The short answer
If an AI assistant never names your business, it is almost always one of these: the AI crawlers cannot read your site, you have no page that answers the question being asked, or nothing outside your own website confirms you exist. Fix those three and the rest is tuning.
The uncomfortable starting point
An AI assistant does not browse the web the way you do when it answers. It does one of two things. Either it recalls what it absorbed during training, which is a snapshot of the web from months or years ago, or it runs a live retrieval, pulls a handful of pages, and writes an answer from those pages.
Both routes have the same requirement. Something readable, on the open web, has to say who you are and what you do. If nothing does, no amount of budget changes the outcome. The model is not ignoring you. It has nothing to work from.
The seven causes, in the order they usually bite
- AI crawlers are blocked. This is the most common and the most invisible. Many hosts and CDNs now block AI bots by default, so a site that looks perfectly healthy in a browser returns a refusal to GPTBot. You did not choose it and nothing warns you.
- You have no page that answers the question. Your site sells. It does not explain. An assistant asked a how or a why question has nothing on your domain to quote.
- Your entity is fuzzy. Your business name, address, phone number and description differ between your website, your directory listings and your social profiles. A retrieval system cannot merge those into one confident business, so it trusts none of them.
- Nothing third party corroborates you. Models weight sources they consider reliable, and your own marketing site is the least reliable source about you that exists. Reviews, directories, roundups and press carry more weight than anything you publish about yourself.
- Your content is written for skimming, not extraction. A hero line, three benefit tiles and a form give a retrieval system nothing to lift. It needs a clear question, followed immediately by a clear answer, in plain sentences.
- You are not in the index the assistant retrieves from. ChatGPT search leans on Bing. If your site is thin or new in Bing's web index, ChatGPT's live retrieval will not surface you even when your content is good.
- You are invisible for the question actually being asked. Buyers do not ask for your product name. They describe a problem. If every page on your site is named after your product, you are optimised for a query nobody types.
How to tell which one is yours
Work through them in order, because the early ones make the later ones pointless. There is no value in writing content for a site that GPTBot cannot fetch.
Start with a crawler fetch. Request your own homepage while identifying as GPTBot and check that the response is a 200 with the real page body in it. A 403, a 503 or a challenge page is the whole answer.
Then search your own site for the question. Take the exact sentence a buyer would ask an assistant and look for a page on your domain that answers it in the first two lines. If you have to scroll or interpret, so does the model.
Then check your own name. Search your business name and read what the open web says. If the first page of results is only your own properties, no third party has confirmed you and the model has no corroboration to lean on.
Finally, ask the assistant directly. Put the buyer's question to ChatGPT and read who it names. Those competitors are your actual benchmark, and their pages will show you the shape of answer that won.
What fixing it looks like
The order matters more than the effort. Crawler access is a configuration change and it is usually the single largest unlock. Answer pages come next, because they are the thing that gets quoted. Entity consistency and third-party mentions compound slowly, so they are worth starting early even though they pay late.
None of this is a trick. The engines are trying to find the clearest, most confirmable answer to a buyer's question. The work is being that answer, in a form a machine can read.