The first instinct when teams start tracking AI visibility is to count mentions: “ChatGPT named us 12 times this week.” It feels like progress. It mostly isn’t. A mention buried at the end of a list, hedged with “though it can be pricey,” is not the same as being the first tool the assistant reaches for.
If you’re going to measure AI visibility, measure the things that change buyer behavior. Here are five.
1. Appearance rate
Because answers vary run to run, a single check tells you almost nothing. Appearance rate is how often you show up across many runs of the same question. Showing up in 9 of 10 runs is a real position; showing up in 1 of 10 is noise you got lucky on. Always measure across repetitions, per question, per model.
2. Recommendation strength
Not all mentions are equal. Recommendation strength captures how you’re recommended, combining signals like:
- Position — first on the list versus an afterthought.
- Endorsement — “the best choice for X” versus “another option is Y.”
- Sentiment — the tone of the surrounding context.
- Context — whether you’re matched to the buyer’s actual use case.
Collapsing these into one comparable score is what lets you track movement over time and benchmark against competitors. In Cypress this is the Recommendation Score.
3. Share of voice vs. competitors
Visibility is relative. The question that matters to a founder isn’t “do we get mentioned?” — it’s “when a buyer asks about our category, who gets recommended, and how often is it us versus them?” Tracking your appearance rate next to your top three competitors, per question, turns a vague worry into a clear scoreboard.
4. Category and prompt coverage
A product can be strong on its branded question (“is Acme good?”) and absent from the questions that actually drive discovery (“best tools for X,” “alternatives to Y,” “X for early-stage startups”). Coverage measures how many of the buyer-intent prompts in your category you appear for at all. Gaps here are usually the highest-leverage place to work, because they represent demand you’re completely missing.
5. Source attribution
When you do get recommended, what is the model leaning on? When you don’t, what is it citing for the competitor that beat you? Tracing recommendations back to their likely sources turns measurement into action: it tells you exactly which review sites, roundups, or threads to go earn a presence in.
How to read them together
No single number is the answer. Read them as a stack:
Coverage tells you where to compete · Appearance rate tells you if you’re competing · Recommendation strength tells you how well · Share of voice tells you against whom · Source attribution tells you what to do next.
A high score on a branded prompt with zero coverage of category prompts means you look healthy and are quietly invisible to new buyers. Strong coverage with weak recommendation strength means you’re in the conversation but losing it on framing. The combination is the diagnosis.
This is precisely the stack Cypress reports on — appearance, Recommendation Score, competitor share of voice, category coverage, and source intelligence — across ChatGPT, Claude, Gemini and Perplexity, refreshed over time so you can see the trend rather than a one-day snapshot. Measure the things that move buyers, and AI visibility stops being a mystery and starts being a roadmap.