AI visibility metrics measure how often AI assistants such as ChatGPT, Gemini, Perplexity and Google's AI Overviews and AI Mode name your brand, how they describe it, which sources they rely on, and what that produces in traffic and demand. The core set is mention rate, share of voice, accuracy, cited sources, AI referral traffic and branded search lift.
None of these come from a single dashboard, and some of them are easy to misread. This guide explains what each AI SEO metric tells you, how to measure it with tools you already have, how often to report it, and which numbers to ignore. Two companion guides go deeper: how to measure AI share of voice and how to measure the ROI of AI SEO.
What are AI visibility metrics?
AI visibility metrics are measures of how a brand appears in AI-generated answers. They cover presence (whether you are mentioned or cited), position against competitors (share of voice), quality (whether the description is accurate and favorable) and outcomes (AI referral visits, branded searches and leads). Together they show whether AI search optimization is working.
They sit alongside traditional SEO metrics, not in place of them. Rankings, impressions and organic clicks still matter, and many AI answers draw on the same pages that rank well. If you are new to the terminology, our explainer on AEO vs SEO vs GEO sets out how the disciplines overlap.
Why can't you measure AI search like rankings?
A ranking is a fairly stable position on a results page that a tool can check. An AI answer is generated fresh each time. The same question can return a different list of businesses depending on wording, the platform, whether it searched the web, the user's location and what was said earlier in the conversation. There is no position 1 to record.
That changes the method. Instead of checking one answer, you ask a fixed set of prompts several times and measure how often something happens. It is closer to polling than to rank tracking, and it means every AI visibility metric is an estimate with some noise in it. That is fine, as long as you measure the same way each time and watch the trend.
Which AI SEO metrics matter most?
| Metric | What it tells you | How to measure | Cadence |
|---|---|---|---|
| Mention rate | How often AI answers name you at all | Share of prompt runs where your brand appears | Monthly |
| Citation rate | How often your own pages are linked as a source | Share of runs that cite a URL on your domain | Monthly |
| Share of voice | Your presence compared with named competitors | Your mentions divided by all brand mentions in the set | Monthly or quarterly |
| Accuracy and sentiment | Whether you are described correctly and favorably | Review the text of brand prompts against your real facts | Monthly |
| Cited-source pages | Which pages AI tools trust for your topic | List every URL cited across runs and count repeats | Quarterly |
| AI referral traffic | Visits and conversions sent by AI tools | GA4 sessions from AI referrers and utm_source=chatgpt.com | Monthly |
| Branded search lift | Whether more people search for you by name | Branded queries in Google Search Console over time | Monthly, read quarterly |
Mention rate and citation rate
Mention rate is the share of prompt runs in which your brand is named. It is the most basic measure of AI visibility and the one to start with. Citation rate is narrower: the share of runs in which the answer links to a page on your website as a source. Perplexity, ChatGPT search, AI Overviews and AI Mode all show sources, so citation rate can be recorded for those. An answer can mention you without citing you, which usually means it learned about you from a directory, review site or article rather than your own pages.
Share of voice
Share of voice compares your mentions with those of a fixed set of competitors across the same prompts. A 40% mention rate means little on its own; it means a lot if your nearest competitor sits at 70%. Scoring can also separate a passing mention from an active recommendation. The full method, with a worked example, is in our guide to AI share of voice.
Accuracy and sentiment
Being named is only useful if the description is right. Ask brand prompts such as "what does [your company] do" and "is [your company] any good", and check the answers against reality: services, locations, prices if you publish them, opening hours, and who you serve. Record errors and the tone of the description. Wrong facts are usually traceable to an outdated listing or page, which makes this the quickest metric to improve.
Cited-source pages
Keep a running list of every URL that AI answers cite for your prompts, whether it is yours or not. After a few rounds, the same handful of pages tend to recur: a directory, a comparison article, a Reddit thread, a competitor's guide. Those pages are shaping the answers in your market. They tell you where to earn a mention and which of your own pages are doing the work.
AI referral traffic in GA4
Some AI answers send clicks, and GA4 records them as referral sessions. Filter session source for the main AI referrers:
- chatgpt.com, plus any session tagged utm_source=chatgpt.com
- perplexity.ai
- gemini.google.com
- copilot.microsoft.com
- claude.ai
OpenAI's Publishers and Developers FAQ states that ChatGPT adds utm_source=chatgpt.com to referral URLs, so those visits are easy to isolate. A custom channel group in GA4 that groups these sources as "AI assistants" keeps the numbers in one place, and lets you see conversions as well as sessions. Treat this as a floor, not a total: many AI answers produce no click, and some clicks arrive without a referrer and land in direct traffic.
Branded search lift in Search Console
People who hear about you in an AI answer often search for your name afterwards. Filter Google Search Console's performance report to queries containing your brand name and watch impressions and clicks over time. Note what Search Console does not do: Google's documentation on AI features says traffic from AI Overviews and AI Mode is included in the overall Web search type data, not reported separately, so you cannot isolate AI Mode clicks there. Branded lift is a signal of growing awareness, not proof of where it came from.
How do you build a prompt set?
The prompt set is the foundation of every AI visibility metric, so build it once, carefully, and keep it fixed.
- Start from real questions: sales calls, support emails, the queries in Search Console and the questions customers ask before buying
- Group prompts by intent: recommendation ("best X for Y"), problem ("how do I fix Z"), comparison ("X vs Y") and brand ("is X any good")
- Write them the way people type into a chat, which is usually longer and more specific than a search keyword
- Include location or industry where it matters, since AI answers for "in Maryland" and "for law firms" differ
- Aim for 20 to 50 prompts for most businesses; more if you serve several markets or product lines
- Version the set: when you add prompts, add them as a new group so older results stay comparable
For the step-by-step routine of running and recording prompts by hand, see our guide to tracking ChatGPT brand mentions.
How many times should you run each prompt?
More than once. Because answers vary from run to run, a single result can make you look better or worse than you are. Run each prompt at least three times per platform per reporting period, in a fresh chat each time, and record each run. Your mention rate is then the share of all runs that named you. Prompts that give the same answer every time are stable; prompts that flip between runs show you where the market is still open.
Keep the conditions consistent. Use the same accounts or logged-out sessions, the same location settings and the same day of the month, and record whether the platform searched the web. Changing the setup between periods changes the results as much as any real shift in visibility.
How often should you report AI visibility?
Monthly reporting suits most businesses. It gives enough runs to smooth out noise and is frequent enough to catch problems such as a wrong description. Quarterly is the right interval for decisions: whether share of voice moved, which cited sources to pursue, and whether branded search and AI referrals are rising. Weekly checks are only worth it for a few weeks after a big change such as a rebrand, a launch or a press campaign.
A useful monthly report fits on one page: mention rate and share of voice by platform, accuracy issues found and fixed, the top cited sources, AI referral sessions and conversions, and branded search trend. Any change should be read against the previous three months, not the previous one.
What should you not trust?
- One-off screenshots. A single answer showing you first, or missing, proves nothing about how often it happens
- Single "AI visibility scores" with no method behind them. If a tool or agency cannot show the prompts, the platforms, the number of runs and how the score is calculated, the number cannot be checked or compared
- Results from changed prompt sets. If the prompts changed, the trend is broken
- API results presented as app results. Some tools query platforms through an API, which can answer differently from the apps customers use. Ask which method is used
- AI referral traffic as the whole story. It counts clicks, and many AI answers are read without one
Note - Before you buy an AI visibility report, ask to see the raw prompt list and a sample of the recorded answers. A report you cannot trace back to real answers is a vanity metric.
How do you measure GEO success?
Generative engine optimization is working when the measures move in a sensible order. First, accuracy improves and your own pages start appearing as cited sources. Then mention rate and share of voice rise across your prompt set. Finally, AI referral sessions, branded searches and leads that mention AI tools increase. If outcomes rise while mention rate is flat, look for another cause. If mention rate rises but nothing else does, check whether you are being recommended or merely listed.
Set expectations by platform. Answers that search the live web, such as Perplexity, ChatGPT search and AI Overviews, can reflect changes within weeks. What a model knows without searching changes only when it is retrained. Our guide to what AI visibility is explains the causes behind low visibility and what changes it.
The short version
Measure AI search performance with a fixed prompt set run several times a month across ChatGPT, Gemini, Perplexity and Google's AI features. Track mention rate, citation rate, share of voice, accuracy and cited sources, then connect them to AI referral traffic in GA4 and branded search in Search Console. Report monthly, decide quarterly, and ignore any number you cannot trace back to real answers. If you want the measurement and the improvements handled together, that is what our AI visibility optimization service does.
