What is AI visibility?

How often an AI assistant names your brand when a buyer asks it a question in your category. It is measured as a rate: the share of answers in which you were named, on a fixed set of real buyer questions, per engine, in a stated period.

Why is AI visibility two numbers rather than one?

Some buyer questions already contain your brand name; most do not. Brands are named far more often on the first kind. Across 27,260 measured answers, brands were named in 77% of answers to questions that carried their name and 28% of answers to questions that did not. A single blended score lands wherever the mix of questions puts it, so the two rates must be reported separately.

Can you pay to appear in AI answers?

Not in the answer itself. Advertising is arriving alongside AI answers as a separate inventory. The recommendation is assembled from sources the engine reads and trusts, which is why presence in those sources, not payment, is what moves it.

Five-minute guide

How AI visibility actually works

Q: Why does my AI visibility number change when I did nothing?

The model behind the answer changes without notice. When 245 identical questions were asked before and after one engine changed its model, 26 answers gained a brand and 16 lost one, with nothing changed on the brands' side. A number without its engine, its model version and its date cannot be compared to anything.

Your buyers now ask an AI before they ask you. This is what happens inside that answer, in plain English, and what you can and can't do about it. Read it before the report.

Roger Graham, 28 Labs · September 2026 · Every number below comes from 27,260 real AI answers, measured April to August 2026, with its base stated.

The short version. A buyer asks ChatGPT, Claude, Gemini or Perplexity who to use. The engine reads a set of sources, picks a few names, and hands them over with reasons. There is no page two. You're in that answer or you're not, and nobody sees the ones you're missing. AI visibility is how often you're in it, measured as a rate.

1What happens when a buyer asks

A buyer types a question the way they'd ask a colleague: "best organic mattress in Dubai", "which agency for a product launch", "is X better than Y for a family of four". The engine doesn't return a ranked list of links. It returns an answer with a handful of names in it, and the buyer treats those names as the shortlist.

Buyer asks a question

Engine reads sources it trusts

Answer names a few brands

You're in it, or you're not

Search gave you a list to sit somewhere on. An answer makes a choice. That is the whole difference.

2Where the names come from

The engine doesn't know your brand the way a person does. At the moment of the question it fetches pages, reads them, and builds the answer from what it read. What it reads most isn't brand websites. In the answers we measured, the most-cited sources were communities, video, social platforms and review sites: YouTube, Reddit, Facebook, Instagram, LinkedIn, Trustpilot.

Two things follow. Your own site is one input among many, not the whole game. And you can't buy the placement: advertising is arriving next to AI answers, but the recommendation itself is assembled from sources, not sold.

Base: 22,111 answers that carried citations, 84,809 cited pages across 29,157 sites, April to August 2026.

3Why "visibility" is two numbers, not one

Some buyer questions already have your name in them ("is [your brand] worth the price"). Most don't ("best organic mattress in Dubai"). Brands get named far more often on the first kind. The first number says people who already know you can find you. The second says strangers can. Only the second brings you anyone new.

77%

of answers named the brand when the question already had its name in it

28%

of answers named the brand when the question didn't

Base: 5,920 and 18,080 answers respectively, four engines, April to August 2026. Corpus averages across 126 measured brands, not any one company's result.

Blend the two into one "AI visibility score" and the score is decided by the questions, not the brand. Add more questions with your name in them and it goes up. Nothing about you changed. So any number quoted to you should come with its question type attached, or it can't be compared to anything, including itself next month.

4Why one engine isn't the answer

ChatGPT, Claude, Gemini and Perplexity read different sources and reach different verdicts. Ask all four the same question and, on the questions where at least one of them named the brand, 57% of the time they didn't all agree. Which engine your buyer happens to use decides whether you're in their shortlist. A check on one engine is one engine's opinion.

Base: 3,084 questions where at least one engine named the brand, April to August 2026.

5What a normal number looks like

Measure a brand on 40 or more real buyer questions and, more often than not, it's named in fewer than one answer in four. In over a quarter of our 204 measurements the brand was named in fewer than one answer in ten. That's not a forecast for your brand. Your number is unknown until it's measured. It's the base rate, and the base rate is low.

Base: 204 measurements of 40 or more answers each, four engines, April to August 2026.

6Why the number moves on its own

The model behind each engine changes without notice. When we asked 245 identical questions before and after one engine changed its model, 26 answers gained a brand and 16 lost one. Nothing had changed on the brands' side. A number from one day is a snapshot. Without the engine, the model version and the date next to it, it can't be re-checked.

Base: 245 matched questions across 3 brands on one engine's model change. Counts only, not a rate.

7What you can and can't do about it

This is not a schema fix or an FAQ page. It's a standing discipline, and it has three parts.

Work on the questions without your name in them. That's where buyers find someone. The branded questions look after themselves.
Be present where the engines read. Reviews, communities, listings, third-party coverage, video. Your site matters, but it's not where the answer gets assembled.
Measure all four engines on a fixed set of real buyer questions, and measure again after every model change. That's the only way a number means anything next quarter.

It moves slower than paid and faster than brand. Source work compounds over months, and a model update can move a number either way without warning. There is no universal "good score"; there is your category's rate, your market's rate, and your own, measured the same way each time.

8What nobody can see

All of this measures what engines fetch and say today. What sits inside a model's training data is not visible to anyone outside the labs, and any vendor claiming to measure or change it directly is overpromising. Every figure above re-derives to a stored answer, an engine and a date. If a number can't do that, it isn't measurement.

Now read the reportFourteen sections, every number next to the count it came from, and a section on what the data can't tell you. Free, no email gate.

Read The AI Visibility Report Check your website free

28 Labs measures how AI engines answer real buyer questions - and what moves the counts. Every number we publish re-derives to specific questions and engines. The AI channel, measured.