In short
"AI visibility" is not one number. It is a ladder of five levels: AI reads your site, cites your pages, mentions your brand, recommends you, and describes you correctly. Each level needs its own metric, and every metric only means something with its setup: the questions asked, the AI model, whether it searched the web, the location, the dates and how many answers are behind it.
- One answer proves nothing: AI answers vary, so report rates over many answers and judge trends over weeks, not days.
- Live visibility (what AI finds on the web today) and Memory visibility (what it knows from training) move for different reasons.
- Measure by buying stage: being found when buyers explore is not the same as being chosen when they decide.
One word, five different things
Ask five tools what "AI visibility" means and you can get five different answers, because the word covers a whole chain of events. Before an AI assistant can recommend a brand, a lot has to go right first. We think of it as a ladder:
- 1Read
AI crawlers visit your site and can actually read the pages that matter. If they can't, nothing above this step happens.
Measured with: server logs / crawl insights - 2Cited
Your pages, or pages about you, are used as sources when AI builds an answer.
Measured with: citation rate, source share - 3Mentioned
Your brand appears in the answer, perhaps as one name in a list of eight.
Measured with: visibility (mention rate) - 4Recommended
AI puts you first or presents you as the best fit for the buyer's need. This is the step that moves revenue.
Measured with: position, share of voice - 5Described correctly
What AI says about you is accurate and positive: the right category, the right strengths, no outdated facts.
Measured with: sentiment, perception
A single "visibility score" usually measures step 3 only. That's useful, but it can hide a lot. A brand can be mentioned often and still never be the one AI recommends. It can be recommended while AI quotes an old price. It can lose citations long before its mentions drop. The first question to ask about any AI visibility number is: which step of the ladder is this?
A prompt is a coin, not a ruler
Ask an AI assistant the same question twice and the answer can change: different brands, a different order, different sources. A single answer is closer to one coin toss than to a measurement. What you actually want to know is the underlying rate: how often, out of many tries, does AI mention you?
The number of answers behind a figure decides how much it can be trusted. A quick illustration, using the standard way of estimating a range for a rate:
| Observed | Shown as | Plausible true rate (95% range) |
|---|---|---|
| 3 mentions in 5 answers | 60% | roughly 23% to 88% |
| 60 mentions in 100 answers | 60% | roughly 50% to 69% |
Both lines print "60%" on a dashboard. Only one of them tells you something. In practice this means three things:
- Ask many different questions. A broad set of prompts that covers how your buyers really ask gives a sturdier picture than the same handful of prompts repeated over and over.
- Judge trends over weeks, not days. Daily numbers wobble by chance. A change that holds across a week or a month is far more likely to be real.
- Separate branded from open questions. "Is [your brand] good for X?" almost always mentions you. "What's the best tool for X?" is where the real competition happens. Mixing them inflates the score.
Live or Memory: which layer moved?
When an AI assistant answers, it draws on two layers. With web search on, it looks things up and builds the answer from the pages it finds right now. With search off, it answers from what it learned in training. We track them as two separate metrics, because they change for different reasons:
What AI finds today
Moves with the pages AI retrieves and cites: your site, reviews, forums, press. When your sources change, it can change within days.
What AI already knows
Reflects what the model learned before its training cutoff. It changes slowly, mostly when a new model version is released.
If your visibility drops, the first diagnostic question is which layer dropped. A live drop points to sources: a review page that stopped ranking, a competitor's new comparison article. A memory gap points to long-term presence: how consistently the web has described you over time.
Measure by buying stage, not just by brand
Buyers don't ask AI one question. They explore, compare, then decide, and AI can treat your brand very differently at each stage:
| Stage | Typical question | What to measure |
|---|---|---|
| Discovery | "What tools exist for X?" | Are you mentioned at all? |
| Comparison | "X vs Y for a small team?" | Are you positioned well against competitors? |
| Decision | "Which one should I buy?" | Are you the one AI recommends, and is the description accurate? |
A brand that is visible in Discovery but disappears in Decision has a different problem from one that is never found at all. One overall score averages the two away. Grouping prompts by stage shows where in the journey you win and where you lose.
Real conversations go further still: buyers add requirements, push back and ask follow-ups, and a recommendation often appears several turns in. Single-question tracking is a clear, repeatable baseline, but it captures the opening of that conversation, not the whole of it. A good measurement says which one it is.
From visibility to value
Visibility is a means, not the goal. Three signals connect it to the business:
- AI crawler activity: which AI bots read your site, why (training, search or a live user question) and which pages. It's the bottom of the ladder, and the earliest warning when something blocks AI from your content.
- AI referrals: visits that arrive from AI assistants, in your analytics.
- "How did you hear about us?": asking new customers directly, and tracking the revenue from those who answer "ChatGPT" or "an AI assistant". Simple, imperfect, and still one of the clearest signals available.
The setup behind every number
Whatever you measure, the number only means something with its setup. Before you trust or compare an AI visibility figure, check that it comes with:
- The questions asked, grouped by topic and buying stage
- The AI apps or models, and whether answers came from the app or the API
- Whether web search was used (Live) or not (Memory)
- The location and language
- How many answers are behind each figure, and over which dates
- The sources AI cited
- Which step of the ladder the number measures
When a number moves, this is where you start looking: did the questions change, the model, the search behaviour, or the sources?
How Vosapio measures
- Every step of the ladder, separately. Crawl insights for read, sources for cited, visibility for mentioned, position and share of voice for recommended, sentiment and perception for described.
- The AI apps your buyers use, every day. Your tracked prompts run daily, for the location you choose, and every answer keeps its cited sources.
- Live and Memory, side by side, so you know which layer moved.
- Trends, not snapshots. Daily data rolled up into weekly and monthly views, so you act on real changes instead of noise.
Frequently asked questions
What does AI visibility actually measure?
It depends on the tool. AI visibility can mean five different things: AI crawlers reading your site, your pages being cited as sources, your brand being mentioned, your brand being recommended, and your brand being described correctly. Most visibility scores only measure mentions, so always check which of these a number refers to.
Why do AI visibility numbers change every day?
AI answers vary from one run to the next, even for the same question. A daily figure based on a few answers is mostly noise. Reliable tracking uses many different prompts, reports rates instead of single answers, and judges changes over weeks rather than days.
How many prompts do I need to track?
Enough to cover how your buyers really ask, across topics and buying stages. A broad set of different prompts gives a more reliable picture than a few prompts repeated many times. A rate based on a handful of answers can be off by tens of percentage points.
What is the difference between Live visibility and Memory visibility?
Live visibility is how often a brand appears when the AI searches the web while answering, so it changes with the pages AI retrieves and cites. Memory visibility is how often a brand appears when the model answers with search off, from what it learned in training, so it mainly changes when a new model version is released.
Is being mentioned by AI the same as being recommended?
No. Being mentioned means your brand appears somewhere in the answer, often in a list. Being recommended means AI presents you first or as the best fit for the buyer's need. A brand can be mentioned often and rarely recommended, which is why position and share of voice matter alongside visibility.
Key takeaways
- "AI visibility" is a ladder: read, cited, mentioned, recommended, described correctly. Know which step a number measures.
- One answer is a coin toss. Use many different prompts, report rates, and judge trends over weeks.
- Separate branded from open questions, and group prompts by buying stage.
- Live and Memory visibility move for different reasons. Find out which layer moved before you act.
- Connect visibility to value with crawler activity, AI referrals and asking customers how they found you.
See every step of your AI visibility
From the crawlers that read your site to the answers that recommend you, with the sources behind every one.