Key takeaways
- AI visibility is how often AI assistants name your brand or cite your pages when buyers ask about your category. It belongs on the same scorecard as paid and organic.
- I track 7 numbers: mention rate, citation share, answer position, source mix, answer accuracy, AI referral sessions and self-reported AI attribution.
- The prompt panel is the instrument. Keep it fixed, count branded and unbranded prompts apart and never score a link chip as a mention.
- At Commas the panel showed 8.4% of citations against 21% to 33% for peers. At Pipe17, 32% of new-business wins came through AI assistants.
01What AI visibility is
AI visibility is the share of AI-generated answers in your category that name your brand or cite your pages, across assistants like ChatGPT, Perplexity, Gemini and Google's AI Overviews. You measure it with a fixed panel of real buyer prompts run on a schedule, and you judge it by the traffic and pipeline it sends.
I track AI visibility like any other channel. At Commas I analyzed 10,560 ChatGPT and Google AI Overview answers in Profound. Commas earned 8.4% of citations against 21% to 33% for peers. That one number changed the conversation from whether to invest in AEO to which pages to build first, and the gap became 221 page opportunities and a 35-topic roadmap.
The program around it is in my AEO playbook. This note is the measurement layer.
02The metrics and their formulas
Compute every metric by engine, intent and topic. A blended number hides the topic you're losing.
| Metric | Formula | What it tells you |
|---|---|---|
| Mention rate | Unbranded answers that name you ÷ all unbranded answers | Whether you're in the consideration set at all. |
| Citation share | Citations to your domains ÷ all citations in the panel | How much of the evidence engines rely on is yours. My headline number. |
| Answer position | Answers where you're the first brand named ÷ answers that name you | Whether you're the recommendation or the afterthought. |
| Source mix | Citations to your own pages ÷ all citations in answers that name you | Whether your pages or third parties carry your mentions, and where to work next. |
| Accuracy and sentiment | Brand answers with no factual errors ÷ brand answers checked, plus the tone of each | Whether the answer about you is right. A wrong price or category costs deals you never see. |
| AI referral sessions | Monthly sessions from AI assistant sources, with their conversion rate | Whether visibility sends people who act. |
| Self-reported AI attribution | New pipeline where the buyer names an AI assistant ÷ all new pipeline | The revenue line, including buyers who asked an assistant and then typed your URL. |
Bing Webmaster Tools now reports citation share per grounding query for Copilot and Bing's AI answers, defined the same way. Many tools also roll these metrics into one AI visibility score. I don't report blended scores. If you use one, keep the method fixed, or the trend moves whenever the vendor changes the formula.
03Building the prompt panel
Most bad AI visibility numbers come from a bad panel. My rules:
- Size. Start with 50 to 150 prompts: several per topic, few enough to review by hand when a number moves.
- Intents. Category, comparison, problem and brand prompts in buyers' own words, pulled from sales calls, support tickets, searches and community threads.
- Engines. The ones your buyers use. The Commas read covered ChatGPT and Google AI Overviews. At Pipe17 I tracked referrals from ChatGPT, Perplexity, Gemini and Claude.
- Runs. Answers change from run to run, so run each prompt more than once per cycle and report rates. Hold settings constant: same location, a clean session, search on.
- Cadence. Weekly for the panel, monthly for traffic and pipeline. I run a weekly AEO job and a monthly LLM-referral report.
- Fixed. New prompts go in as a new cohort, reported beside the original. Swapping prompts in place breaks the trend.
- Branded and unbranded apart. Prompts with your name in them mention you almost every time. Use them for accuracy and sentiment and leave them out of mention rate.
Count strictly
Write the counting rules down before the first run.
- A mention is your brand name in the answer text.
- A citation is your URL shown as a source for the answer, as a link chip or a numbered reference.
- A link chip with no mention of your name in the text is a citation only.
- A URL that only shows up in a related or more-links list counts as neither.
- A cited third-party page that names you goes into source mix. It's a mention only if the answer text names you too.
04An AI visibility audit in a day
An audit is the first panel run plus the checks around it, and it fits in a working day.
- Write the fact sheet.Name, category, who it's for, pricing model, key numbers and main competitors. Every accuracy check runs against it.
- Draft the prompts.40 to 60 to start, across the 4 intents, with branded prompts marked.
- Run the engines.ChatGPT with search on, Google AI Overviews and AI Mode, Perplexity and Gemini, with identical settings. Save every answer and cited URL.
- Score the answers.Mentions, first position and citations, using the counting rules above.
- Classify the sources.Tag each cited URL by owner and type: your pages, competitors, lists, review sites, communities, documentation. The most-cited third-party pages become the outreach list.
- Check accuracy.Read each branded answer against the fact sheet and log every error with its likely source.
- Check access.Robots.txt and firewall rules for OAI-SearchBot, PerplexityBot, Claude-SearchBot and Bingbot, plus the raw HTML of your top pages.
- Write the readout.The metrics by topic, 10 fixes ranked by buyer intent and the date of the next run.
To improve AI search visibility from there, fix access first, then the pages closest to a buying decision, then the third-party sources. The page rules are in how to rank in ChatGPT.
05AI visibility tracking tools
There are 3 ways to run an AI visibility tracker, plus free first-party data to read beside any of them.
| Option | Good for | Watch for |
|---|---|---|
| Dedicated AI visibility platforms | Large panels across engines, competitor benchmarks and stored answers. I used Profound at Commas for the 10,560-answer read. | Whether you can export raw answers and citations, set your own prompts and see how mentions are counted. |
| SEO suites with AI tracking | Teams already working in Semrush or Ahrefs, which have both added AI visibility tracking. | Prompts generated from keyword lists can drift from how buyers ask. |
| A do-it-yourself panel | Full control: model APIs with web search plus a SERP API for AI Overviews, run by an agent on a schedule. | API answers can differ from what a logged-in user sees. Spot-check a sample in the apps. |
| Free first-party data | GA4, Bing Webmaster Tools' AI Performance report and Search Console's generative AI report for AI Overviews and AI Mode. | Bing and Google report only their own AI surfaces, and GA4 only sees the visits. |
My rule: if a tool can't export every answer and citation, I can't audit its number. My weekly AEO run is a scheduled agent job covering the panel, citation changes and the next pages to ship. The setup is in how I run growth marketing on Claude Code and Codex.
06GA4 setup for AI referral traffic
GA4 now has an AI Assistant channel in its default channel group. Google's documentation names ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples, doesn't publish the full list and excludes AI Overviews and AI Mode. I still build my own channel so I control the definition.
- In Admin, under Data display, open Channel groups and click Create new channel group. It starts as a copy of the default.
- Add a channel named AI assistants with one condition, Source matches regex:
.*(chatgpt|openai|perplexity|gemini[.]google|claude[.]ai|copilot|deepseek|grok).* - Reorder the list so it sits above Referral. A session lands in the first channel it matches.
- Save the group. Custom channel groups apply to past data, and a standard property can have 2.
- Each quarter, check Referral and Unassigned for AI sources the pattern missed.
Match on source because OpenAI says ChatGPT adds utm_source=chatgpt.com to the links it sends. A session that keeps the tag but loses the referrer can arrive with no medium and fall into Unassigned. A source rule still catches it.
Know what GA4 can't see. Clicks from AI Overviews and AI Mode count as Google organic, so read their impressions in Search Console's generative AI report. App visits with no referrer land in Direct, which is why the self-reported field matters.
07Reporting it next to pipeline
AI brand visibility earns budget when it sits next to revenue. My readout has 2 halves.
- Weekly, from the panel. Citation share and mention rate by topic and engine, answer position, new accuracy errors and pages shipped.
- Monthly, from analytics and the CRM. AI referral sessions and conversions, plus new pipeline and wins where the buyer named an AI assistant. Pass the session source into the CRM with every form fill so pipeline can be grouped by it.
At Pipe17, rebuilding attribution around what buyers said, with monthly referral tracking, showed that 32% of new-business wins came through AI assistants, revenue the CRM had credited to paid search. Pipeline is the number leadership acts on, so it goes on the same page as the panel.
My decision rules: a topic with low citation share and high deal value goes to the top of the page list. If citation share rises but referrals and self-reported mentions don't move, check that the prompts match how buyers ask before shipping more pages.
08Mistakes that make the numbers lie
- Screenshots as evidence. One run of one prompt is an anecdote.
- Loose counting. Link chips scored as mentions, or branded prompts mixed into mention rate.
- Swapped prompts. Changing the panel mid-trend and calling the jump a result.
- Trusting the default channel. Using GA4's AI Assistant channel without checking which assistants it catches.
- Visibility without revenue. A score with no pipeline next to it.
FAQQuestions people ask
- What is AI visibility?
- AI visibility is the share of AI-generated answers in your category that name your brand or cite your pages, across assistants like ChatGPT, Perplexity, Gemini and Google's AI Overviews. It's measured with a fixed panel of buyer prompts and judged by the traffic and pipeline it sends.
- How do you measure AI search visibility?
- Run a fixed panel of real buyer prompts on a schedule across the engines your buyers use and record mentions, citations and the sources cited. Report mention rate and citation share by topic, then add AI referral sessions and self-reported AI attribution to tie it to pipeline.
- What is an AI visibility audit?
- A one-time baseline of how AI assistants describe and cite your brand: a prompt panel run across engines and scored for mentions, citations, accuracy and sources, plus checks on crawler access and AI referral tracking. It ends with a ranked list of fixes.
- What is an AI visibility score?
- A single number some tools use to blend metrics like mention rate, citation share and position. The formula differs by vendor, so report the components, and if you use a score, keep the method fixed.
- What are the best AI visibility tools?
- Dedicated platforms such as Profound are built for it, and SEO suites such as Semrush and Ahrefs have added AI visibility tracking. A do-it-yourself panel run through model and SERP APIs gives the most control. The best tool is the one that lets you fix the prompt list and export every answer and citation.
