A customer asks ChatGPT for a provider of exactly your service, and the answer lists three firms. Yours is not one of them. Annoying, yes, but mostly it is information: the model leans on sources where others are documented better than you are.
Around 900 million people use ChatGPT every week, according to OpenAI (February 2026). A growing share of them ask for providers and suppliers instead of opening a classic search engine. Who the AI names helps decide who still makes the shortlist at all.
Why does ChatGPT recommend a competitor and not me?
Because the model assembles its answer from sources in which the competitor appears more often and more clearly. Not because they are the better company, but because in the texts the model saw or fetched at answer time, they are the better documented one.
A language model holds no judgment about professional quality. It reconstructs, from many pieces of text, which names sound plausible next to a service and a region. A firm that shows up consistently across trade portals, directories and press becomes the probable name. A firm that is missing there simply does not exist for the model in that context.
First find out which of the three stages is failing
Before you change anything, pin down what is actually not happening. "The AI recommends a competitor" can mean three different things, and each calls for a different response.
- Mention: your name appears in the answer, with no source. The model saw you during training.
- Citation: the AI points to a verifiable source, your website or a press article. This happens mostly when web search is on.
- Recommendation: the AI does not just name you, it puts you forward as a fitting option. This is the stage with the highest business value.
Merge the three into a single metric and you draw the wrong conclusions. Being mentioned but never recommended is a different problem from not appearing at all. In the first case the model knows you and ranks you down, in the second you are missing from the sources entirely.
The cause is almost always the source landscape
The competitor rarely wins on a trick. They win because the material the AI draws from describes them more densely and in better context.
Four kinds of source carry most answers. Trade portals and directories supply structured entries with service, location and category. Press and specialist media supply context and framing. Review platforms supply evidence that the offer exists in practice. And your own website supplies the self-description everything else aligns to. Remove one of those layers and your name drops out of exactly the answers that rest on it.
The distinction between training and retrieval matters. Some answers come from what the model once learned, others from a live search at the moment of the question. Block every bot wholesale and you can vanish from the second kind of answer without noticing. According to Cloudflare (August 2025), over 2.5 million websites block AI training entirely, often without weighing the effect on their own discoverability.
Why the most visible provider is rarely the technically best
Visibility in AI answers measures evidence density, not quality. A provider with strong marketing and many directory entries looks to the model like the obvious pick, even when a more specialised competitor solves the task better.
This does not disparage the competitor and it is not a fault in the AI. It is the logical result of how a model works: it prefers the well documented over the quietly competent. For you that means two things. You do not have to displace anyone, and you can close the gap by improving your own evidence, not by talking others down. The lever sits in your sources, not in the rival's.
What the model finds about you is something you set first
Before an AI can recommend you, it needs a coherent picture of who you are and what you do. Contradictory or thin information pushes the model, when in doubt, toward the more clearly described competitor.
Make sure your self-description reads the same everywhere first: same service, same location, same category on the website, in directories and in profiles. A plain, factual page about your firm helps the model match you to a query. And technically the page has to be readable without JavaScript and reachable for the relevant bots, or the best copy counts for nothing. How that fits together with data sovereignty and local processing we lay out under local AI.
What you can actually do
The first step is not a rebuild, it is a measurement. Without knowing at which stage and on which questions you fall out, you are changing things blind.
- Collect the questions a customer really asks and check, for each, whether you are mentioned, cited or recommended. How that self-test runs in detail is in Does ChatGPT mention my company.
- Note which competitor the AI cites. It usually reveals the sources you are missing.
- Close the weakest layer first: missing directory entries, an unclear website, references with no backing.
- Measure again across many runs. A single screenshot proves nothing, because the same question can name a different firm five minutes later.
One caveat for perspective: in 2024 Gartner forecast that classic search volume would fall 25 percent by 2026. By mid-2026 that has not happened in that form. AI answers arrive as a second channel, they do not replace the search engine. A firm that is findable on both channels today sits easier than one betting on a single scenario.
From diagnosis to measurement
The self-test tells you whether a problem exists. It does not tell you how big it is or on which queries it hurts.
A reliable measurement works with a battery of questions rather than a single prompt, with many repetitions, across several platforms and in your customers' languages. It keeps mention, citation and recommendation apart and reports frequencies instead of snapshots. How such a measurement is built is described on our page on AI visibility.
There are no guaranteed placements, and anyone who promises them has not understood the mechanism. You can improve the conditions under which a model names you instead of the competitor, and you can measure the progress. That much, and no more.
If you want to know how often the major AI systems already name a competitor in your place, a pilot project is the fastest route to real numbers. Reach us through the contact page and we will look at your case.