Open ChatGPT and ask it to recommend a company in your industry. If your name is not in the answer, you are not losing on price or on quality. You are not in the room at all.
That is what changed. For twenty years, being found meant ranking on a page of ten blue links. A growing share of buyers never sees that page: they ask an assistant, read one answer, and act on it. The shortlist is written before you get a chance to compete.
What AI visibility means
AI visibility is how often an assistant mentions you, and in what terms, when someone asks a question your business could answer. It is not a ranking position. There is no "first place" in a chat answer — there is being named, or not being named.
It is measured differently too. A keyword has a position; a brand has a citation rate: out of a hundred relevant questions, how many answers include you.
How does an assistant decide who to cite?
Behind a short answer there are almost always three steps:
- Retrieval. The question becomes one or more searches, and the system pulls a few dozen pages that look relevant.
- Reading. The model reads those pages and extracts the passages that directly answer the question.
- Composition. It writes one answer and attaches citations to the sources whose passages it actually used.
Two consequences follow, and they explain most of what works:
- You have to be retrievable before you can be cited. If your page never enters the candidate set, nothing else about it matters.
- You have to be quotable. A page that answers the question in a clean paragraph gets used. A page that buries the answer inside a sales pitch gets skipped, even when it is in the set.
The four doors your name can come through
| Door | How it works | How fast you can influence it |
|---|---|---|
| Training data | What the model absorbed before release | Slow — months to years |
| Live web search | The assistant searches while it answers | Fast — days to weeks |
| Third-party mentions | Lists, reviews, directories and press the model finds | Medium — weeks |
| Structured data | Schema markup the crawler reads literally | Fast — days |
Most advice online is about the first door, which is the one you can do least about. The return is in the middle two.
Why is my company invisible to AI?
In practice it is almost never one dramatic problem. It is a pile of ordinary ones:
- Nobody links to you. Retrieval leans on the same signals search has always used. A site nobody references is a site the system has little reason to pull.
- Your pages answer nothing. A homepage that says "innovative solutions for your business" cannot be quoted, because it contains no fact.
- You exist only where the model cannot read. A business whose entire presence is a social profile is largely invisible to a text retrieval system.
- Your name is ambiguous. If the brand is also a common word, the model cannot separate your mentions from the noise.
What actually moves the needle
In rough order of return on effort:
- Be mentioned somewhere other than your own site. A model that sees your name on twenty independent sites treats it as a real entity. A model that only sees your own homepage has one source — and one source is an advertisement.
- Answer the question in the first paragraph. Forty to sixty words, no preamble. That paragraph is what gets quoted.
- Use headings that are questions. Retrieval matches question to question.
- Publish facts, not adjectives. Prices, numbers, dates, limits. "Plans start at $29 a month" is quotable; "affordable plans" is not.
- Mark up what you are. Organization and FAQPage schema tell the crawler literally what the page is, instead of making it infer.
- Say the uncomfortable part. Pages that mention limits and trade-offs get cited more than pages that only sell, because they read as information rather than as marketing.
How do I check whether AI mentions my brand?
Write twenty questions a customer would actually ask — not your brand name, which proves nothing. Ask each one in ChatGPT, Gemini and Perplexity, in a fresh session with no memory of you. Count how many answers name you, and write down who gets named instead.
Repeat it monthly. The number is crude, but it is real, and it moves. Searching for your own brand name and finding yourself is not a test: the model is reading back what you just told it.
Where do backlinks fit?
Links did not stop mattering when the answer box arrived. They changed job. They used to be a vote that moved you up a list. Now they are the evidence that you exist at all, and the path the retrieval step follows to reach you.
A page nobody links to is a page the system has to be told about. A page referenced from dozens of other sites is one it finds on its own, repeatedly, from different directions — and repetition across independent sources is exactly what an assistant treats as confidence.
That is the whole job: be referenced in enough places to be easy to find, and write pages that are easy to quote once found. Everything else in this category is detail.
