Why does ChatGPT recommend your competitor and not you?

Why does ChatGPT cite your competitor and not you?

Because the assistant recommends the brands it can read, already knows, and that match the criteria in the question. If your competitor checks those three boxes better than you do, it shows up and you don’t.

The scenario is common. A prospect asks ChatGPT “what invoicing software should a contractor use?”. Three names come back, with one sentence on each. Your direct competitor is on the list. You are not. The same prospect asks Gemini, then Grok: same result.

This is not advertising, and it is not paid favoritism. OpenAI states that ads, tested since February 9, 2026 in the United States on the Free and Go plans, do not influence answers: “Ads run on separate systems from our chat model, and advertisers have no ability to shape, rank, or alter ChatGPT’s responses” (Ads in ChatGPT). The cause lies elsewhere, and you can diagnose it.

An assistant answers in one of two ways. Either it searches the web: OpenAI says ChatGPT “may search the web automatically when your question would benefit from current information” (Searching the web with ChatGPT). Or it answers from memory, with what the model learned up to its knowledge cutoff date, listed on each model page (OpenAI, Models). The six causes below affect one mode or the other.

In what order should you look for the cause?

From the most technical to the most commercial: access first, then content, then reputation, and finally the offer itself. There’s no point rewriting your pricing page if OAI-SearchBot is blocked.

flowchart TD
  A[Competitor cited, not you] --> B{Crawlers allowed?}
  B -- No --> B1[Fix robots.txt and CDN]
  B -- Yes --> C{Does a page answer?}
  C -- No --> C1[Create the dedicated page]
  C -- Yes --> D{Cited by third-party sites?}
  D -- No --> D1[Comparisons, reviews, press]
  D -- Yes --> E{Clear offer with numbers?}
  E -- No --> E1[Publish price, audience, timelines]
  E -- Yes --> F[Compare the criteria]

Which cause applies to you? The 10-question test

The ten questions below pinpoint it in two minutes: answer honestly, “I don’t know” included, and every “no” points you to the fix.

Is your site blocking AI assistant crawlers?

This is the first thing to check, because it’s the only cause that is guaranteed to keep you out of answers that use web search. A site ChatGPT isn’t allowed to read won’t appear there as a source, no matter how good its pages are.

In other words: blocking OAI-SearchBot removes you from ChatGPT search answers. GPTBot, on the other hand, is used for training: blocking it doesn’t stop you from appearing in search, and the two settings are independent. After you fix your robots.txt, OpenAI says it takes about 24 hours for its systems to adjust. Its publisher FAQ sums it up: to be included in ChatGPT’s summaries and snippets, don’t block OAI-SearchBot.

Google works differently. To be a link in AI Overviews or AI Mode, a page must be “indexed and eligible to be shown in Google Search with a snippet” (AI features and your website). For Gemini, the Google-Extended token controls whether your content is used to train future Gemini models and to ground answers in Gemini apps (grounding feeds them content from the Google Search index). It has no effect on your presence in Google Search (Google’s common crawlers). In the xAI API, Grok has tools to search both the web and posts on X (xAI, Tools Overview), so your mentions on X can be read too.

A less visible trap sits in your CDN. On July 1, 2025, Cloudflare switched to blocking AI crawlers by default (Content Independence Day), and its documentation announces new defaults for new domains starting September 15, 2026: training crawlers and agents blocked on pages that show ads, search crawlers allowed (Block AI bots). So a site can have an open robots.txt and still turn some crawlers away. The crawler-by-crawler breakdown is in the guide robots.txt, Cloudflare and AI crawlers: is your site blocking ChatGPT?.

Do you have a page that answers the exact question asked?

Often not: your competitor has an “invoicing software for contractors” page, while you only have a generic homepage. The assistant cites the page that answers, not the brand that could answer.

Assistants don’t run your question as is. OpenAI explains that ChatGPT search “typically rewrites your query into one or more targeted queries” sent to partner search providers (Searching the web with ChatGPT). Google describes the same principle for AI Overviews and AI Mode, called “query fan-out”: several related searches, one per subtopic (AI features and your website). Every sub-question is a chance to be found, as long as you have the page that matches it.

To check, type your prospect’s question into Google and Bing, then two short rewordings (“invoicing software contractor”). If your pages don’t show up for any of them, the assistant is unlikely to find them.

To fix it:

1. List the 5 to 10 questions a buyer asks before choosing, without your name in them.
2. Check that a page on your site answers each one in its first few lines.
3. Create the missing pages: one use case, audience or industry per page, with facts rather than adjectives.

The full method, including what is documented and what is myth, is in the guide GEO: how to get cited by ChatGPT, Gemini and Grok.

Are other sites talking about you?

If the sources the assistant reads talk about your competitor and not you, it will cite your competitor. An assistant that searches the web can only recommend what it finds, and it rarely finds just your own site.

Part of this is documented. For products, OpenAI lists among the factors it considers structured metadata (price, description) provided by the merchant or by third parties, and “other third-party content”, and specifies that review summaries are “based on reviews from public websites” (Shopping with ChatGPT Search). The rest is observed: a University of Toronto study published in September 2025 finds that AI search engines lean heavily toward third-party sources (“earned media”) over brand-owned content, far more than Google does (Chen et al., arXiv). That’s a study, not a rule published by the providers.

To check, open an answer where your competitor is cited and click Sources. Write down the domains: comparisons, business directories, forums, press articles. Those are the pages where a mention of your brand carries the most weight.

Does AI know your brand without searching the web?

Not necessarily: when the assistant answers without searching, it relies on what the model learned up to its cutoff date. A new brand, or one rarely mentioned on the web at the time, isn’t in there. A competitor that has been around for ten years is.

To check, ask ChatGPT “What do you know about [your brand]?” and see whether the answer shows sources. No sources means the model’s memory is talking: a wrong or vague description shows it barely knows you. With sources, look at which ones.

There’s no shortcut: a model’s memory changes with the next models, trained on a web where your brand is, or isn’t, more present. Until then, web search makes up for it, which is why the two previous causes matter. Worth knowing: blocking GPTBot asks OpenAI not to train its models on your content. A legitimate choice, as long as you make it knowingly.

Is your offer described with comparable facts?

An assistant compares facts: price, target audience, timelines, terms, specialties. If your offer page talks about “tailored solutions” with no price and no audience, the assistant can’t match it to a question like “for a 20-person SMB, budget €200 per month”.

OpenAI illustrates this for products: ChatGPT may consider “available options, price, reviews, and ease of use”, and if the user gives a $30 budget, it focuses more on price (Shopping with ChatGPT Search). An offer with no listed price has little grip on questions that include one.

To check, reread your offer page like a prospect in a hurry: for whom, what, how much, how fast? Then do the same read on your competitor’s page. The gap is often obvious right away.

Does your competitor simply answer the question better?

Sometimes, yes, and you have to accept it. If the question is about “the cheapest” and your competitor is, no amount of optimization will put you ahead on that particular question.

Read the sentence the assistant writes about each brand it cites: it reveals the criterion it used (price, specialty, track record, reviews). There are three cases:

• The criterion is right and you lose on it: change the playing field. Target the questions where your offer truly wins (an audience, a use case, a timeline) and create the page that proves it.
• The criterion is right but the assistant ignores you: the problem is in your published facts or your mentions (previous sections).
• The criterion is wrong (outdated price, discontinued service): fix the information on your site and on the third-party sites that carry it.

This work happens question by question. Choosing which questions to track, and how to phrase them without your name in them, is covered in the guide choosing which questions to track in AI assistants. To put a number on the gap with your competitors, the guide calculating your AI share of voice gives the formula.

How do you check without misdiagnosing?

By measuring several questions, across several assistants, several times. A single answer proves nothing: it depends on the wording, the context and whether the assistant searched or not.

OpenAI says so for its product results: ChatGPT “considers your query and context (such as Memory or Custom instructions)” (Shopping with ChatGPT Search). Your own account is therefore not a reliable witness. A few rules:

• Never put your brand in the question: you would only measure the echo of your own wording.
• Ask the question like a buyer, with their criteria (“for a contractor”, “under €50 a month”).
• Compare at least two assistants: ChatGPT may go through partner search providers, while Gemini relies on Google’s index.
• Measure again after each fix, and at regular intervals.

Their limits: the test asks one question, once, with web search, so it says nothing about model memory or change over time. Recurring tracking on paid plans covers ChatGPT, Gemini and Grok, but not Claude, Perplexity or AI Overviews: if those assistants matter in your industry, check them by hand. And the audit reads robots.txt: a block set at the CDN level doesn’t show up there, so check it in your CDN settings. To see which brands already come up, the AI Visibility Index publishes rankings by industry.

What should you do this week?

Start with what you can fix in a day, then move on to what takes weeks.