Why ChatGPT Doesn't Recommend Your Business (and How It Picks Who It Does)

"Why doesn't ChatGPT recommend my business?" often comes up after someone typed "best [their category]" into ChatGPT and got a list of five competitors. The question behind it is fair. You have a real product, a website and customers. Why do the same three names come back every time?

Most answers to that question are general advice. This one uses MentionAgent's own AI visibility scans, which put the questions a buyer would ask to ChatGPT's model, Gemini and Claude for 41 small brands, twice a month, and record whether the brand was named. Below are the six reasons we see, how to tell which one is yours, and what to fix first. The full numbers and how they were collected are at the end.

ChatGPT answers from memory or from a search

For every question, ChatGPT either answers from what the model learned in training or runs a web search and summarises the results. OpenAI's help center says ChatGPT "may search the web automatically when your question would benefit from current information", and users can also start a search themselves. The two paths treat a small business very differently:

Brand rates are across all 2,213 ChatGPT answers; competitor rates cover the 35 brands that listed competitors.

From memory, ChatGPT almost never names a small brand: 4 times in 562 answers with no sources. It still names the established competitors nearly half the time, most likely because they appear far more often in the text it was trained on. A web search is where a small business gets its chance, and even there the brand was named in fewer than one answer in ten.

That split gives you the first diagnostic question for every miss: did ChatGPT show sources?

Reason 1: ChatGPT never searched

If the answer has no source links, ChatGPT answered from training data. Nothing you publish this week changes that answer, because the model is not reading the web for it. Your business is either in the model's memory of your category or it is not.

You can't make your buyers' ChatGPT search (they can, you can't), but you can see which of your buyer questions get answered from memory. In our scans ChatGPT skipped the search on a quarter of questions overall, and the split by question type is lopsided: it answered from memory on 66.9% of "how do I" questions, 15.2% of "best X for [audience]" questions and only 4.1% of "best X" and "alternatives" questions. Those questions are long-term work: being named in enough articles, reviews and discussions that the next model learns you belong in the category. Treat them as a months-long project and measure them separately.

The questions where ChatGPT does search are the ones you can move now, so start there.

Reason 2: The lists ChatGPT reads don't name you

When ChatGPT searches on a "best X" question, it mostly reads other people's lists. We sorted the 12,973 links ChatGPT cited by URL. In answers to "best [category]" and "[competitor] alternatives" questions:

• 66.6% of answers cited at least one list, review or comparison article ("best", "top 10", "alternatives", "vs", "review" in the URL).
• 28.9% of all cited links were that kind of page, and 18.0% were a company's homepage.

An answer built on a search draws on the pages ChatGPT just read and links them as sources. On "best X" questions those are mostly lists and reviews, so whether those particular articles name you matters more than anything on your own site. If none of them mention you, ChatGPT has no text in front of it that does.

This is the most fixable cause. Ask ChatGPT your top three buyer questions, click the sources on each answer, and check which list articles leave you out. Those authors are your outreach list; the list articles ranking for the same questions on Google and Bing are worth adding. Our guide to listicle outreach covers how to ask, and how to get cited by ChatGPT covers what makes a page the one it links.

Reason 3: Your competitors own the category words

35 of the 41 brands told us their competitors when they signed up. We checked every answer for those names:

ChatGPT named a brand's own competitors more than eight times as often as the brand. For 15 of the 35 brands, ChatGPT named at least one competitor during the three months and never named the brand at all. That is the "ChatGPT recommends my competitor" experience, measured.

The most likely reason is exposure, not preference: an established competitor is named in far more reviews, comparison posts and roundups for your category than a small brand is, both in the text the model was trained on and in the pages a search turns up. Even when ChatGPT answered from memory, it named a listed competitor 47.3% of the time. Every independent page that names you alongside your category and your competitors moves that balance, and pages that compare you with a named competitor are worth prioritising, because they match the "[competitor] alternatives" questions buyers ask.

Reason 4: You only show up for one kind of question

The panels mix three kinds of question, and the rate at which ChatGPT named the brand differed by type:

Category questions are where small brands do best, and they are also where ChatGPT almost always searched: it answered only 4.1% of them from memory. Problem questions are where ChatGPT is weakest for them, and reason 1 explains why: it answered two-thirds of them from memory, where small brands almost never appear. If you only tested "best [category]", test the other two shapes too. A brand that never shows up on problem questions needs pages, its own and other people's, that answer the problem and name the product as one way to solve it. These test prompts cover all three shapes.

Reason 5: OpenAI's search crawler can't reach your site

OpenAI runs separate crawlers for separate jobs. GPTBot collects training data. OAI-SearchBot is the one used for ChatGPT's search answers, and OpenAI's crawler documentation says sites that opt out of it "will not be shown in ChatGPT search answers, though can still appear as navigational links."

Some sites added rules to block AI crawlers, through robots.txt or a bot setting in their CDN, and a broad rule can catch OAI-SearchBot along with the training crawlers. Open yoursite.com/robots.txt and look for a Disallow under OAI-SearchBot or under User-agent: *, then check your CDN's bot settings. This reason only affects your own pages being cited. ChatGPT can still name you from other people's pages, which is one more reason the list articles matter more than your own site.

Reason 6: You were named once, and that was luck

Some people ask this question because ChatGPT did recommend them, once, and then stopped. ChatGPT's answers vary from one ask to the next. On questions where ChatGPT named a brand at least once and the question was asked two or more times, it named the brand in 61.4% of the asks. Gemini was at 57.3% and Claude at 53.8%.

So "it recommended us yesterday and not today" is normal, even for brands it knows. The cure is to stop judging from a single screenshot. Ask each buyer question several times, on different days, and count how often you appear. Our guide to checking whether ChatGPT mentions your brand sets that up by hand. MentionAgent's paid AI visibility scan asks the questions for you twice a month; the free AI Mention Checker checks a few public signals (Wikipedia, Hacker News, GitHub and PageRank) rather than asking ChatGPT itself.

The engines also disagree with each other. On 674 questions asked to all three, Gemini or Claude named the brand on 94; ChatGPT missed it on 56 of those. On 20 questions only ChatGPT named it. Visibility in one engine says little about the others, which is why the engines need checking separately.

How to find which reason is yours

Pick your three most important buyer questions, one of each shape from reason 4, and ask each one in ChatGPT three times. Then read the answers against this table:

Many businesses hit more than one of these. Reasons 2 and 3 are usually the same problem seen from two sides: the pages that name your competitors are the pages that don't name you.

What to fix first

1. Rule out reason 5. It takes five minutes and nothing else works on your own pages while it is true.
2. List the pages ChatGPT cites for your category questions. Click the sources on the answers where it searched. This is the most concrete outreach list you will get, because ChatGPT has told you which pages it reads.
3. Get onto those pages, or pages like them. Ask the author to add you to an existing list, offer a comparison or a data point, or write for the site. The ChatGPT mention playbook lists the placements in priority order.
4. Turn existing mentions into usable ones. Articles that already name you but are thin on detail, or name you without a link, are the cheapest wins. See unlinked brand mentions.
5. Measure the rate, by engine, over time. Answers vary between asks (reason 6), so compare how often you are named across several rounds of asking, not single answers.

One measurement tip: OpenAI's publishers FAQ says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs, so filtering your analytics on that tag shows which of your pages ChatGPT sends people to. Answers from OpenAI's API tag links differently: all 12,973 links in our scans carried utm_source=openai.

For the bigger picture on how the engines pick sources, our AI citation study covers which domains each engine cites and how fast those sources turn over, and the GEO guide covers the rest of the field. If ChatGPT search specifically is your target, ranking in ChatGPT search goes through it step by step.

If you run a Shopify store or a local business

Our scans covered product and service brands, not single stores or single-location businesses, so we have no separate numbers for either. What carries over is the method, not a guess about which sources ChatGPT prefers for your market.

Shopify and other online stores. Ask ChatGPT the product questions your customers ask ("best [product] for [use]"), several times, and click the sources on every answer that has them. Those pages are your outreach list, exactly as in reason 2. Check reason 5 too, including any bot-blocking setting in your CDN or store platform.

Local businesses. Do the same with "best [service] in [town]" questions. Note which sources ChatGPT cites and which competitors it names, then find where those competitors are mentioned and you are not.

The full numbers

Each brand gets a panel of about 15 buyer questions written so they never contain the brand's name: "best [category] tools", "[competitor] alternatives", "best [category] for [audience]", and "how do I [solve the problem]". Every question goes to all three engines on scan days. Between 8 July and 5 October 2026 that produced 6,587 answers to 674 questions.

So if ChatGPT does not recommend you, you are in good company. Half of these brands, all of them MentionAgent customers building links, were not named by ChatGPT once in three months. 13 of the 41 were not named by any of the three engines.

Method and limits

• Source. MentionAgent's AI visibility scans, stored in its production database: 41 customer brands, 674 buyer questions, 6,587 answers, 8 July to 5 October 2026. Questions were written for each brand from its category, competitors and audience, and never contain the brand name. Placement-probe questions are excluded.
• What "ChatGPT" means here. The scans call OpenAI's API (gpt-4o-mini) with the web search tool switched on, and the model decides whether to use it. The ChatGPT app uses larger models and can draw on a user's history, so its answers will differ. Gemini is gemini-2.5-flash with Google Search grounding; Claude is Claude Haiku 4.5 with web search.
• "Named". A brand counts as named when its name appears in the answer text. A competitor counts when any name from the brand's own competitor list (four characters or longer) appears. Short or common-word names can over-count; the competitor lists were written by the brands themselves and include some large companies.
• "List article". Classified from the cited URL only ("best", "top 10", "alternative", "vs", "review", "compar" in the path). Some list pages have plain URLs, so the true share is likely higher.
• Small sample of brands. 41 brands, mostly small sites building links. Read the rates as what happens to a small business, not to brands in general.

Frequently asked questions

Related guides