Why AI assistants recommend your competitor and not you

AI assistants do not rank businesses. They summarise sources. So when ChatGPT names three suppliers in your category and none of them is you, the useful question is not "why am I ranked lower" — it is "which pages did it read, and why was your competitor on them". Almost always the answer is that the assistant never looked at your website at all.

This is worth sitting with for a second, because it inverts how most people think about the problem. You are not losing a ranking contest. You are absent from the material being summarised.

Once you see it that way, the six reasons below stop being mysterious and start being addressable — some of them quickly.

1. It is reading third-party pages, not your website

Ask an assistant "who are the best suppliers of X" and watch what it cites. It is rarely a vendor's own site. It is a roundup article, a review platform, a community thread, a trade publication, a comparison page written by somebody with no stake in the answer.

That is not a quirk. For a recommendation question, a page that claims "we are the best" is close to worthless as evidence, and the systems behave accordingly. Your competitor is not winning because their homepage is better written. They are winning because they appear on the four pages the assistant actually read.

This is the single largest factor, and it is the one most agencies skip, because fixing it means work off your own website where nothing is under your control.

2. It cannot tell confidently what you are

Language models resolve entities — they try to work out that "the business called X" is a specific, identifiable thing with a category, a location and a set of attributes. That resolution is easy when your name, category description and details are consistent everywhere you appear, and hard when they are not.

If your company is described one way on your site, another on LinkedIn, with a different trading name on directories and a category that shifted two years ago, the model has a fuzzy and contradictory picture of you. A fuzzy entity is easy to omit from a shortlist, because the safe move is to name the businesses it can describe confidently.

Nothing about this is exotic. It is the same reason a customer gets confused, made mechanical.

3. You describe yourself in your language, not the category's

Buyers ask questions using the words of their problem. Businesses describe themselves using the words of their positioning. When those two vocabularies do not overlap, you are invisible to the retrieval step that happens before any answer gets written.

If everyone in your market searches for a thing by its plain name and your site calls it something proprietary, you will not be retrieved for the plain name. You have to be findable in the words the question uses, which sometimes means writing the boring term next to the clever one.

4. Your competitor is on the platforms these systems lean on

Review platforms, marketplaces, industry directories, established publications and active community forums carry disproportionate weight, because they are the places where third parties talk about businesses rather than businesses talking about themselves.

If your competitor has fifty recent reviews on the platform that matters in your category and you have four from three years ago, they will be named more often. This is one of the few areas where the fix is unglamorous and completely within reach.

5. You do not rank for the questions behind the answer

Assistants overwhelmingly cite pages that already perform in conventional search. Retrieval leans on the same signals of quality and authority that search engines use, so a site that ranks poorly for its category's questions tends to be absent from the answers built on top of them.

Google is explicit that its AI features work this way. Its guidance for site owners states there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" — the same fundamentals apply. Which means if your conventional search foundations are weak, that is the first problem to solve, not a separate one to solve later.

6. Your pages answer "what we do", not "who this is right for"

Most business websites describe services. Very few state plainly who they are the right choice for, who they are wrong for, what the work costs, how long it takes, and what the trade-offs are.

Those are exactly the questions a buyer asks an assistant, and a page that answers them directly is far more quotable than a page of capability statements. Google's own guidance on helpful, people-first content points the same way: content written to answer a person's question beats content written to occupy a keyword.

Being willing to say "we are not right for X" is, in practice, one of the strongest things you can publish. It is specific, it is unusual, and it gives a summariser something worth quoting.

What does not fix this

A short list, because there is a lot of noise in this area.

Special files and AI-specific markup are not the answer — Google states directly that you do not need to "create new machine readable files, AI text files, or markup" and that there is "no special schema.org structured data" required for its AI features. We cover what is and is not genuinely different about AI search in more detail, including where that guidance stops applying, because Google is not the only assistant that matters.

Blocking or unblocking crawlers is a prerequisite rather than a strategy. If the retrieval crawlers — OpenAI's OAI-SearchBot, Perplexity's PerplexityBot and their equivalents — cannot reach your pages, nothing else you do will register. But allowing them does not earn a citation either. It removes a reason you cannot have one. Which crawler is which, and which ones you can block without losing citations, is worth getting right before anything else here.

One honest caveat

The same question asked twice can return different names. Answers vary between users, sessions and weeks, and every one of these systems changes without notice. Anyone promising you a guaranteed mention is describing something they cannot control.

What can be measured is share: take twenty questions your buyers actually ask, run them on a schedule, and record how often you are named versus each competitor. A number that moves over a quarter is real. A single lucky answer is not.

Where to start

If you only do three things, do these.

1. Run the questions yourself. Ask five assistants the ten questions a buyer would ask before choosing someone in your category. Write down who gets named and which sources the answer cites. This costs an afternoon and tells you more than any tool.
2. Fix your entity. One name, one category description, one set of details, everywhere you appear. This is tedious and it is the cheapest structural improvement available.
3. Get onto the sources being cited. Look at what the assistants actually read in step one, and work out what it would take to be accurately represented there.

If you would rather have someone do the first step properly and tell you what it found, tell us your category and your competitors and we will run it — including when the answer is that you are already being named and the problem is somewhere else entirely.