The assistant named somebody else.
It is information, not a verdict. It means the assistant found more to say about them than about you, in the places it looked — and that is a different problem from being worse, with a different fix.
You asked an assistant the question your customers ask, and it recommended two companies. Neither was you. The reflex at that moment is to conclude the model has an opinion about your quality. It does not. It composed an answer out of what it could find written about your category, and somebody else was easier to write about.
First, confirm it is real
One ask is not a measurement, and acting on one is how a quarter gets spent on the wrong thing. Answers to the same question vary between runs, and an answer produced inside your own account is not the answer a stranger gets — your history and your location both lean it toward you, which means the version you saw is the friendly one.
- Ask from a clean session with no account history, set to the market your customers are actually in.
- Ask more than one phrasing. A buyer does not use your internal words for your category.
- Ask more than one assistant. They disagree with each other more than they disagree with themselves.
- Ask again in a week. If the names moved, you were reading noise the first time.
What it actually means when it holds up
Three things are true when a competitor is named consistently and you are not, and only one of them is about your product.
- There is more written about them, in the sources the assistant reads, than about you. This is the usual one.
- What is written about them says plainly what they are for. Models quote sentences that are already clear; they do not do the work of making a vague description specific.
- Their name is unambiguous in text. A brand that is also an ordinary word is genuinely harder for a retrieval system to attribute, and that is a real disadvantage rather than an excuse.
What nobody can honestly tell you is why the model chose them. Anyone selling you the ranking factors of a language model is selling a guess in a confident voice. What can be told you is exactly what it said, exactly which pages it cited, and whether that moved since last month.
The fix is almost never on your own website
This is the part that surprises people who arrive from classical SEO. In our own first full run the assistants leaned on forum threads, video descriptions and review sites far more than on any company website, including the websites of the companies being recommended. Your homepage is not where the answer came from.
The sources it cites are your press list. Not the domains with the best ratings — the pages that actually influenced the answer about your category.
So the work is being present, and described accurately, in the places it already reads. Pull the cited sources for the questions you lose, and you have a list of specific pages rather than a content strategy. Some of them will be reachable this month: a category listing you are missing from, a comparison article that omits you, a review site with three reviews on it, a question in a forum nobody from your company has answered.
Two things not to do
Do not publish a "best tools in our category" post that puts you first and expect an assistant to quote it. Models are unimpressed by a vendor ranking itself, and the pattern is common enough to be discounted. Write the piece that is genuinely useful about the decision instead, including where you are the wrong answer, because that is the passage that gets quoted.
And do not try to buy the placement. There is no slot to buy. An assistant composing an answer is not selling inventory, and anybody quoting you a position is describing something they do not control.
How you will know whether it worked
By measuring the same questions, the same way, repeatedly — and by refusing to call a difference a change until it clears the noise. Two readings differ reportably only when their confidence intervals do not overlap, and there is no interval under three runs. A tool that shows you an arrow after one rerun is showing you variance with a direction drawn on it.
The short answers
Why does ChatGPT recommend my competitor instead of me?
Almost always because there is more written about them in the sources the assistant reads, and because what is written says plainly what they are for. It is a retrieval and description problem far more often than a quality judgement, and the model has no opinion about which product is better.
Can I pay to be recommended by an AI assistant?
No. There is no placement to buy. Assistants compose answers from what they can find written about a category, which makes earned coverage, reviews and clear factual descriptions the things that actually move it.
How do I find out which sources the AI used?
Engines with web search on cite the pages they read, and those citations are the actionable half of the reading. Collect them for the questions you lose and you have a specific list of pages to be present on rather than a general content plan. Our own first run collected 460 cited URLs across 190 domains from 15 questions.
Should I write a blog post ranking myself first?
No. A vendor ranking itself is a recognisable pattern and gets discounted. The piece that gets quoted is the genuinely useful one about the decision, including the part where you are the wrong answer, because that passage is the one nobody else is willing to write.
How long does it take to change what an assistant says?
Longer than a website change and on a schedule you do not control, because the assistant has to re-read a source that has to exist first. Treat it in months, measure monthly, and do not read the first rerun as a result.
How do I know a change is real and not random variation?
Ask the same questions the same way at least three times per reading, and compare intervals rather than points. Two snapshots differ reportably only when their confidence intervals do not overlap. A single measurement dressed as precision is how credibility gets spent.
Find out in under ten minutes
One free check, no account and no card. You get the answers the assistants gave, not a score.
Every figure quoted here comes from a real question put to a real assistant, saved with the date and the assistant that answered it. This piece was written to answer the search "ai recommends competitor not me".