Your boss asks whether the money you moved from link building into AI search is working, and your rank tracker has nothing to say. Positions one through ten do not exist inside a ChatGPT answer. If you cannot put a number in front of the people who approve budget, the shift looks like a bet instead of a plan, and it gets cut at the next review.
This article gives you a repeatable way to measure AI answer visibility using nothing more than a spreadsheet, a fixed prompt list and a little discipline. It also says plainly what the method cannot tell you. AskEO, the tool behind this blog, audits the pages you control; it does not sample live AI answers, so the two jobs complement each other.
What is AI answer visibility, and why can't a rank tracker show it?
AI answer visibility is the share of relevant prompts for which an AI answer engine names your brand or cites your URL in the response it generates. It is a frequency, not a position, because an AI answer has no ranked list for you to occupy.
A rank tracker records one deterministic result for one keyword at one moment. An answer engine builds a fresh response each time, drawing on retrieved pages, its own training and the wording of the prompt. Ask the same question twice and the cited sources can differ. That is why a single screenshot proves nothing, and why the measurement has to be a rate across repeated runs.
Two separate things get lumped together here, and you should record them separately. A mention means the engine writes your brand name in its answer text. A citation means it shows your URL as a source. You can be mentioned without being cited, and cited without being named in the sentence. Each tells you something different about how the engine sees you.
How do you measure AI answer visibility step by step?
You measure it by running a fixed prompt set against each engine on a schedule and logging the outcome of every run. The steps below are deliberately boring, because consistency is what makes month-over-month numbers comparable.
- Write 20 to 40 prompts that a buyer would actually type, covering problem, comparison and purchase questions. Use real phrasing from sales calls and support tickets, not keywords.
- Freeze the list. Give every prompt an ID and never reword it. Add new prompts under new IDs so old trend lines stay valid.
- Choose the engines that matter to your audience, for example ChatGPT, Perplexity, Gemini and Google AI Overviews, and decide whether you test with browsing or search enabled. Keep that setting identical every month.
- Run each prompt three to five times per engine in a fresh session with no prior chat history, so earlier answers cannot bias later ones.
- For every run, log four fields: mentioned (yes or no), cited (yes or no), which competitors appeared, and the URL that was cited if any.
- Calculate your rates and review them on a fixed date each month. Do not check daily; daily numbers are mostly noise.
Use a clean browser profile or the API where the engine offers one, and note the date of every run. Personalised history and location can shift results, so the log should record the conditions as well as the outcome.
Which metrics belong in the report?
Four metrics are enough: mention rate, citation rate, share of voice against named competitors, and source-page concentration. Anything more makes the report harder to read without changing a decision.
| Metric | How to calculate it | What it tells you |
|---|---|---|
| Mention rate | Runs that name your brand divided by total runs | Whether the engine knows you exist for these topics |
| Citation rate | Runs that cite one of your URLs divided by total runs | Whether your pages are trusted as sources |
| Share of voice | Your mentions divided by all brand mentions across the set | Your standing against the competitors you named in advance |
| Source concentration | Share of your citations that point to your top three URLs | Whether one page carries all your visibility |
Report each metric per engine, not blended. A strong rate in Perplexity and a weak one in ChatGPT point to different causes: one may be a retrieval problem, the other an index or crawler-access problem.
Source concentration is the metric most teams skip and the one that most often changes the plan. If nearly all your citations point at one guide, you have a single point of failure. Losing that page to a redesign or a robots.txt mistake would erase most of your visibility overnight.
How many prompts and runs are enough?
Twenty to forty prompts with three to five runs each is a practical floor for a small site. That gives you somewhere between 60 and 200 data points per engine per month, which is enough to see a large change and not enough to see a small one.
The reasoning is arithmetic. With only 10 runs, a single extra citation moves your rate by ten percentage points, so you would chase noise. With 100 runs, one extra citation moves it by one point. Decide in advance what size of change you will act on, and treat anything smaller as unchanged. As a working rule, we would not call a shift real until it persists across two consecutive monthly reviews.
This is a stated opinion, not a published standard. Nobody has an official sample size for sampling AI answers, so write your threshold down before you look at the data. That prevents the common failure of deciding the threshold after seeing which way the numbers moved.
Where does the manual method break down?
It breaks down when you need to know why a number moved. The log tells you that your citation rate fell from one month to the next. It does not tell you whether a crawler was blocked, a schema block broke or a page stopped being readable without JavaScript.
That diagnosis is page-level work, and it is where an audit helps. AskEO fetches a page, checks its meta tags, JSON-LD, headings and AI crawler permissions, and returns SEO and AEO scores with a list of detected issues. Pairing the two gives you the outcome from your prompt log and the likely cause from the audit. See the AskEO plans and limits to compare the free scan with Pro, which adds historical score tracking.
To be exact about scope: AskEO does not query ChatGPT or Perplexity on your behalf and does not report your citation rate. Your spreadsheet does that part.
What if the numbers are too noisy to trust?
Noisy numbers are still useful if you compare like with like. The usual objection is that non-deterministic answers make any measurement meaningless, so why bother. The answer is that noise is the reason to use repeated runs and a rate. A rate across many runs is stable enough to compare month to month, even when any single answer is not.
What you should not do is compare across changed conditions. If you reword a prompt, switch the browsing setting or add an engine, start a new series and label the break in the chart. Keep the old series running in parallel for one month so you can see how much the change itself moved things.
You can also reduce the noise you introduce yourself. Run prompts at similar times, keep the account state clean and have the same person log the results, since two people will judge borderline mentions differently.
What should you do with the result?
Use the log to pick two or three prompts where competitors are named and you are absent, then fix the pages that should answer them. Read the AskEO blog for the specific page-level fixes, from crawler access to schema, and re-run only those prompts the following month.
Start by auditing the page you most want an engine to cite. Create a free AskEO account, run the homepage scan, and compare the issues it lists against the prompts where you are missing. The scan returns in roughly 30 seconds, and the free tier includes three lifetime homepage audits.
Frequently asked questions
How do you measure AI answer visibility for free?
You can measure it for free with a spreadsheet and a fixed list of prompts. Run each prompt several times in each engine, log whether your brand is mentioned and whether your URL is cited, then divide by total runs. The only cost is the time to run and record the prompts each month.
What is a good AI citation rate?
There is no published benchmark, so a good rate is one that rises against your own baseline and compares well with the competitors you named. Set the baseline in your first month, then judge changes only after they persist across two reviews. Avoid quoting a universal target, because it depends heavily on topic and market.
Is a mention the same as a citation in AI search?
No. A mention means the engine writes your brand name in the answer text, while a citation means it lists your URL as a source. You can have either without the other, so log them separately. Citations usually drive clicks, while mentions shape brand awareness inside the answer.
How often should you check AI answer visibility?
Check once a month on a fixed date. Answers vary from run to run, so daily or weekly readings mostly show noise. A monthly review of the same prompt set gives enough data points to see a real change, and it matches the pace at which content and technical fixes take effect.
Can an SEO audit tool show my citation rate?
Not unless it samples live AI answers, and AskEO does not. An audit checks the pages you control: meta tags, schema, headings and AI crawler permissions. Use it to find why a page is not being cited, and use your prompt log to measure whether the fixes changed the outcome.