You cannot improve what you do not measure. Most teams know their AI visibility is weak. Almost none can tell you their mention rate, their citation accuracy, or how often they lose the recommendation to a specific competitor.
This article defines the metrics that matter, how to collect them, and the cadence that keeps the data honest.
What Ai Visibility Means
AI visibility is whether AI systems name your brand when buyers ask questions about your category. It has three levels.
- Mentioned: your brand appears somewhere in the answer.
- Cited accurately: the facts the model states about you are correct.
- Recommended: the model positions you as a choice the buyer should consider.
Most brands overestimate their level. They appear once in ten answers and assume they are visible. Measurement replaces that guess with a number.
The Five Metrics That Matter
- Mention rate: of the prompts that matter to your business, the percentage where your brand appears at all.
- Citation rate: the percentage of answers where your domain or owned content is used as a source.
- Recommendation share: of answers that recommend options in your category, the percentage where you are one of them.
- Description accuracy: whether what the model says about you matches what you actually sell. Score each answer correct, partially correct, or wrong.
- Competitor share: which rivals appear, in how many answers, and ahead of or behind you.
Track all five. Mention rate without accuracy flatters you. Recommendation share without competitor share hides who is beating you.
Build Your Prompt Set
Your prompt set is twenty to thirty questions written the way buyers actually ask. Not marketing language. Buyer language.
Cover four types:
- Category questions: "what is the best CRM for small law firms"
- Comparison questions: "Acme vs Competitor for X"
- Problem questions: "how do I fix Y without Z"
- Shortlist questions: "top tools for X in 2026"
Write them once. Freeze them. Every measurement cycle uses the same set so the trend line means something.
Run the Measurement Cycle
Run every prompt in a clean session on ChatGPT, Claude, Gemini, and Perplexity. Prior context contaminates answers.
Record four things per answer: did you appear, were you described correctly, which sources were cited, which competitors appeared.
A spreadsheet is enough. Columns: date, platform, prompt, appeared, accurate, cited sources, competitors. One hour per week keeps the data current.
What Good Looks Like
Directional benchmarks beat fake precision. Judge yourself against these:
- Strong: you appear in most category answers, your description is correct, you are named alongside the top two competitors.
- Moderate: you appear sometimes, descriptions are partly right, competitors dominate shortlist prompts.
- Weak: you rarely appear, or the model describes a product you stopped selling years ago.
The trend matters more than any single reading. Flat week over week means nothing you shipped worked. Look at what changed before the move.
Manual Tracking vs Tools
Run it manually for your first month. Manual work teaches you how models retrieve, describe, and recommend. That context makes every later tool more useful.
When volume outgrows the spreadsheet, automate the prompt runs and reporting. The Analyst on this site runs a live diagnostic for free and returns your AI Visibility Score out of 100. It is the fastest way to get a baseline today.
Whatever you use, keep the frozen prompt set. Tools change. The baseline is what makes your numbers comparable.
