How it works
How citecheck measures AI visibility
Updated September 30, 2026
AI visibility is how often an AI assistant recommends you when a buyer asks it for a product, a service or a place. citecheck measures it by asking the assistants the questions your buyers ask, reading every answer, and scoring how prominently you are named. This page explains each step, so you can judge the number before you trust it.
1. The questions
An audit starts from your website, your category and your competitors. citecheck writes buyer questions in the shapes people really use, such as:
- Best of: "best screenshot tool for indie hackers", "best dentist in Austin".
- Comparison: "CleanShot X vs Shottr".
- Alternatives: "alternatives to Snagit".
- Problem first: "how do I capture a scrolling web page on a Mac".
- How to and direct questions about you by name.
The mix depends on what you sell. Software buyers ask more comparison and switching questions; local buyers mostly ask "best X near Y".
2. The assistants
citecheck asks seven assistants: ChatGPT, Claude, Gemini, Grok, DeepSeek, Meta AI and Qwen. Answers come from each vendor's API, not from their consumer apps, so they are close to what a buyer sees but not identical.
ChatGPT, Claude and Gemini search the web before they answer, the way their apps do for a question like this. Their answers make up your score.
Grok, DeepSeek, Meta AI and Qwen answer from training data through the APIs citecheck uses. What they say is still worth knowing, because it is what they learned about you, but it is a different measurement. They are reported beside the score and never averaged into it.
The free plan asks ChatGPT, Gemini, Claude and DeepSeek. Paid plans ask all seven, from $29 a month.
3. Reading each answer
An AI judge reads every answer and records where you appear, if at all, and which competitors are named. Each answer gets one verdict:
- Recommended first: you are the top pick. 3 points.
- Recommended: you are recommended, but not first. 2 points.
- Listed: you are named among options, without a recommendation. 1 point.
- Cited only: you are not named, but one of your pages is a source. 0.5 points.
- Absent: you are not there. 0 points.
citecheck also records the pages each answer cites or reads, and marks which of them name you.
4. The score
Your score for an assistant is the points you earned as a share of the most you could have earned, out of 100. Recommended first on every question is 100; absent from every answer is 0.
A failed call is never counted as absent. If an assistant times out or refuses a request, that answer is left out of the score and the report says so. An outage is not evidence that an assistant doesn't know you.
5. Asking more than once
Assistants rarely give the same answer twice. A tracked run asks every question five times on every assistant and scores the pattern, so one lucky or unlucky reply doesn't decide your number.
6. The pages that decide it
Assistants that search recommend what the pages they read recommend. citecheck counts every page cited across a run and splits them into the ones that already name you and the ones handing the answer to a competitor. Those are the pages to get onto. The guide to getting recommended by AI explains how.
7. What the number can't tell you
API answers are close to app answers, not identical: apps add their own instructions, personalisation and memory. Answers also vary by day and by wording. Treat one audit as a snapshot and a trend over weeks as the signal. Questions are in the FAQ.