ExplainerAI visibility
AI Visibility Metrics: What to Measure and What to Ignore
by the AEO GEO Labs team6 min read
The AI visibility metrics worth measuring are mention rate, citation rate, share of voice, recommendation rate and the sources engines cite, each calculated across repeated runs of a fixed prompt set. Ignore single-run rank positions, blended scores you can't decompose, and prompt volume figures presented as fact. AI answers vary too much between runs for anything measured once to mean much.
These metrics apply the measurement ideas from what AI visibility is to a reporting setup.
The AI visibility metrics that matter
Five AI visibility metrics carry most of the useful signal. Each is a rate, calculated per engine first and then combined.
| Metric | Formula | What it tells you |
|---|---|---|
| Mention rate | Runs mentioning your brand ÷ total runs | How often you exist in the answer |
| Citation rate | Runs linking your domain ÷ total runs | How often your pages are used as a source |
| Recommendation rate | Runs where you're recommended or listed first ÷ total runs | How often you win the answer, not just appear |
| Share of voice | Your mentions ÷ mentions of you plus tracked competitors | Your position relative to the field |
| Source share | Citations by domain across all runs | Which sites decide your category's answers |
Mention rate is the base metric. If you track 30 prompts on three engines with three runs each, that's 270 runs. Being named in 81 of them is a 30% mention rate.
Citation rate is separate because engines often name a brand without linking it, or link a page without naming the brand in the text. Both matter, for different reasons: mentions build awareness, citations can send clicks.
Recommendation rate catches the difference between "Acme is one option" and "For your situation, Acme is the best choice." Define your rule once and apply it consistently.
Share of voice puts your numbers in context. A 20% mention rate is strong if competitors sit at 10% and weak if they're at 60%. The full calculation, including weighting choices, is in AI search share of voice.
Source share is the most actionable metric on the list. It tells you which review sites, publications and forums engines rely on for your prompts, which is where outreach should go.
Supporting metrics worth tracking
Supporting AI visibility metrics add context but shouldn't be the headline.
- Sentiment mix: share of mentions that are positive, neutral or negative. Useful for spotting reputation problems. Automated sentiment scoring is rough, so spot-check it.
- Accuracy rate: share of mentions that describe your product, pricing and category correctly. Errors here point to outdated sources.
- AI referral traffic: visits from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com in your analytics. It shows whether mentions turn into visits, though it undercounts, because many readers never click.
- First-party citation counts: Microsoft's AI Performance report in Bing Webmaster Tools counts how often Copilot and Bing AI summaries cite your pages. As of October 2026, it's the only engine-reported citation data we know of from a major engine.
Undercounting is real. Pew Research Center found that people who saw a Google AI summary clicked a link inside the summary on just 1% of visits to results pages that showed one, in March 2025. Referral traffic tells you about the clickers, not the readers.
AI visibility metrics to ignore (or demote)
Some AI visibility metrics look precise but mostly measure noise. Demote these in your reports.
- Single-run rank position. "We're number two in ChatGPT" means little. SparkToro and Gumshoe found roughly a 1 in 1,000 chance of two answers listing the same brands in the same order.
- Blended visibility scores you can't decompose. A single number from 0 to 100 hides which engine, prompt type or competitor moved. Use it only if you can see the inputs.
- Prompt volume as fact. No tool can see what people type into ChatGPT. Volume figures are modeled from search data or panels. Fine for prioritizing, not for forecasting.
- Daily movements. Day-to-day changes in a 30-prompt set are mostly sampling noise. Read monthly trends.
- "Impressions" in AI answers. Some tools multiply mention rate by estimated prompt volume. That stacks one estimate on another.
Warning: the more decimal places a metric has, the more suspicious you should be. AI visibility is a sampling exercise, and samples have error bars.
How many runs make a metric reliable
A metric is reliable when the sample behind it is large enough that the next month's change isn't just chance. There's no published standard for AI visibility, but basic sampling logic gives a guide.
- Per prompt: at least three runs per engine for a manual check. More is better.
- Per report: a few hundred total runs before you compare months. Thirty prompts, three engines and three runs gives 270.
- Per segment: if you split by engine or prompt type, each slice needs enough runs on its own. Thirty runs per slice is a sensible floor.
- Per change: treat movements under about 5 to 10 percentage points on a few hundred runs as noise until they persist for two or three months.
These are rules of thumb, not statistics you should quote. The point is direction: more runs, fewer slices, longer windows.
Building an AI visibility metrics report
An AI visibility metrics report should answer three questions: are we showing up, against whom, and what changed. A simple monthly structure:
- Headline: mention rate and share of voice, overall and per engine, against last month and three months ago.
- Competitors: share of voice table for you and three to five rivals.
- Sources: top 20 cited domains and whether you're present on each.
- Accuracy: any answers that misstate your product, with the source cited.
- Traffic: AI referral sessions and conversions, labeled as a floor, not the full impact.
- Actions: what you'll change next month based on the sources and errors.
For collecting the data, see how to track AI brand mentions. For connecting these metrics to business outcomes across GEO work, see how to measure GEO results.
Frequently asked questions
What are the most important AI visibility metrics?
The most important AI visibility metrics are mention rate, citation rate, recommendation rate, share of voice against named competitors, and source share, meaning which domains engines cite for your prompts. All should be calculated as rates across repeated runs of a fixed prompt set, per engine. Sentiment, accuracy and AI referral traffic are useful supporting metrics.
Which software has the most accurate AI visibility metrics?
No software is clearly most accurate as of October 2026, because every tool measures its own test sessions rather than real user conversations. Accuracy depends on how many times each prompt runs, which locations are used, how answers are parsed and whether results are shown as rates. Ask vendors about runs per prompt and raw answer exports before trusting any score.
Is AI referral traffic a good AI visibility metric?
AI referral traffic is a useful but incomplete metric. It shows real visits from ChatGPT, Perplexity, Gemini and Copilot, and those visits can be tied to conversions. But most people who read an AI answer don't click, so referral traffic undercounts the reach of your mentions. Report it alongside mention rate and share of voice, labeled as a floor.
How often should AI visibility metrics be reported?
Monthly reporting suits most teams, because daily and weekly changes in AI answers are dominated by run-to-run noise. Use the same prompt set each month so results are comparable, and compare against both the previous month and three months earlier. Weekly data is worth collecting during launches or campaigns, but judge results on the monthly trend.
If you want these metrics tracked and reported for you, AEO GEO Labs runs GEO, AEO and SEO programs for B2B and SaaS teams. See our services.