AI visibility isn't about ranking like SEO. It's about getting recommended.

Your website looks fine. Your Google ranking is decent, maybe even good. And you're still invisible where it counts.

Someone types a question into ChatGPT, Google's AI Overviews, or Claude instead of scrolling a page of blue links, and your business doesn't come up. That gap is what AI visibility actually means. It's whether an AI system names your business when a potential customer asks who's good at what you do.

It's about whether you get recommended. Not whether you rank.

The shift

For twenty years, visibility meant ranking. You wrote a page, tuned it for a keyword, and if you did it well enough you showed up on page one. Annoying, but knowable.

AI search doesn't work that way.

When someone asks an AI assistant "who's a good GEO consultant" or "which accountancy firm specialises in creative agencies," the system isn't running a keyword match against ten million pages. It's synthesising an answer from whatever it has read and trusted, then naming a small number of businesses. Usually a handful. Sometimes just one.

I built Ghostart before I fully understood any of this. The original product had nothing to do with search. It existed to fix an older problem: professional writing that sounds like nobody. LinkedIn posts, website copy, case studies smoothed of anything specific until they could belong to any company in the sector.

I called it "beige." I built a scoring tool, the Beige-ometer, to measure how alive or dead a piece of writing was — voice, emotion, rhythm, story, whether a real person seemed to be behind it. Built for humans reading human writing.

It turned out to describe almost exactly what AI models reward too.

What the evidence actually says

Semrush analysed over 300,000 AI-generated answers and found that individual voices, named people writing with a specific point of view, outperformed generic company pages at getting cited. Not marginally. Consistently, and by a wide enough margin that it reads as pattern rather than noise.

The AI systems doing the recommending were pulling from content that sounded like a specific person had written it under real conditions, not from polished corporate copy that could have come from anyone.

That finding didn't send me in a new direction. It confirmed the one I was already in.

I'd built the Beige-ometer to fight generic writing for reasons of craft — flat, safe, committee-approved copy is boring to read and doesn't earn trust. I hadn't built it with AI search in mind at all. But the mechanism an AI model uses to decide who's worth citing and the mechanism a human reader uses to decide whether to trust what they're reading turned out to be the same mechanism: specificity, a real point of view, evidence that someone actually did the thing they're claiming expertise in.

That's the proof underneath everything Ghostart GEO now does. The fix for "we sound generic" and the fix for "AI never mentions us" are the same fix, arrived at from two different starting points, years apart.

An illustrative case: the consultant who couldn't find herself

Here's a worked example to make the mechanism concrete. It's illustrative, not a real client story, but it maps to a pattern I hear from founders constantly.

Picture a fractional CFO who's been in business twelve years, with a genuinely strong reputation built on referrals. She asks ChatGPT: "who's a good fractional CFO for a Series A startup in the UK?"

The answer names three firms. Larger firms, with content teams, publishing volume, and case studies with numbers attached. Her name doesn't come up.

She isn't less qualified. She's less legible to the model.

Her website has a bio page and a services page, both written in the safe, competent tone most consultancy sites default to: "We provide strategic financial guidance to growth-stage businesses." That sentence is true of every fractional CFO in the country. It gives the AI nothing to distinguish her with, so it defaults to whoever has published enough distinctive, specific content that the model can point to and say, with some confidence, this is what this person actually does and how they do it differently.

The fix isn't more content. It's more specific content: a real answer to a real question her clients actually ask her, written in her own voice, with an opinion in it that a competitor firm couldn't have written the same way.

That's the whole discipline. It has nothing to do with keyword density or schema markup. It's a sentence that could only have come from her.

Where most businesses get it wrong first

The instinct, when a founder hears "AI doesn't know about us," is to publish more, faster, often with AI generating the drafts. Understandable. Mostly backwards.

Publishing more generic content doesn't make you more visible to an AI model. It makes you a slightly larger pile of the same undifferentiated material the model already has too much of.

If your published answer to "what makes a good financial advisor" reads exactly like the answer on the top five competitor sites, the model has no reason to single you out, no matter how many pages you add.

The businesses that start showing up in AI answers are usually the ones willing to say something a competitor wouldn't say the same way: a specific opinion, a method with a name, an honest account of a mistake and what it taught them. That's harder than publishing volume. It's also the only lever that actually moves you from being indexed to being named.

Found for what

Most businesses with a website are visible in the sense that a model has crawled them. The real question is what you're visible for.

Being found and being found for something true aren't the same achievement. A business can be all over the training data and still get skipped in the answer, because nothing in what they've published gives the model a reason to name them for anything specific.

The businesses that get recommended are the ones an AI system can describe in one confident sentence. Not because they've mastered a new technical trick, but because they've written down, clearly and specifically, what they're actually good at and why.

That was true of good marketing before AI search existed. It's just no longer optional.

If you want to know what an AI system currently says about your business when someone asks who's good at what you do, ask it. Type the question a potential client would type. See who gets named. If it isn't you, that's the starting point.

You can join the waitlist for Ghostart on the homepage: ghostart.io