pages, whether your product data reaches it and is right, what your pages say, and what reviews and other sites say about you. The steps below take them in that order.

Step 1: Let ChatGPT read your store

OAI-SearchBot is the crawler behind ChatGPT search. GPTBot is a different crawler, for training OpenAI's models, and the two are controlled separately: you can block GPTBot and still let OAI-SearchBot in. A site that blocks OAI-SearchBot isn't shown in ChatGPT's search answers, though it can still appear as a plain navigational link. After you change robots.txt, OpenAI says it takes about 24 hours for search results to catch up (OpenAI: crawlers).

ChatGPT search sometimes also sends rewritten versions of a shopper's question to search providers it partners with. OpenAI doesn't name them, though it links Microsoft's and Shopify's privacy policies for them. To be eligible, OpenAI asks only that you allow OAI-SearchBot and let traffic from its published IP addresses through, and it says placement isn't guaranteed (OpenAI Help: Searching the web with ChatGPT).

Two places to check:

• robots.txt. Our free crawler check shows whether yours lets OAI-SearchBot reach your homepage, a product page and a category page, and which line decides.
• Your CDN or firewall. These can turn crawlers away whatever robots.txt says. Cloudflare, for one, lets a site block AI search crawlers across the whole site (Cloudflare).

Step 2: Get your catalog into ChatGPT

How your products reach ChatGPT depends on your platform (ChatGPT: product discovery for merchants):

• Shopify and Etsy. Their catalogs are already connected, so there's nothing to apply for. On Shopify, Agentic Storefronts, on by default for eligible stores, list your products in AI channels including OpenAI's (Shopify Help Center). Check that yours is on and that your products are eligible.
• WooCommerce, BigCommerce and other platforms. You can apply to share a product feed. OpenAI says a feed isn't required if ChatGPT already crawls your site, but it gives you more control over what ChatGPT knows about your products. Purchases happen on your own site.

OpenAI's merchant feed program currently targets the US. Its feed format asks for nine fields on every product: an ID, title, description, page URL, brand, seller name, image, availability and price. Identifiers such as GTINs, plus sizes, materials, shipping, return terms and review counts, are optional. OpenAI recommends sending the full feed at least daily (OpenAI: product feed spec). Whatever your platform sends, those are the facts to get right.

Step 3: Make each product page answer what buyers ask

A buyer doesn't ask ChatGPT for "espresso grinder". They ask for the best espresso grinder under $300 for a beginner who makes two drinks a day. An answer can only match your product to that question if your page states the facts it turns on: the price, the burr type, the grind settings, who it's for and who it isn't for.

Research on AI answers points the same way. In E-GEO, a test of rewritten Amazon product descriptions across five AI models, the descriptions that did best matched what the shopper was looking for, opened with a summary, used headings and bullets, listed use cases, answered buyers' questions and kept the facts unchanged. Length alone didn't help. In the original GEO paper, adding quotations, statistics and citations raised a page's visibility in AI answers most, and keyword stuffing did slightly worse than changing nothing.

For a product page, that comes to:

1. A two- or three-sentence summary at the top: what it is, who it's for, and the one or two facts that set it apart.
2. The specs buyers compare, written out in the text, not only in an image or a dropdown.
3. Use cases, and the questions buyers ask before they buy, answered on the page.
4. Specifics in place of adjectives: numbers, materials, test results, and quotes from real reviews.
5. Price and availability that match your feed.

Step 4: Earn reviews and mentions on public sites

ChatGPT's review summaries come from reviews on public websites, so the reviews it can see are the ones on your product pages, at retailers that carry you, and on review sites. Ask real buyers to leave reviews where people can read them.

The written answers draw on more than your own pages. Ask ChatGPT the questions your buyers ask and read the sources under each answer: review sites, roundups and community threads often appear there. Those are the places to be described accurately, by being reviewed, being included in a roundup, or answering questions in a community, always with your involvement disclosed.

Two rules from the FTC apply here. Its rule on reviews and testimonials, in force since October 21, 2024, bans fake reviews, including AI-written ones, and buying them. It also bans offering a reward for a review that says something positive, posting reviews from staff or their relatives without saying so, and suppressing reviews with threats, or hiding the negative ones while implying you show them all (FTC: rule questions and answers). And under its Endorsement Guides, a reviewer who got your product free or at a steep discount has to disclose it (FTC: Endorsement Guides).

Step 5: Keep price and availability right

When several sellers carry your product, ChatGPT weighs availability and price. A feed that shows an old price, or a product as out of stock, works against you there. Update the feed when prices or stock change, and keep the price on the page, in your structured data and in the feed the same.

Check where you stand

Ask ChatGPT ten or twenty questions your buyers would ask, each a few times, and write down which brands each answer names and which sites it cites. Turn memory and custom instructions off first, because ChatGPT can use them when it searches and picks products, and your own chats would shape what you see. One answer tells you little, because the same question can get a different answer each time. Repeat the same questions every month to see what moves.

The AI Shelf Report does this at scale: 20 buyer questions, asked three times each through the APIs of OpenAI, Perplexity and Claude with web search, with every brand counted. API answers aren't exactly what a shopper sees in the ChatGPT app, so each report also includes three labeled screenshots from the app.

The other AI engines

The same steps apply elsewhere, with a few differences:

• Perplexity. Its crawler is PerplexityBot (Perplexity crawlers). Perplexity says its answers are more likely to recommend merchants that supply detailed product data, such as availability, reviews, pricing and specs, and that its product listings aren't sponsored (Perplexity Help: What is Instant Buy?).
• Google AI Mode and AI Overviews. These are part of Google Search, so the work is SEO, with Merchant Center for product data. Our GEO vs SEO guide covers them.
• Copilot. Microsoft builds Copilot's answers on Bing's crawling and indexing, and recommends IndexNow to tell Bing when pages change, so answers don't cite outdated pages (Bing Webmaster Guidelines).
• Claude. Its search crawler is Claude-SearchBot. Anthropic says its crawlers follow robots.txt, and that blocking Claude-SearchBot may reduce a site's visibility in Claude's search results (Anthropic). Anthropic lists Brave Search among the providers behind Claude's web search (Anthropic: subprocessors).

What doesn't work

• Keyword stuffing. In the GEO research it did slightly worse than leaving the page alone.
• Paying to change the answer. ChatGPT shows ads on some plans, but they're labeled and shown separately below the answer, and OpenAI says they don't influence answers. Product results aren't ads (OpenAI Help: Ads in ChatGPT).
• Fake or bought reviews. They're the review summaries' raw material, and the FTC's rule bans them.
• An llms.txt file for ChatGPT. OpenAI documents robots.txt for its search crawler; it doesn't say ChatGPT reads other sites' llms.txt files.

Questions

Can I pay to be recommended by ChatGPT? Not in the answers or the product results. OpenAI says ads don't influence answers and product results aren't ads.

How long until ChatGPT recommends my brand? We can't give you a timeline. ChatGPT is run by OpenAI and changes without notice, which is why we measure the same questions on a fixed schedule rather than promise a result. A robots.txt change is the one thing with a stated time: OpenAI says about 24 hours.

Is this different from SEO? Mostly not: ChatGPT search has to be able to crawl and read your pages, as any search engine does. What's added is product data, reviews on public sites, and being named in a written answer rather than ranked in a list. The GEO vs SEO guide goes through the differences.