ChatGPT does not recommend your competitors because it "prefers" them. When asked for a recommendation, it assembles its answer from two materials: what it retained from training, and what the web pages it consults at question time actually say. If your competitors keep coming up and you do not, it is almost always because those sources talk about them, and not (or barely) about you. On top of that, answers vary from one run to the next: a single answer proves nothing, in either direction.
What happens when you ask an AI for a recommendation
Ask ChatGPT, Claude, Gemini or Perplexity "which company would you recommend for [your trade] in [your city]?". In most cases, the engine rephrases the question, runs a web search, reads a handful of results, then writes a synthesis naming a few companies (and, often, citing its sources).
Two direct consequences:
- The answer depends on the sources consulted that day. The engine does not "know" your market: it restates what the pages it deemed relevant say (comparison articles, directories, press, reviews, service pages).
- The same question does not always produce the same answer. Asked twice, it can return two different lists. That is why any serious measurement relies on repetitions: our checks ask each question several times per engine and compute a citation rate, never a yes or no.
The most frequent reasons, as observed in measurement
- The consulted sources talk about your competitors, not about you. Engines lean heavily on third-party content: "best X in Y" articles, specialised directories, local press, review platforms. If you are absent from those, the engine simply has no material to cite you with.
- Their online presence is easier to pick up. A name written the same way everywhere, pages that state plainly who does what, where and for whom: that is what an engine can extract and restate effortlessly. A blurry or scattered presence cites poorly.
- Your visibility lives elsewhere. Word of mouth, closed networks and messaging apps leave no public trace: what is not written on the open web does not feed AI answers.
- Language and geography matter. A question asked in French mostly pulls French-speaking sources; a company visible only in another language starts with a handicap on that ground.
- The draw went against you. In our measurements, a company can appear in one answer out of two, or one out of ten. If you only tested once, you saw one draw, not a trend.
What our measurements show
Every week since June 2026, we have been asking panels of questions to the four major engines (ChatGPT, Claude, Gemini, Perplexity), with repetitions, across several markets. A few findings come back consistently:
- Each engine has its own podium. On the same question, the names one engine cites most are not the ones another does. "Why does ChatGPT recommend my competitors" is therefore a question to ask engine by engine: the answer differs.
- Being cited as a source is not being recommended. A website can rank among the pages engines consult most without the company ever being named in the answers. The reverse exists too.
- A good reputation and an absence of recommendations coexist. A company can be treated very well when you ask the AI about its name, and almost never appear when a customer asks for a recommendation without naming it. These are two distinct measurements.
- Engines love content that compares. The most reused sources are often third-party pages that rank, compare or answer a question, more than the companies' own websites.
What to do, concretely
1. Measure before concluding. Establish a baseline: the questions a customer would ask, asked continuously to several engines, with repetitions. That is exactly what an AI ranking tracker does; the full method is described on our how-it-works page.
2. Look at the sources, not just the names. The pages engines cite on your questions are your work list: that is where you need to exist. A comparison article you are missing from, a directory where your listing is empty, a press article that does not mention you are concrete projects.
3. Publish what engines can pick up. Pages that answer your customers' questions plainly, a consistent presentation of who you are and what you do, the same information everywhere. We detail these levers in "How to appear in AI answers".
4. Re-measure over time. Corpora and engine behaviours evolve; only longitudinal tracking separates real progress from an accident of the draw.
One honest note to close: nobody can guarantee that an AI will cite you, and you should be wary of anyone who promises it. What can be steered is your presence in the sources engines draw from, and the measurement of the effect over time.
Read next
Start by checking where you stand: "How to know if ChatGPT recommends your business" gives a fifteen-minute manual test method, and its limits. For the vocabulary of the discipline, see "What is GEO?".