EtymaEtyma
    Back to blog
    AEO

    Why Does ChatGPT Recommend My Competitor and Not Me?

    Someone types your category into ChatGPT, gets back four names, and yours is not among them. There is nobody to complain to, but there is something to fix. The five reasons this happens, in the order we would check them.

    Etyma Team
    September 16, 2026
    8 min read
    Why Does ChatGPT Recommend My Competitor and Not Me?

    Someone types your category into ChatGPT, gets back four names, and yours is not among them. You can't complain anywhere, but you can fix it. Whatever the model said about your category came from somewhere, and the work is finding out where.

    Here are the five reasons this happens, roughly in the order I would check them.

    1. You were not in the pages it read

    For commercial questions, the AI is not recalling your brand from training. It runs a search, reads a handful of pages, and summarises them. So the real question is not "what does ChatGPT think of us", it is "which pages did it open, and were we on any of them".

    Those pages are mostly not yours. Similarweb's analysis of around 600,000 citation events found Wikipedia and Reddit alone account for a quarter of ChatGPT's citations. Your competitor is probably not winning because their homepage is better written. They are winning because they appear in a comparison post, a forum thread and a review listing that the model happened to read.

    This is the first thing to check and the most common answer. Look at the sources behind the answer that annoyed you, then work out how to be present in them.

    2. You are in the tail, and the tail rotates

    Ask the same question twice and you often get different names. Conductor tested this across 14,000 API calls: at comparison intent the leading brand stayed the same 91% of the time, but overall only about six in ten brands reappeared between any two runs.

    That pattern matters. If you are the category leader, you are stable. If you are anyone else, you are in a rotating cast, and the single answer your CEO screenshotted may not be the answer anyone else got that morning. Before you rebuild your content strategy around one result, run the prompt ten times and count.

    The source mix moves too, sometimes violently. Semrush tracked the most-cited domains across 100 million citations and watched Reddit fall from roughly 60% of ChatGPT citations to about 10% within six weeks in late 2025 before settling. Anyone who had spent that quarter building a Reddit-first strategy learned something expensive.

    3. The crawler cannot read your site

    Unglamorous, and more common than anyone admits. When Vercel analysed AI crawler behaviour, no major AI crawler executed JavaScript, and ChatGPT's crawler wasted 34.8% of its fetches on pages that returned 404, against 8.2% for Googlebot. That study is from late 2024 and deserves a refresh, but the mechanism has not changed.

    If your product pages render client-side, or your key content sits behind an interaction, the crawler may be seeing an empty shell. Turn off JavaScript in your browser and load your own pricing page. It takes a minute and occasionally ends the investigation right there.

    4. Your page is too new, or your category has an incumbent problem

    AI answers lean older than you would expect. Ahrefs looked at nearly 17 million cited URLs and found AI assistants cite content averaging 1,064 days old, with ChatGPT the freshest at 958 days. New pages do get picked up, but a page published last week is competing with one that has had three years to accumulate links and mentions.

    Concentration compounds it. In Semrush's index of 126 million prompts, the top three brands in news and media held 82.9% of the visibility in their category. Some categories are simply won, and the honest answer for a challenger is that this is a two-year project, not a two-week one.

    The route in is not to outrank the incumbent on your own site. It is to appear inside the pages the incumbent is already cited in, which is a far cheaper piece of work and one most teams have never explicitly tried.

    5. You are measuring something the model was never asked

    Two measurement traps.

    The first is prompt selection: teams track the question they wish buyers asked rather than the one they do.

    The second is more technical. Research from Graphite found only around 7% of US ChatGPT responses carry citations at all, and that forcing web search in a tracking tool shifted measured visibility by an average of 20 percentage points on prompts that would not normally trigger a search.

    So a tool can report that you are invisible on a prompt that no real user ever fires with search enabled. That is not a brand problem, it is an instrumentation problem, and it costs teams real money in misdirected content work.

    The reason that is not on this list

    None of this is a penalty. The model has not judged your brand, there is nothing to appeal, and nobody at OpenAI has a list you can be added to. If someone offers to get you "registered" or "listed" with the AI engines, they are selling something that does not exist. The engines read the open web, and the open web is where the work happens.

    What I would do about it, in order

    • Run the prompt ten times, across at least three engines, and write down what comes back. One screenshot is an anecdote.
    • List every domain cited in those answers. That list, not your content calendar, is the work queue.
    • Check your own site renders without JavaScript, and that the pages you care about return a 200.
    • Go and be present in the cited sources: the comparison posts, the review platforms, the forum threads, the YouTube reviews. This is PR and content work, and it is slow.
    • Re-measure in ninety days, not ninety minutes.

    Where a tool fits

    You can do the first pass by hand, and I would encourage anyone to try before buying anything. It stops being practical at about the third week, which is where monitoring earns its money.

    What you want from it is narrow: the same prompts asked of every engine daily, the full answers stored so you can read what actually changed, and the cited sources ranked. Etyma does that across ChatGPT, Gemini, Google AI Overview, Claude and Perplexity on every plan, which is why it is what I point people at when the question is "who is the model listening to". If your problem is technical, a tool that reads your server logs will serve you better.

    Sometimes the model recommends your competitor because your competitor is genuinely better documented on the open web: more reviews, more third-party coverage, clearer pricing pages, more people talking about them. No dashboard fixes that. What it can do is tell you exactly where the gap is. This is worth fixing, given that 71% of B2B buyers visit a vendor's website after their AI mentions it.

    Sources

    All posts

    Enjoyed this article?

    Get in touch