Ghostart Labs

The AI visibility industry has plenty of claims. We'd rather test them.

AI tools are increasingly influencing which businesses people discover and consider.

At the same time, a new industry offers confident advice about what makes businesses more visible to AI.

The problem is that surprisingly little of it has had a proper test.

Ghostart Labs is where we test those claims, including the ones our own product depends on.

Before we collect data, we record what we're testing and how we'll measure it. We publish results whether they support our thinking or not. And wherever we can, we make the underlying data available so others can examine the evidence for themselves.

Current research

What we're testing

Collecting data

The Ghostart SME Visibility Panel — Edition 1

How visible are UK small businesses in AI answers?

We're tracking 240 publicly listed UK small businesses across 9 sectors and 18 cities to understand how often businesses like these appear when customers ask AI who to use.

We're looking at questions including:

  • Which sectors are most and least visible?
  • What gets recommended when a local business doesn't?
  • How much do different AI platforms agree?
  • How much do answers change when the same question is repeated?
  • Does visibility differ between discovery and validation questions?
Study type:
Standing measurement
Collection began:
August 2026
View study details →

Question locked · starting soon

Does substance predict citations? An observational cohort study

Customer content with higher Beige-ometer substance and citability scores is associated with higher observed AI-answer citation rates than lower-scoring content, controlling for engine and topic coverage.

We're testing it because we don't yet know the answer.

Study type:
Observational analysis
Question locked:
10 September 2026
View study details →

Question locked · starting soon

Does the content we suggest actually get cited? An observational study of Ghostart-suggested publishing

Target answers that receive a Ghostart-suggested, published piece transition from absent to cited at a higher rate, and sooner, than target answers tracked for the same business over the same period that received no piece. Secondary: the published artefact itself appears in the engine's cited sources, at a rate varying by channel and engine.

We're testing it because we don't yet know the answer.

Study type:
Observational analysis
View study details →

The AI Visibility Index

Building a picture of how visible small businesses really are.

Most AI visibility research focuses on major brands.

We're interested in something different: the businesses people choose between every day.

Our standing panel contains 240 real UK small businesses across 9 sectors and 18 cities, including accountants, builders, physios, restaurants, shops and agencies.

Each collection cycle asks AI platforms the kinds of questions their customers might genuinely ask and records which businesses appear.

And because AI answers can change between asks, we don't treat a single answer as a ranking.

businesses
240businesses
sectors
9sectors
cities
18cities
every question, repeated
×3every question, repeated

Rates, never ranks.

We repeat questions and measure how often businesses appear.

We won't tell you that a business “ranks #3 in ChatGPT” when asking the same question again could produce a different answer.

Over time, the same panel should let us see something more useful:

not simply what AI said today, but how AI visibility is changing.

How Labs works

Evidence before conclusions.

Lock the question first

We record what we’re testing and how we’ll analyse it before looking at the results, reducing the temptation to change the question to fit the data.

Repeat the measurement

AI answers vary. We use repeated observations and report rates and sample sizes rather than pretending a single answer is a stable ranking.

Publish the nulls

If a study finds no meaningful relationship, or contradicts something we expected, that result still gets published.

Let people check

We publish enough methodology to understand how a study was run and, where we can do so responsibly, provide the underlying data.

The full protocol, including how the panel was built and what we won't publish, is on How Labs works.

What we won't do

No rankings. No cherry-picking. No early teasers.

We won't identify businesses included in the SME panel.

We won't turn unstable AI answers into claims that somebody “ranks #3 in ChatGPT”.

We won't publish findings from a single sample.

And we won't release enticing “early data suggests...” figures while a study is still running.

If the evidence doesn't support what we expected, that's part of the result.

No results yet. Deliberately.

Our first studies are underway.

The full methodology for the SME Visibility Panel is planned for publication around Beyond the Beige in January 2027, before the first results, so it can be examined and challenged independently of what those results eventually say.

Edition 1 will follow once the required collection cycles are complete.

There isn't a publication date to hit. There's an evidence threshold to meet.

Ghostart Labs

We'd rather find out than assume.

Ghostart Labs exists to test some of the assumptions shaping AI visibility, including our own.

See how Ghostart measures AI visibility →