GUIDE · PART OF AI VISIBILITY TRACKING
AI visibility measurement,
with the workings shown.
An AI visibility score looks precise, and usually it isn't. The number depends on which prompts were asked, how many times, and how the tool did its sums. This guide sets out the measures worth using, the formula behind each one, and how much weight each can bear.
What is AI visibility measurement?
AI visibility measurement is the process of turning AI answers into numbers that can be compared over time. It begins with a sample: a set of prompts, run on a set of engines, a set number of times. Each answer is checked, and the checks are added up into metrics such as mention rate, citation rate and share of voice. Because the person measuring chooses the sample, the results describe that sample and nothing wider.
Key facts
- Every AI visibility metric is a count over a sample of answers
- The sample is prompts multiplied by engines multiplied by runs
- Change the prompts and every number changes with them
- Scores from two different tools can't be compared, because each builds its score in its own way
- The reliable reading is your own trend, on the same sample, against a recorded baseline
- In SEORCE, the Visibility Score is calculated from the prompt runs in your own project
Last reviewed 15 October 2026 by the SEORCE team.
A score to one decimal place can still be a guess.
The arithmetic is the easy part. Three choices sit underneath every figure, and few reports mention any of them.
You chose the questions
Forty prompts about your strongest product will score higher than forty about your weakest. Neither is "your AI visibility".
One name, many formulas
Two tools can both report a "visibility score" and mean different things by it. The label is shared and the method isn't.
Ten answers isn't a trend
With few prompts and few runs, a change of one answer moves the number by several points.
A sample, some counts and a baseline.
That is all AI visibility measurement is. Get these three right and the numbers will mean something.
Fix what you're measuring.
Write the sample down and leave it alone. Every comparison you make later depends on it staying the same.
- The promptsA fixed list, sorted by stage.
- The enginesThe same ones each time.
- The runsHow many times each prompt is asked, and how often.
Know the sum behind each number.
None of the formulas is complicated. Knowing them is what lets you answer the question "how did you get that?"
- Mention rateAnswers that name you, divided by answers tracked.
- Citation rateAnswers that used one of your pages, divided by answers tracked.
- Share of voiceYour mentions, divided by all brand mentions on the same prompts.
- SentimentThe average score of the answers that mention you. SEORCE scores from 0 to 100.
Measure change, on the same sample.
A single reading tells you little. Two readings of the same sample, some weeks apart, tell you whether anything moved.
- Record it before you change anythingSo the later figure has something to stand against.
- Keep the sample fixedIf you add prompts, start a new baseline for them.
- Report the change, with the datesAnd say what was done in between.
Measuring it so the number holds up.
Five steps, and the first two are the ones people skip.
Fix the prompt list
Decide the questions and sort them by stage. Don't keep swapping them.
AI mention tracking →Fix the engines and runs
The same engines, the same number of runs, on the same schedule.
Count
Apply the formulas to every answer in the sample.
Record the baseline
Date it, and keep the answers behind it.
Measure again
On the same sample. Report the change and what was done in between.
Free checks that come before measuring.
Measurement needs repeated runs, so it lives in the platform. What you can do without an account is check that your pages are readable, which is the first thing to rule out when a number is low.
How far to trust each number.
| Metric | The sum | Good for | Not good for |
|---|---|---|---|
| Mention rate | Answers naming you ÷ answers tracked | Your own trend, and gaps by engine or stage. | Comparing with another company's published figure. |
| Citation rate | Answers using your pages ÷ answers tracked | Seeing whether your pages are being used. | Estimating traffic. A citation isn't a click. |
| Share of voice | Your mentions ÷ all brand mentions | Comparing with rivals on the same prompts. | Anything, if rivals were measured on different prompts. |
| Sentiment | Average score of answers that mention you | Following tone over time. | Reading much into it when mentions are few. |
| Visibility score | Different in every tool | One line on a chart, for your own project. | Comparing tools, or benchmarking against others. |
No formula is given for a visibility score, because every tool builds its own.
Your sample, your numbers.
AI Beacon runs the prompts you choose on 7 AI engines + Google AI Overview, AI Mode & Search, keeps every answer, and counts from there. The Visibility Score is calculated from the prompt runs in your project, so it is a measure of your own trend.
Frequently asked questions.
An AI visibility score is a single number that sums up how present a brand is across a set of tracked AI answers. Each tool builds it differently, so the same brand can score differently in each.
By sampling. A fixed set of prompts is run on a fixed set of engines a fixed number of times. Each answer is checked for the brand and its pages, and the checks are counted into metrics such as mention rate and citation rate.
Because they use different prompts, different engines, different numbers of runs and different formulas. The scores share a name and little else. Compare your own trend inside one tool.
There's no fixed figure. Twenty to fifty prompts, run several times on each engine, is a workable start. The test is whether the number stays steady from week to week when nothing has changed.
Yes, if they are measured on the same prompts as you. That is what share of voice does. A figure a competitor publishes about themselves was measured on their prompts, so it can't be set beside yours.
There isn't a universal one. A score only means something next to your own earlier score, on the same sample.
You can start on the free forever plan, which runs your prompts and keeps the answers. The free tools and browser extension don't measure visibility. They check whether your pages can be read.
Set up a sample you can measure against.
Add your site and your first prompts. From that day you have a baseline, and every later reading has something to stand against.
Not ready for an account? Use the free browser extension or any of the free tools.