AI visibility, or GEO, is maybe two years old as a discipline. It's an industry running on confidence and hot air. It's grown a full set of tactics and playbooks, some of them sold with guarantees that don't exist.

Very little of it has been tested against anything.

The claims nobody has checked

An analysis from Citedar found this: _nobody can guarantee that ChatGPT or Google's AI Overviews will cite or recommend a business_.

A tactic that works in one competitive setting can do nothing in another.

Then there's the volume question. A lot of advice in this space still amounts to 'publish more'.

Digitalapplied's research into AI Overview citations, published in April 2026, found something more specific: structure, extractability, named sources and topical relevance predicted citation better than sheer output did.

Answer-first formatting. Headings. Lists a machine can lift a clean sentence from.

The advice buried in that data is more useful than 'publish more': shape what you publish so there's _something to quote_.

I built Ghostart on a conviction that predates any of this: writing that sounds like a real person, with a real point of view, outperforms writing that doesn't.

Watching the AI visibility data start to agree with that, rather than contradict it, is satisfying. But satisfying isn't proven, and I'm wary of anyone, including me, who treats a hunch as settled just because it's convenient.

> We're not guessing our way through this. We're letting the facts lead the product, even when the facts are inconvenient.

That's the reasoning behind Ghostart Labs.

What Ghostart Labs does

Labs is our research arm, and the rules are simple by design.

Before we collect a single data point, we publish the question we're asking and exactly how we'll answer it. The numbers can't rewrite the question after the fact. Every result gets published, including the ones that prove us wrong.

The raw data ships with every study, so anyone can check our working rather than take our word for it.

We promise to continually iterate the Ghostart AI Visibility platform in line with the best practice we're uncovering.

The first study: UK SMEs and AI visibility

Nobody, anywhere, has measured whether small businesses know that AI tools are already answering their customers' questions without them.

Our first study asks that directly: a survey of UK businesses from one to 250 employees, on how their customers actually find them, whether they've ever checked what AI tools say about them, and what they'd do about it if the answer's wrong.

We'll publish here on Medium as well as on the Ghostart blog, with the full method, the questionnaire and the raw responses attached, whatever it finds.

The AI Visibility Index

Alongside the one-off studies like above, we're running something with a longer horizon.

We've assembled a panel of 240 real UK small businesses, accountants, builders, physios, restaurants, shops and agencies, across nine sectors and eighteen cities.

Every business was found in a public place, and we recorded the source.

Each month our systems put the questions those businesses' own customers ask, "who are the best accountants in Leeds?", to the AI tools people actually use, and record who gets mentioned. We ask each question three times, because AI answers move between asks.

The analysis is preregistered. We wrote down which cuts of the data will publish and which won't before anyone looked at a single result, the same discipline medical trials use against cherry-picking. We report rates, never ranks. You won't see us claim a business "ranks third in ChatGPT," because that number can change on the next ask.

We've also named our own bias up front. Panel businesses came from public directories, the same places AI engines read, so our numbers will likely land rosier than the true national picture. The Index describes publicly listed UK small businesses. Not all of them.

Collection started in August 2026. The full method publishes at the next Beyond the Beige summit on 30 September, before a single result, so it can be challenged while it can still be improved. The first edition follows in October, built on two complete monthly cycles.

What this means for Ghostart

Ghostart's product will keep changing shape as Labs tells us what's real, not what's convenient.

That's the deal we've made with ourselves, and now, in public, with you.

If you want to watch where this goes, follow Ghostart on Linkedin and here on Medium.

The first survey and the Index's first edition are coming, and we're publishing all of it, including the parts that don't flatter the AI Visibility / GEO industry or its current methods.

We intend to be an absolute authority in the AI visibility world, so the labs and panel are fundamental for our business. We aren't guessing, we're letting the facts lead us and our product development.