How to measure your visibility in AI search answers
Measure AI visibility from four sources at once: a fixed panel of prompts you test by hand each month, a referral channel for AI tools in your analytics, crawler activity in your server logs, and the one first-party citation report that exists, Bing's. None of them is complete. Search Console counts AI Overviews and AI Mode inside ordinary Web search data with no filter to separate them, most AI apps give site owners no report at all, and the same prompt can cite different sources each time it runs.
That means AI visibility is something you sample, not something you read off a dashboard. Keep the method fixed, look at trends over several months, and record what the answers say about you as well as whether they link to you.
Build a prompt panel you can repeat
Write 20 to 40 prompts and keep them fixed for at least two quarters. Mix three kinds:
- Category prompts, the way a buyer would ask: "which accounting software suits a two-person design studio in Canada".
- Problem prompts: "why does my sourdough collapse after the second rise".
- Brand prompts: "what is [your brand]", "[your brand] vs [competitor]", "is [your brand] legit".
Run each prompt in every engine you care about (ChatGPT with search, Perplexity, Gemini, Copilot, Claude with web search, Google AI Mode, plus the AI Overview on a normal Google search). Use a fresh or temporary session with memory off, note your location, and run each prompt twice. Log the results in a sheet like this:
date,engine,prompt_id,run,cited_us,our_url,mentioned_us,competitors_cited,answer_accurate,notes
2026-10-01,perplexity,P07,1,yes,/pricing/,yes,"brandx.com;review-site.com",yes,
The "answer_accurate" column matters. Being mentioned with the wrong price or an outdated feature list is a problem you can often fix on your own pages.
Give AI referrals their own channel
In GA4, AI referrals are mixed into the general Referral channel by default. Create a custom channel group under Admin, Data display, Channel groups, add a channel called "AI assistants" with the condition "Session source matches regex", and move it above Referral so it catches those sessions first.
chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai
Then read the landing pages report for that channel. The pages that receive AI visits are the ones being cited and clicked, which tells you more than the total. Two caveats: many app clicks arrive with no referrer and land in Direct, and clicks from AI Overviews and AI Mode arrive as ordinary google / organic traffic, so they never reach this channel at all.
What Search Console can and cannot show
Google's documentation says traffic from its AI features is included in the Search Console Performance report under the Web search type. There is no AI Overviews or AI Mode filter, and Google's help pages say all links inside an AI Overview share the overview's single position, which makes average position hard to interpret for those queries.
You can still find proxies:
- Long queries. AI Mode encourages conversational questions, so a query filter such as
^(\S+\s){7,}\S+$isolates eight-word-plus queries for separate tracking. - Click-through changes. A query whose impressions stay flat while clicks fall often has a new AI Overview above it. Search it yourself to confirm.
- Page comparisons. Group pages by type and compare their click-through over time.
Bing is the exception. Its AI Performance report in Bing Webmaster Tools shows citation counts for Copilot and Bing's AI summaries, along with sample grounding queries. The Copilot page explains how to use it.
Use server logs as the early signal
Before a page can be cited, a search agent has to fetch it. Logs show two kinds of activity: index crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Bingbot, Googlebot) discovering pages, and user-triggered fetchers (ChatGPT-User, Perplexity-User, Claude-User) reading a page because someone's question needed it right then. A burst of the second kind on one URL is often the first sign that it is being used in answers.
grep -E "OAI-SearchBot|ChatGPT-User|PerplexityBot|Perplexity-User|Claude-SearchBot|Claude-User" access.log \
| awk '{print $7}' | sort | uniq -c | sort -rn | head -20
User-agent strings can be faked, so check the IP addresses against the ranges each operator publishes before drawing conclusions. The RankWave AI crawler directory lists each agent's token and purpose.
Automated trackers, including ours
Third-party tools now run prompt panels for you across several engines. They are sampling too, so ask how many runs they make per prompt, from which locations, logged in or out, and how they handle answers that change between runs. A tool that reports a precise "share of voice" from one run per prompt is overstating what it knows.
Hydrogen SEO's AI visibility and citation tracking is coming soon and is not available yet. Until it is, the manual panel above, the analytics channel and your logs give you a workable picture for the cost of an hour or two a month.
Common questions
Can I see AI Overview clicks separately in Search Console?
No. Google counts AI Overview and AI Mode traffic inside the Web search type with no separate filter. Manual checks and click-through trends are the practical workaround.
Why does some ChatGPT traffic show up as direct?
Clicks from apps and copied links often carry no referrer. ChatGPT adds a utm_source tag to many of its links, but not every visit keeps it, so some AI traffic always lands in Direct.
How often should I run my prompt panel?
Monthly is enough for most sites. Run each prompt at least twice per session, because single answers vary, and compare months rather than days.