What AI Visibility Is, and What Actually Moves It

This morning I looked at a search term one of my own pages ranks eleventh for. Google's AI Overview answered the question above the results and named four products. Mine wasn't one of them.

That's AI visibility. Not a score, not a ranking — whether the machine says your name when someone asks it a question you ought to be the answer to.

AI visibility is how often, and how favourably, an AI assistant names your brand when someone asks it a buying question.

That's it. A client asks ChatGPT "what's the best tool for a small agency" and gets four names back. You're in that list or you're not. There's no position two. There's no "we came up on page one." The answer is a paragraph, and either you're in the paragraph or you don't exist for that question.

And increasingly it isn't ChatGPT at all — it's the Google search bar. Somebody types that same question into ordinary Google, and Google answers it above the results. The AI Overview pushes every organic listing down the page, hands the searcher a finished answer with a few products named in it, and most people stop reading there.

You can rank well and still lose, because the answer arrived before your listing did and your name wasn't in it. That's the part that catches agencies out: nothing in their rank tracker moved. The page is still where it was. An LLM simply served the answer first, and served it without them.

Which is why the old measurements don't transfer. A rank tells you where you sit in a list of ten blue links. An AI answer isn't a list of ten anything.

Google's own position is that optimising for AI is still SEO, and they're not wrong about the plumbing — the assistant has to be able to crawl and read your page like anything else.

But when I measured it, the plumbing turned out to be the least interesting part.

I ran a field study across 21,213 citations for ten national brands, looking at what the assistants actually read before answering. The reading list was Reddit, YouTube and Amazon. Not the brands' own websites.

That's the part that changes the job. You can have a technically flawless site — I do, it grades an A — and be invisible in AI answers, because the assistant is forming its opinion of you somewhere else. Somewhere you don't own.

It isn't traffic. An AI answer often means the person never clicks anything. You can be highly visible and see nothing in analytics. Measure this by visits and you'll conclude it isn't working when it is.

It isn't sentiment. Being mentioned isn't being recommended. "X is cheaper but limited" is a mention. Any honest measurement separates named from named well.

It isn't one number. Visibility is per-question. You might be the obvious answer to one buying question and absent from the next, and those two facts have completely different causes.

Not by asking ChatGPT about yourself once. That tells you almost nothing — the answer varies by session, by phrasing, by which assistant, and by the day.

The method that survives contact with a client:

1. Write the questions your buyer actually asks. Nobody types "is StyleForge good." The real ones are category questions: best tool for X, X vs Y, how do I do Z.
2. Lock the list. The same questions, every time. A changing question set makes every comparison meaningless.
3. Ask across several assistants. They read different things and disagree constantly.
4. Record who else got named. The competitor list in an answer is the most useful thing in it — those are the brands the machine currently considers the category.
5. Keep the sources. Which pages did it cite? That's your work order, and usually it isn't your site.
6. Re-run monthly, unchanged. The delta is the evidence. A single reading is a photograph, not a measurement.

You can do all of that by hand with a spreadsheet and an afternoon, and that's genuinely how I'd start — you learn more in the first hour of reading real answers than from any dashboard. It stops scaling around the second client, which is when it's worth automating. StyleForge runs that loop as AI Visibility Pulse, asking a locked question pack across ChatGPT, Claude, Gemini, Perplexity and Google's AI Overview and tracking what moves.

Here's where I'll argue against my own sale.

Most of what shifts AI visibility isn't on your website, and a lot of it isn't work an agency can bill for. If the assistants are reading Reddit, YouTube and Amazon, then being present and well-regarded in those places matters more than another page on your domain.

Which means the honest answer to "how do we improve this" is often earn a reputation somewhere you don't control — slower and less comfortable than a technical audit.

What does help, and is in your hands:

Answer the question on the page, in the words it was asked. Assistants lift question-and-answer pairs. A page that states the question and answers it plainly is far more liftable than the same information buried in a narrative.

Be specific enough to be worth citing. Assistants cite what has a fact in it. A page saying "we're the leading platform" gives nothing to quote. A page saying what a thing costs, what it does and what it doesn't, gives plenty.

Say what you don't do. The answers I've seen that name a brand favourably almost always describe its limits too. A page that admits a boundary reads as a source; a page that claims everything reads as an advert.

Fix the plumbing, then stop. Let the assistants crawl you, mark up your FAQ, keep your titles readable. That's a day's work — necessary but not sufficient, which is exactly why so many agencies do it and see nothing.

You cannot make an assistant recommend you. There's no submission form, no ranking factor to buy, and anyone selling guaranteed AI visibility is selling something that doesn't exist.

You also can't fix a bad reputation with an audit. If the discussion about a brand is negative in the places the machines read, the answer will reflect that — correctly.

Worth saying to a client before you start, not after. The measurable promise is: we'll find out what the assistants say about you, why they say it, what part of that is fixable, and we'll show you the same measurement again next month. That's a real service. "We'll get you into ChatGPT" isn't.

AI visibility is whether the machine says your name when it matters — and the work is less about your website than most people selling it will admit.

Which is the real shift, and it is bigger than a new acronym. For twenty years the discipline pointed inward — your site, your pages, your technical health, your content. That was where discovery happened, so that was where the work went.

The answer engines moved the centre of gravity off your domain. They build an answer out of wherever the conversation already is: a thread, a review, a video, somebody else's roundup. Your own pages are one input among many, and frequently not the loudest one.

So exposure on the broad web matters more now than it ever did when search was the front door. A brand has to know where it is being talked about, not only what its own site says — which forums, which videos, which retailers, which comparison posts, and whether what sits there is accurate or three years out of date. That is closer to reputation work than to a technical audit, and it is the part that decides who gets named.

• Why your brand isn't showing up in AI answers — the diagnostic, cheapest checks first.
• GEO vs SEO: what's actually different — the field study in full.
• How to make money with AI visibility — what to sell, who pays, what to charge.