How AI visibility is measured

Four assistants, measured separately

People meet AI answers in four main places: ChatGPT, Gemini, Google’s AI Overview at the top of an ordinary search, and Google’s AI Mode. We measure each one on its own.

They regularly disagree about the same business. One may name you in most of its answers while another barely mentions you. An average across all four would hide exactly the difference you need to act on, so we never blend them into a single score.

Real questions, not keywords

People ask assistants whole questions: “I want to get cosmetic dentistry in Leeds but I’m nervous, where should I go?” rather than “dentist Leeds”. The way a question is put changes which businesses get named, so we write questions the way people actually ask them.

The questions cover the different things people want to know: what something costs, who to choose, how the options compare, whether it suits them, whether it will be uncomfortable, how long it takes and whether they can spread the cost. A business can be visible for one kind of question and invisible for another.

Asked more than once

The assistants do not give the same answer twice. Behind the scenes, an assistant often turns one question into several searches of its own, and the results shift from one ask to the next. So we ask every question five times on every assistant, and report how often you were named rather than whether you were.

What we count

Held

Questions where an assistant named you every time it was asked.

Winnable

Questions where it named you some of the time. The assistant already knows you, which makes these the place to start.

Lost

Questions where it recommended a competitor and not you.

Share of recommendations

Of every recommendation the assistants made across the audit, the proportion that were for you.

Who is named instead

Every business the assistants put forward, ranked the same way for each of them.

Pages quoted

The web pages the assistants cited as sources when answering, yours and everyone else’s.

Counted, never characterised

“Named in 41 of 50 answers” is a measurement. “Better at marketing than you” is an opinion. Every figure in a Visility report is a count, shown with what it is out of, and every question is printed in full, so nothing rests on our say-so.

What the numbers cannot tell you

AI assistants change. Their answers move from week to week, sometimes with no visible cause, and nobody outside the companies that run them controls what they say. A measurement tells you where you stand today and how that changes over time. It does not promise where you will stand next month.

That is why we never promise a ranking, a percentage or a timeframe. Anyone who does is guessing.

Check it yourself

Every question in the report is printed in full. Ask any of them in your own browser and you will see the kind of answer we counted. It may not match word for word, because the assistants vary. That variation is the reason we ask each question several times.