Measure AI visibility without vanity metrics by testing a fixed set of 20 to 50 real buyer questions across several AI platforms, then recording mention rate, citation rate, position, sentiment, accuracy and the URLs cited. Repeat on a schedule. Avoid a single opaque score, one-off screenshots and counts from a single run.
AI visibility score: a single number that summarizes how visible a business is in AI answers. There is no shared standard, so a score from one tool cannot be compared with another.
Which metrics are worth tracking?
Track mention rate, citation rate, position, sentiment, accuracy and cited URLs, and compare them with competitors through share of voice.
| Metric | What it shows | How to calculate | Fictional example |
|---|---|---|---|
| Mention rate | How often you are named | Answers that name you divided by total answers | 4 of 30 = 13% |
| Citation rate | How often your page is linked as a source | Answers that link your site divided by total answers | 3 of 30 = 10% |
| Position | Where you appear when an answer lists businesses | Average rank across answers that list you | Average 3.2 |
| Sentiment | How the assistant describes you | Share of mentions that are positive, neutral or negative | 3 positive, 1 neutral |
| Accuracy | Whether what is said is right | Mentions with correct details divided by all mentions | 3 of 4 = 75% |
| Cited URLs | Which of your pages, and which other sites, are used as sources | List every URL cited, grouped by owner | 2 of your pages; 14 other sites |
| Share of voice | Your share of all business mentions | Your mentions divided by all mentions of you and competitors | 4 of 52 = 7.7% |
What to do next: Record all of these from the same set of answers, so each number has a denominator you can explain.
What counts as a vanity metric?
A vanity metric looks good but does not show real change, and the common ones are a single opaque score, one good screenshot and a mention count with no total.
| Vanity metric | Why it misleads | Use instead |
|---|---|---|
| One opaque score | You cannot see what it includes, and no two tools calculate it the same way | The separate metrics above, with the questions behind them |
| A screenshot of one good answer | Answers vary from run to run | Mention rate across many questions and runs |
| Total mentions with no denominator | Twelve mentions could be a lot or a little | Mention rate and share of voice |
| Visibility on one assistant only | Assistants differ sharply | Results split by assistant |
| Mentions that are wrong or negative | Being named inaccurately can hurt | Accuracy and sentiment alongside mentions |
What to do next: If a report gives you one number and no questions or answers behind it, ask for the evidence.
How many prompts do you need?
Use 20 to 50 real buyer questions for a reliable baseline, and treat 10 as the minimum to start.
More questions give steadier results. Run each question two or three times in new chats, report rates instead of yes or no, and treat small changes as noise. A move from 12% to 14% means little. A move from 12% to 40% after you made changes is worth noticing.
What to do next: Write 20 to 50 questions, mixing broad, specific and comparison questions, and run each more than once.
How should you cut the results?
Split them by assistant, by question type and by stage of the customer’s decision, because a single combined number can hide a problem.
- By assistant: ChatGPT, Gemini, Perplexity and Claude often differ sharply.
- By question type: broad (“best dentist in Austin”), specific (“emergency dentist open Sunday”) and comparison questions behave differently.
- By stage: early research, comparison and ready-to-buy questions.
A fictional audit: 30 buyer questions across four assistants
| Assistant | You were named | Top competitor named |
|---|---|---|
| ChatGPT | 4 of 30 (13%) | 18 of 30 (60%) |
| Gemini | 8 of 30 (27%) | 20 of 30 (67%) |
| Perplexity | 6 of 30 (20%) | 17 of 30 (57%) |
| Claude | 2 of 30 (7%) | 13 of 30 (43%) |
What the numbers say: visibility differs by assistant, and a competitor leads in all four. A combined average of 17% would hide that Claude and ChatGPT are the weakest.
This is a fictional example to show the format. It is not a CoreTxt test result and not data about a real business.
What is a good AI visibility score?
There is no universal good score, because there is no shared standard for how a score is built.
Compare yourself with your own earlier results and with the competitors named in the same questions. CoreTxt’s Deep Audit scores nine weighted categories, including AI Visibility (20%), Entity Clarity and Technical GEO, and marks a category not scored when data is missing. See the sample report.
How do you start?
Write 20 to 50 customer questions, run them across the assistants, record the results and repeat on a schedule.
- Write the questions a customer would ask, in their words, with your city or service area.
- Run each two or three times in new chats in each assistant.
- Record mentions, citations, position, sentiment, competitors named, errors and every URL cited.
- Calculate your mention rate and your top competitors’ rates.
- Re-run monthly, or every two weeks while you are making changes.
What to do next: The test guide has example questions and a recording sheet.
Frequently asked questions
How do you measure AI visibility?
Ask a fixed set of real customer questions in ChatGPT, Gemini, Perplexity and Claude, repeat each several times, and record mention rate, citation rate, share of voice and accuracy.
How do you track AI visibility?
Keep the same questions in a spreadsheet and re-run them on a schedule, monthly or every two weeks, so you can compare results over time. Re-run after major changes to your site.
What is an AI visibility score?
A single number that summarizes how visible a business is in AI answers. There is no industry standard, so each tool calculates it differently. Ask what goes into it before you compare numbers.
What are the main AI visibility metrics?
Mention rate, citation rate, position, sentiment, accuracy and the URLs cited, plus share of voice against competitors. Cut them by assistant and by question type.
How many prompts do I need to measure AI visibility?
Use 20 to 50 real buyer questions for a reliable baseline. Ten is the minimum to start. Run each one more than once across several assistants.
What is a vanity metric in AI visibility?
A number that looks impressive but does not show real change, such as one opaque score, a screenshot of a single good answer, or a mention count with no total to compare it with.
How do you measure brand mentions in ChatGPT?
Ask ChatGPT your customer questions in new chats, run each more than once, and count the share of answers that name your business. Record the competitors it names instead.
No one controls what an AI assistant says. These steps improve your chances of being understood and recommended. They do not guarantee a mention, a citation or a position.
Related guides
- How to test if AI recommends your business
- What is AI visibility?
- Free AI visibility audit tool
- How to choose an AI visibility tool
Want help with this?
CoreTxt monthly plans re-run your prompts every two weeks and report mentions, citations and accuracy by email. See pricing.
Run your free AI Visibility Snapshot
See how ready your website is for ChatGPT, Gemini, Perplexity and Claude. Free, no credit card.
Get My Free Snapshot