“Are we showing up in ChatGPT?” is usually answered with a screenshot. Someone asks the assistant a question, sees the brand named, and the question is considered closed. Ask again tomorrow, or ask a different engine, and the answer may be different.
AI visibility is the attempt to replace that screenshot with a number that means something. This post defines the term, explains how it is measured, and covers the mistakes that make the number say more than it should.
What is AI visibility?
AI visibility is how often and how favorably AI answers name your brand for the questions your buyers ask. It describes your presence inside the answer itself, not in a list of links beside it.
Each part of that definition carries weight:
- How often. Visibility is a rate across many questions and many runs. One good answer is an anecdote.
- How favorably. An answer that names you as the option to avoid still names you. A mention count alone cannot tell those apart.
- The questions buyers ask. Visibility only means something against a defined set of real buyer questions. Change the questions and you change the number, without anything changing in the market.
Visibility is a rate, not a result
AI visibility is measured as the share of answers that name your brand, across a fixed prompt set, for each engine. The unit is the answer, and the denominator is every answer to every tracked prompt on that engine.
A measurement that holds up follows the same process every time:
- Write a fixed set of prompts in the words a buyer would use. The guide to which prompts to track covers how to choose and size them.
- Ask each prompt on each engine from a clean session, with no logged-in history or personalization.
- Repeat on a regular schedule, because answers vary between runs.
- Record whether each answer names you, and save the full text.
- Report the share of answers naming you for each engine separately, then look at the combined view.
The schedule is what separates measurement from sampling. Engines do not give the same answer twice, so the only stable signal is the pattern across many runs.
How is AI visibility different from search rankings?
A search ranking is your position in a list of links. AI visibility is whether you are named in a written answer the buyer may read instead of clicking anything.
There is no page two
An answer names a short list of brands. If you are not on it, there is no lower position to fall back to and no scroll that might reach you. The outcome is closer to in or out than to a rank.
The engine writes your description
A search result shows your own title and snippet. An AI answer describes you in its own words, assembled from whatever sources it relied on. Two brands can both be named while one is recommended and the other is qualified.
The same question gets different answers
Rankings move slowly. AI answers can change from one run to the next, which is why a single check proves little and why visibility has to be tracked over time.
Buyers ask in sentences
Search queries are short. Buyer prompts include budgets, team sizes and integrations, and those details change which brands get named. Strong rankings for a short keyword do not carry over automatically to the long question.
The metrics that sit next to visibility
Visibility tells you whether you are named. Four related metrics tell you how you are named, against whom, and why.
Share of voice
Your share of all brand mentions in the answers, competitors included. Visibility can hold steady while share of voice falls, if competitors are being named more often alongside you. The full definition is in what is AI share of voice.
Position
Where in the answer you appear. The first brand recommended and the last brand in a closing list are both mentions, and readers do not treat them the same way.
Sentiment
How favorably the answer describes you. Sentiment should be scored only on answers that name you, so that being absent is never mistaken for being described neutrally.
Sources
The pages the engine cites. Sources explain the other four metrics, because they show which of your pages and which third-party pages are shaping the answer.
Why most visibility numbers mislead
Most misleading visibility figures are not measurement errors. They come from a handful of choices about what gets counted and when it gets read.
Branded prompts in the headline number
A prompt that contains your brand name will name you almost every time. Mixed into the same figure as category prompts, branded prompts lift the total without saying anything about whether new buyers find you. Report them separately and use them to check accuracy.
A prompt set that keeps changing
Every prompt added or removed breaks comparability with the history before it. Adding prompts where you already do well raises visibility on paper and nowhere else. Fix the set, grow it in dated batches, and never remove a prompt because its answer is unwelcome.
Reading single days
Because answers vary between runs, a daily figure moves for reasons unrelated to your work. Read the trend over several weeks and look for a sustained change before looking for its cause.
Averaging engines too early
ChatGPT, Gemini, Perplexity, Claude and Grok answer the same question differently. A combined figure can hide strength on one engine and absence on another, which are two different problems with two different fixes.
What to do first
In rough order of return on effort:
- Write down the buyer questions that matter, in buyer wording, and freeze that list.
- Split branded prompts out before calculating anything.
- Measure each engine separately from clean sessions on a regular schedule, and keep the full answers.
- Add share of voice, position, sentiment and sources once the visibility baseline is stable.
- Judge changes on multi-week trends, not on the day after a change ships.
AI visibility is not a vanity metric when it is measured this way. It is the closest available answer to a simple question: when a buyer asks an assistant for a recommendation, are you in the answer?