Why does ChatGPT recommend my competitors instead of me?

Usually because they have been described in many places over years and you have not, and because their pages match how people phrase the problem while yours describe a category. The first part accumulates slowly. The second is fixable by writing a page for each situation your product handles, in the words a person would use.

How does ChatGPT choose which tools to name?

Two things combine: what the model already holds about the category from training, and what it retrieves when searching to answer the specific question. New products cannot influence the first quickly but can compete immediately on the second, which is driven by being reachable, matching the described problem, and being described by third parties.

Does my own website matter for AI recommendations?

Yes, as the place that answers the specific question, but third-party descriptions carry more weight for the recommendation itself. A page where somebody who does not own the product describes it is better evidence than a homepage, which is expected to be favourable.

Should I say what my product is bad at?

Yes, and it works better than it sounds. A page that honestly states this is good for one case and poor for another is a more useful source for an assistant trying to match a person to a tool, and the resulting recommendation reaches people the product actually suits.

Can I pay to be recommended by AI assistants?

No, and services offering it are selling either ordinary content work at a markup or something that will eventually embarrass you. The route that works is being genuinely good at a specific thing, describing it clearly, and being described accurately in places you do not own.

What is the single highest-leverage change?

Writing one page per situation your product handles, naming the situation in the words a person would use to describe it, and answering it directly near the top. People describe situations to assistants rather than asking for categories, and that is what the matching works on.

How an assistant decides which tool to name

Somebody types a question that your product answers perfectly, and the assistant names three competitors. Understanding why is more useful than being annoyed by it, because most of the reason is fixable.

Run the free check

An assistant answering a recommendation question is composing from sources it can retrieve and from what it already holds about the category. Being named depends on being reachable by the search crawler, being described in words that match how people phrase the problem, being described somewhere other than your own site, and having a page that answers the specific question rather than describing the product in general.

Two different things are happening

It helps to separate them, because they respond to different work.

The first is what the model already holds about a category from having been trained on a large amount of text. This is why well-established products get named constantly: they have been written about for years, in many places, by many people. You cannot influence this quickly, and pretending otherwise would be dishonest.

The second is retrieval. When the assistant searches to answer the question in front of it, it finds current pages and composes from them, citing what it used. This part is influenced by ordinary work on your own pages, and it is where a new product can compete immediately.

So the strategy for a new product is to be excellent at the second while accepting you will lose the first for a while.

Being reachable comes before everything

If the search crawler cannot fetch your pages, none of the rest of this page applies to you.

OpenAI documents OAI-SearchBot as the agent used to surface websites in ChatGPT's search features, which is separate from the training crawler. If your robots.txt blocks it, you are not a candidate for retrieval at all, regardless of how good your content is.

The same applies to rendering. If your pages are assembled entirely by script after loading, and nothing meaningful is in the HTML that gets fetched, then a crawler that does not execute your JavaScript receives very little to work with.

Match the problem, not the category

This is the biggest controllable factor and the one most builders get backwards.

People do not ask assistants for a category. They describe a situation. Somebody does not ask for a project management tool, they ask what to use when three freelancers keep losing track of who is doing what. The assistant is matching that described situation against descriptions of tools.

If your pages describe your product in abstract or aspirational terms, there is very little for that matching to catch on. If your pages describe specific situations, specific problems, and what happens in each one, there is a great deal.

The practical version: write a page for each situation your product handles, name the situation in the words a person would use, and answer it directly. That is what makes you findable for the question rather than for the category.

Other people describing you counts more than you describing you

An assistant composing a recommendation draws on aggregated and third-party sources heavily: comparison pages, directories, forum threads, review sites, listings.

There is a straightforward reason. A page where somebody who does not own the product describes it is better evidence about the product than the product's own homepage, which is expected to be favourable.

This is why the listing work described elsewhere in this cluster matters more than its direct traffic suggests. Twelve directory listings, each with a genuinely different description, is twelve pieces of third-party evidence about what your product does. A forum thread where somebody recommends you in context is worth more still.

It is also why writing each listing fresh matters. Twelve copies of one paragraph is closer to one piece of evidence than twelve.

Answer the question the way an answer is shaped

Given two pages that both cover a topic, the one that gets used is generally the one from which a clean, self-contained answer can be lifted.

That means the answer near the top rather than after several paragraphs of preamble. It means paragraphs that make sense when quoted alone, without depending on the sentence before them. It means concrete specifics, since a passage full of adjectives contains nothing to quote.

It also means saying what your product does not do. Counter-intuitive, and it works, because a recommendation naming the limits is more useful and reads as more trustworthy. A page that honestly says this is good for X and poor for Y is a better source for an assistant trying to match a person to a tool.

What you cannot control, and should not fake

A short honest list, because effort spent here is wasted or worse.

You cannot quickly change what the model already holds about your category from training. That accumulates over years of the internet discussing you.

You cannot make an assistant recommend you for something you are not good at, and attempting it produces a bad recommendation, a disappointed user, and eventually worse evidence about your product than you started with.

You cannot buy your way in, and the services that offer to are selling either ordinary content work at a markup or something that will embarrass you.

The honest route is the slow one: be genuinely good at a specific thing, describe it clearly and specifically, get described accurately elsewhere, and let that accumulate.

Part of a larger guide

This page is one part of The AI channel. The other parts:

Questions people ask

Why does ChatGPT recommend my competitors instead of me?: Usually because they have been described in many places over years and you have not, and because their pages match how people phrase the problem while yours describe a category. The first part accumulates slowly. The second is fixable by writing a page for each situation your product handles, in the words a person would use.
How does ChatGPT choose which tools to name?: Two things combine: what the model already holds about the category from training, and what it retrieves when searching to answer the specific question. New products cannot influence the first quickly but can compete immediately on the second, which is driven by being reachable, matching the described problem, and being described by third parties.
Does my own website matter for AI recommendations?: Yes, as the place that answers the specific question, but third-party descriptions carry more weight for the recommendation itself. A page where somebody who does not own the product describes it is better evidence than a homepage, which is expected to be favourable.
Should I say what my product is bad at?: Yes, and it works better than it sounds. A page that honestly states this is good for one case and poor for another is a more useful source for an assistant trying to match a person to a tool, and the resulting recommendation reaches people the product actually suits.
Can I pay to be recommended by AI assistants?: No, and services offering it are selling either ordinary content work at a markup or something that will eventually embarrass you. The route that works is being genuinely good at a specific thing, describing it clearly, and being described accurately in places you do not own.
What is the single highest-leverage change?: Writing one page per situation your product handles, naming the situation in the words a person would use to describe it, and answering it directly near the top. People describe situations to assistants rather than asking for categories, and that is what the matching works on.

See how your product is currently described

The free Get Found Check looks at what your pages say about you and whether an assistant has anything specific to work with.

Run the free check Have Licheo do it for you