Skip to content

Module 08 · GEO · Guide 4 of 4

Measuring AI Visibility

Measure AI search visibility in 2026: Search Console's generative AI report, Bing AI Performance, GA4's AI Assistant channel, server logs and prompt panels.

By Updated 5 min readAdvanced

What you'll learn

  • Read Search Console's Generative AI performance report and know what it leaves out.
  • Use Bing Webmaster Tools AI Performance to see Copilot citations and grounding queries.
  • Segment AI referrals in GA4 with the native AI Assistant channel and fill its gaps.
  • Grep server logs for current AI user agents and interpret the three bot types.
  • Run a prompt panel, then decide whether a paid tracker adds anything.

GEO without measurement is faith-based marketing. Until 2026 the honest answer was "you mostly can't measure AI search". That changed: Google, Microsoft and Google Analytics now ship first-party AI data. Use it first, then add your own logs and prompt panels, and only then consider paid trackers.

Layer 1: First-party data (free, start here)#

Search Console: Generative AI performance report#

Google's report (Search Console > Performance > Generative AI) launched for UK sites on 2026-06-03 and rolled out worldwide on 2026-08-31.

What it showsWhat it does not show
Impressions: how many times links to your site appeared in AI Overviews or AI ModeClicks
Breakdown by page, country, device and date (daily, weekly or monthly)Queries
Search type: "Web: text-based" and "Web: multimodal"Position
A separate report for generative AI features in DiscoverSearch Labs experiments

Two counting rules matter: several results from your site in one AI response count as one impression, and the report needs a minimum volume before it shows data.

AI traffic is still counted in the normal Performance report too, under the Web search type. Google's methodology since June 2025: a click on an external link in AI Mode is a click, a follow-up question is a new query, and all links in an AI Overview share a single position. So you can see AI impressions on their own, but AI clicks stay blended with classic clicks.

Bing Webmaster Tools: AI Performance#

Released in public preview on 2026-02-10, it reports citations across Microsoft Copilot, AI summaries in Bing and select partners:

  • Total citations and average cited pages per day
  • Grounding queries: a sample of the phrases the AI used when it retrieved your content
  • Page-level citation activity and trends

Bing added intents, topics, citation share and competitor comparison in June 2026. There is no click data, and it does not cover ChatGPT, Perplexity or Google. Grounding queries are the closest thing any engine offers to "which AI queries cite me", so check them monthly.

GA4: the AI Assistant channel#

Since 2026-05-13, GA4's default channel group includes AI Assistant: the medium is set to ai-assistant when the referrer matches Google's list of assistants, which its documentation illustrates with ChatGPT, Gemini, DeepSeek, Copilot and Grok. It is not retroactive. Two gaps to fill:

  • Google's own AI features are excluded. Clicks from AI Overviews and AI Mode arrive as google / organic, in Organic Search.
  • Not every assistant is on the list. Check Referral for the rest and add a custom channel or exploration filter:
Referrer hostnames worth matching (regex)
chatgpt\.com|perplexity\.ai|claude\.ai|copilot\.microsoft\.com|gemini\.google\.com|chat\.deepseek\.com|grok\.com|meta\.ai

ChatGPT also adds utm_source=chatgpt.com to links it cites, so those visits are identifiable even when the referrer is stripped. Some AI visits still arrive with no referrer and land in Direct; treat AI referral numbers as a floor.

For context, AI referrals are still a small slice of traffic: Conductor measured about 1% of website visits across 13,770 domains (May to September 2025). Track the trend, landing pages and conversions, not just the volume.

Layer 2: Server logs (free, needs access)#

Logs show which AI bots read which pages, and they separate the three bot types:

  • Training crawlers (GPTBot, ClaudeBot, meta-externalagent, Amazonbot): your content may feed models; no direct citation value
  • Search and answer crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Amzn-SearchBot, meta-webindexer, Applebot): these drive citations
  • User-triggered fetchers (ChatGPT-User, Claude-User, Perplexity-User, Google-Agent, meta-externalfetcher): a person asked an assistant about your page right now. This is the closest log signal to "an AI answer is using me"
AI bot hits by page (combined log format)
grep -iE "gptbot|oai-searchbot|chatgpt-user|claudebot|claude-searchbot|claude-user|perplexitybot|perplexity-user|google-agent|meta-externalagent|meta-webindexer|amzn-searchbot|applebot" access.log \
  | awk '{print $7}' | sort | uniq -c | sort -rn | head -20

Two caveats. Google-Extended is a robots.txt token, not a crawler, so it never appears in logs; Google's AI features use ordinary Googlebot. And user agents can be spoofed, so verify important bots against their published IP ranges. The full, maintained list is in AI Crawlers & Bot Access.

Expect crawling to dwarf referrals. In July 2025, Cloudflare measured roughly 1,091 crawls per referral for OpenAI and 38,065 for Anthropic. Logs prove you are being read, not that you are being cited.

Layer 3: Prompt panels (free, manual)#

The AI equivalent of rank tracking: ask the engines your important questions and record whether you are cited or named.

  1. Take your top 20 to 50 commercial and informational queries from your keyword map, plus the grounding queries Bing reports
  2. Rephrase each as a person would ask a chatbot ("what's the best X for Y?", "how do I...?")
  3. Monthly, run each prompt in ChatGPT (with search), Perplexity, Gemini, Copilot, Claude, and Google AI Mode
  4. Record: were you cited (linked), mentioned (named, unlinked) or absent? Who was cited instead?

Layer 4: Paid trackers (optional)#

A fast-changing category of tools (Profound, Peec, Otterly, and AI visibility features in Semrush, Ahrefs and others) runs prompt panels at scale and trends your share of answers. They save time; they do not see inside the engines. Google's May 2026 guide warns to "be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics", because no third-party tool has access to its ranking or AI systems. Treat their numbers as sampled estimates, and evaluate vendors when you buy; the market shifts quarterly.

Share of voice and brand signals#

Zoom out from single prompts:

  • Competitive share of voice: across your prompt panel, what share of answers cite you versus each competitor? This is the GEO equivalent of average position, and the number executives understand.
  • Branded search demand: Search Console's branded queries filter (available to eligible sites since 2026-03-11) shows whether AI mentions are turning into people searching for you, the lagging indicator of mention-without-click value.
  • Citation versus CTR: Seer Interactive (September 2025 data) found pages cited in an AI Overview earned higher organic CTR than uncited pages on the same queries, though both fell year over year. Citation is worth tracking even when clicks drop.

A monthly AI visibility report that works#

1. AI impressions by page                 (GSC Generative AI report, trend)
2. Copilot citations + grounding queries  (Bing AI Performance)
3. AI referral sessions + conversions     (GA4 AI Assistant + custom regex)
4. Search-bot and user-fetch hits         (server logs, by page)
5. Citation rate across prompt panel      (per engine, frequency, trend)
6. Share of voice vs. top 3 competitors   (prompt panel)
7. Branded query volume                   (GSC branded filter, trend)
8. Actions: passages to retrofit, gaps where competitors are cited and you are not

The last line matters most: every prompt where a competitor is cited and you are absent is a concrete work item, usually a quotability retrofit, a missing page or a bot-access fix. For turning this into a stakeholder-ready report, see SEO Strategy & Stakeholder Reporting.

GEO module complete. The final module, Expert Practice, turns everything so far into professional workflows, starting with the full technical audit.

Check your understanding

0 of 4 answered

  1. 1.Your Generative AI performance report shows a page with 40,000 impressions last month. What can you NOT learn from that report?
  2. 2.In GA4, a visit that clicked a link in Google AI Mode will most likely appear in which channel?
  3. 3.Your logs show thousands of GPTBot and ClaudeBot hits but almost no OAI-SearchBot or Claude-SearchBot. What does that suggest?
  4. 4.You run the same 30 prompts in ChatGPT twice in one day and get different citations. How should you report citation presence?

Sources

Primary documentation this guide was checked against, last reviewed 5 October 2026.

  1. Generative AI performance report (Search) · Google Search Console Help
  2. Introducing Search generative AI performance reports in Search Console · Google Search Central, June 2026
  3. Introducing AI Performance in Bing Webmaster Tools (public preview) · Microsoft Bing, 2026-02-10
  4. [GA4] Default channel group (AI Assistant channel) · Google Analytics Help
  5. Publishers and developers FAQ (utm_source=chatgpt.com) · OpenAI Help Center
  6. The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals · Cloudflare, 2025-08-29
  7. AIO impact on Google CTR: September 2025 update · Seer Interactive, 2025-11-04

Spotted something wrong or outdated? Report itHow we review

Finished reading?

Mark it complete to track your progress. Saved in this browser, no account needed.