AI visibility can feel vague until you measure it, and it is more measurable than most founders expect. The core idea is simple: take the questions your customers actually ask, put them to the AI engines that matter, and count what comes back. Are you named? Are you cited as a source? How often, compared with your competitors, and in what tone? Those counts turn a fuzzy worry, are we showing up in AI search, into a set of numbers you can track and move.
- The base unit is a prompt run against an engine. Everything is counted across many prompts and repeated runs, not from a single question asked once.
- The metrics that matter are mention rate, citation rate, share of voice against competitors, and sentiment. Each answers a different question.
- One check is noise. A rate measured across repeated samples is signal, because engines give different answers to the same prompt between sessions.
- Rough benchmarks: under ten per cent of relevant prompts is near-invisible, twenty to thirty per cent is healthy, above forty per cent is category leadership. A young product often starts in low single digits.
Start with the right unit: prompts, not one question
The mistake is to ask an assistant one question, see your name or not, and conclude something. Engines are not deterministic. Ask the same thing twice and the wording, the tools named, and the sources cited can all change. So the honest unit of measurement is not a question, it is a set of buyer-intent prompts run repeatedly across each engine. You assemble the prompts a real customer would use, name the problem, ask for the best tool, ask for alternatives to a competitor, then run each one several times per engine and record what happens. From that you get rates, and rates are what you can trust.
The four metrics that matter
Mention rate
Mention rate is the share of relevant prompts where the engine names your product at all. It is the most basic measure of presence: out of the questions your customers ask, how often do you come up. Being mentioned is not the same as being recommended, but you cannot be recommended if you are never mentioned, so this is the floor everything else stands on.
Citation rate
Citation rate is how often the engine points to your own pages as a source for its answer. This matters because assistants that search the live web attach the pages they read, and being one of those cited sources both drives referral visits and reinforces the model's sense that you are a credible authority on the topic. A product can be mentioned from training memory yet never cited live, or cited heavily yet described thinly. Watching both tells you which lever to pull.
Share of voice
Share of voice sets your presence against your competitors for the same prompts. Being named in twenty per cent of answers means one thing if nobody else appears and quite another if a rival is named in eighty. Share of voice is the metric that turns a private number into a competitive one, and it is usually the most motivating, because it shows the gap you are actually trying to close.
Sentiment
Sentiment is how you are described when you are named. Being called a simple, affordable option for solo founders is a different result from being called a limited tool that lacks integrations, even though both count as a mention. Accuracy sits here too: engines sometimes place a product in the wrong category or attach the wrong use case, and a confident wrong description can cost you more than an absence.
The diagnostic behind these numbers: why you are invisible in ChatGPT →
How to read the numbers honestly
Measure across several engines rather than one, because ChatGPT, Gemini, Google's AI Overviews and Perplexity source and weight differently, and a strong showing in one can hide a blank in another. Track the trend rather than the snapshot, because the value is in whether the rate moves after you do the work, not in any single reading. And resist vanity prompts: measuring questions that contain your own name inflates the numbers and teaches you nothing. The prompts worth tracking are the ones a customer would ask without knowing you exist.
- Under ten per cent mention rate on relevant prompts: effectively invisible, the priority is presence.
- Twenty to thirty per cent: a healthy, established presence in the category.
- Above forty per cent: you are one of the default answers, the work shifts to defending and widening it.
- Low single digits: normal for a young product, and the honest starting line rather than a failure.
Where to go next
You can gather all of this by hand: write your prompts, run them across the assistants a few times each, and tally the mentions, citations, competitors and tone in a spreadsheet. It is tedious but it works, and doing it once teaches you what the numbers feel like. AfterLaunch automates the same measurement, running your buyer-intent prompts across four engines on a schedule and scoring mention rate, citations, share of voice and sentiment over time. The free Growth Snapshot gives you the first reading in about a minute so you have a baseline to improve against.
The wider shift these metrics track: the new playbook for the AI search era →What is the single most important AI visibility metric?
Mention rate on customer-intent prompts, measured across several engines and repeated runs. It is the floor: presence in the answers people actually ask for. Once you have presence, share of voice against competitors and sentiment tell you how much further there is to go.
Why measure a rate instead of just asking once?
Because engines are not deterministic. The same prompt can return different tools and sources between sessions, so a single answer is noise. Running each prompt several times and reporting the share that named you gives a stable number you can trust and track.
What is a good AI citation rate?
There is no universal figure, because it varies by category and how much of the web writes about your space. The useful comparison is against your own competitors on the same prompts, and against your own trend over time. Moving the rate up after you do the work is what matters, not hitting an absolute target.
Is being mentioned the same as being recommended?
No. A mention means the engine named you. A recommendation means it named you as a good choice for the specific need. Sentiment is what separates the two, which is why it is worth measuring how you are described, not just whether you appear.
Can I measure this myself for free?
Yes. Write the prompts a customer would ask, run each a few times across two or three assistants, and record mentions, citations, competitors and tone. It is manual but honest. The free Growth Snapshot from AfterLaunch does the same measurement automatically across four engines if you would rather start with a baseline.