For twenty-five years, being found meant one thing: appear high in a list of links, and let the buyer do the rest. The list was the product. Your job was to get into it.
That list is quietly being replaced by a paragraph. Someone asks ChatGPT which companies they should be looking at, and gets four names and a sentence about each. They ask Google, and the answer sits above the links — often good enough that the links go unclicked. Nobody scrolled. Nobody chose from ten. Three companies were named, and the rest of the market was not in the room.
The uncomfortable part
You can rank first for your category and go unmentioned in every one of those answers.
This is not a hypothetical. The two things are read by different systems for different reasons. A search engine ranks a page. A language model recalls an entity — a company it learned about because enough other people wrote about it, in enough places, clearly enough that it could tell who you were and what you sold.
Your own website is evidence of the first and barely evidence of the second. Everyone's website says they are the leading provider. A model that believed website copy would recommend every company equally, so it doesn't; it leans on what the rest of the internet says about you, which is a much harder thing to manufacture and a much better signal because of it.
The short version
Ranking is about pages. Being named is about your company. The second is earned mostly off your own website, which is why it is harder to fake and worth more when you have it.
Two different questions, two different answers
It helps to split "AI visibility" into the two things it actually is, because a brand can pass one and fail the other, and the fixes point in opposite directions.
From memory. Ask ChatGPT, Claude or Gemini a question with web search off and the answer comes from training — from what the industry was saying about you long before anyone asked. This moves slowly. It is close to impossible to fake, and being absent from it usually means one thing: outside your own site, almost nobody writes about you by name.
From retrieval. Ask Perplexity, or read Google's AI Overview, and the machine is reading the live web while it answers. This moves fast. Being absent here is usually mechanical rather than reputational — pages that cannot be crawled, or that carry nothing quotable when they are.
Being strong in memory and absent from retrieval often means a single line in a robots.txt file is undoing years of reputation. Being present in retrieval but absent from memory usually means your content is fine and nobody outside your company has ever cited it. Same symptom, opposite work.
What actually moves it
In rough order of leverage, and none of it is a trick:
Let the crawlers in. GPTBot, ClaudeBot, PerplexityBot, Google-Extended. Plenty of sites blocked these in 2023 on principle and never revisited the decision. If yours is one of them, nothing else on this list matters until it changes — you have opted out of the thing you are trying to win.
Get named by other people. The category roundups, the comparison pages, the industry directories, the interviews. This is the heaviest signal and the slowest to move, and there is no shortcut, which is exactly why it works.
Make it easy to know who you are. Organisation schema, a consistent name and description across your site, LinkedIn, Wikidata and the places your industry lists companies. A model that cannot resolve which company you are will not risk recommending you.
Put real people on the About page. Named, credentialed humans. An anonymous company is difficult to vouch for.
Answer the questions buyers actually type, in structured, quotable form. A marked-up answer can be lifted whole. A paragraph gets paraphrased away, or skipped.
What nobody can tell you yet
Two honest caveats, because this field has more confident claims in it than it has evidence.
Model answers drift. Ask the same question a fortnight apart and you will get a different set of companies, without anybody having changed anything. That means a single reading is a snapshot, not a verdict — and it means that anyone quoting you one number as your AI visibility score is quoting you the weather on the day they looked.
And the answers are personalised, increasingly, by who is asking and where from. A measurement taken in English, from a server in the United States, is a measurement of that. If you sell in Bogotá or São Paulo, it is a measurement of somebody else's market with your name on it.
Neither of these makes the exercise pointless. They make the trend the thing worth watching rather than the reading — which is a familiar idea to anyone who has done this work before, just applied to a new surface.
Where to start
Find out whether you are named at all. Not whether you rank — whether a model, asked the question your buyer would actually type, puts you in the answer. It takes about a minute and it is the only part of this you cannot reason your way to from the outside.
