Most “track AI search traffic” guides written before mid-2026 fall into one of two camps: GA4 regex workarounds for referrer domains, or Search Console tips. As of June 2026 there are three official sources — GA4 referral data, Google Search Console’s generative AI performance report, and Bing Webmaster Tools’ AI Performance report — and almost no page covers all three, let alone what each one is and isn’t telling you. This is that page.
What each source actually measures (and what they don’t)
Get this table right first, because every other step follows from it:
| Source | It measures | It does not measure |
|---|---|---|
| GA4 referral rows | Sessions that landed on your site from an AI surface (the browser referrer says ChatGPT, Perplexity, Gemini, Claude, etc.) | Searches, impressions, citations, and any AI visit that lost its referrer |
| GSC generative AI performance report | Impressions of your site inside Google Search’s own generative AI features (AI Overviews, AI Mode) | Clicks, positions, citations, and everything happening outside Google Search |
| Bing WMT AI Performance | Your site’s performance in Bing’s own AI features | ChatGPT or any other engine — it counts Bing’s AI features, full stop |
The three are complementary, and none of them answers the question that matters most for B2B SaaS: are you cited in AI answers to your buyers’ money queries? That measurement lives outside all three official sources — more on that at the end.
GA4: referral setup, done properly
Step 1 — Know what the rows look like. AI-surface traffic arrives in GA4 as ordinary referral traffic: the source/medium reads like a referral from the surface’s domain (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and the relevant Microsoft/Copilot domains). Expect small numbers — for most B2B SaaS sites, single to double-digit sessions per month per surface. That is normal, and it is exactly why you track direction, not level.
Step 2 — Segment by referrer domain, and know why. Create an audience (or custom channel group) that groups the AI surface referrers so you can report them as one slice. Segment on the referrer domain deliberately, for one reason that trips people up:
Step 3 — The UTM-drop caveat. When an AI surface surfaces one of your links, the user’s click frequently arrives without your campaign UTM parameters — the link the assistant presents is not your tracking link, and some surfaces rewrite or strip parameters in between. Do not build your AI-traffic definition on UTM values; you will undercount. Referrer domain is the stable signal. There is a second, quieter caveat: some AI referrals arrive with the referrer stripped or routed through an intermediate hop, so they land in (none)/(direct) instead. Treat your GA4 AI-referral number as a floor, not a census — it is the minimum, and it will understate the true total by an amount you cannot see.
Step 4 — Report monthly, one row per surface. Surface, sessions, and month-over-month delta. Resist the urge to add conversion metrics in the first months — the sample is too small for any of them to mean anything, and the honest report is the trend line.
GSC: the generative AI performance report
Google launched dedicated generative AI performance reports in Search Console on June 3, 2026, with separate reports for Search and for Discover (Google Search Central, June 2026: https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports). Per Google’s help center, the reports were rolled out to all websites worldwide as of August 31, 2026 (https://support.google.com/webmasters/answer/16984139). That same date is when the Search generative AI control — the AI opt-out toggle — also reached all websites worldwide (https://support.google.com/webmasters/answer/16908024). It lives at Settings > Search generative AI in Search Console and covers AI Overviews, AI Mode, and the generative AI features in Google Discover. Inclusion is the default; excluding your site removes impressions and traffic from those features, Google says the control is not a ranking or inclusion signal anywhere else in Search, and it is not a training opt-out — that is the separate Google-Extended robots.txt token, covered in the LLM crawlers reference. Exclusions generally take effect within 1–2 days.
What the Search report contains, per the official documentation:
- Impressions only — how many times links to your site were shown to a user in a generative AI feature on Google Search. No clicks, no positions, no citations.
- Features covered: AI Overviews and AI Mode (Google says the list will grow over time). Experiments in Search Labs are excluded.
- Dimensions: pages (assigned to the canonical URL), countries, dates, and devices.
- If you do not see the report, Google says this usually means the site has not received enough impressions in generative AI features yet — or the site has been excluded from Search generative AI features, which you can check in the same settings area.
What to do with it:
- Open the report and identify which of your pages earn AI-feature impressions at all. For a B2B SaaS site, that list is short — usually category pages, comparison pages, and a handful of guides. That list is your exposure map.
- Record the monthly impression total and the per-page split. A page’s AI impressions can move for reasons unrelated to your content (Google changes which queries get Overviews), so read single-month moves with care.
- Cross-reference against the CTR evidence: AI Overviews are associated with lower click-through on the queries they appear on, and we break the published studies down in Ahrefs says −58%, Seer says −61%: what the AI Overviews CTR numbers actually mean. The GSC report tells you where you are featured; the studies tell you what featured-but-not-clicked can mean.
The Discover report is a separate view for Google Discover’s generative AI features; same logic, same impressions-only caveat.
If you want the report itself interpreted rather than just set up — how the impression definition works, why chart totals and table totals differ, what the toggle actually excludes — we wrote a companion piece: Google’s new generative AI report, decoded. This post is the setup guide; that one is the interpretation guide. They are not duplicates.
Bing Webmaster Tools: AI Performance
Bing Webmaster Tools’ AI Performance report went into public preview on February 10, 2026, and a Topics Portfolio beta followed on October 13, 2026, grouping the queries that led to your citations into thematic clusters (Bing’s announcements: public preview, Topics Portfolio beta; Bing Webmaster Tools: https://www.bing.com/webmasters/help/ai-performance-9f8e7d6c).
Two things to keep straight:
- Scope: per Bing’s own blog, the report covers “Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations,” and it measures citations of your pages, not impressions. It is not a ChatGPT report and not a cross-engine report. Microsoft owns the data; it reflects Bing’s family of AI surfaces, not just Bing search.
- Setup: verify your property in Bing Webmaster Tools, connect it to your Search Console account if you want the shared verification, and then read the report the same way you read the GSC one — which pages appear, in what volume, and how the trend moves month over month.
For most B2B SaaS sites the Bing numbers will be small — in the public referral-share data, Bing-adjacent surfaces like Copilot hold a single-digit share of AI chatbot referrals, and Bing’s search traffic itself trails Google’s, as the referral-traffic data covers. Set it up anyway: it is ten minutes of work, it is the only official number you can get for that surface, and a baseline is worth having before you need it.
What the numbers will NOT tell you
Walk through the three sources one more time, in the negative:
- Impressions are not clicks. The GSC report tells you Google showed your page inside an AI feature. It does not tell you the user saw it, understood it, or clicked.
- Clicks are not citations. A referral session from ChatGPT is downstream of a citation that happened in a conversation — and many citations produce no click at all. High citations with low referrals is a normal, healthy pattern.
- None of the three measures citations in chat assistants. GSC covers Google’s surfaces; Bing’s report covers Bing’s; GA4 sees the clicks that survive to your site. Whether ChatGPT, Perplexity, Gemini or Claude names your brand — with a link, without one, or not at all — when your buyers ask the questions that matter to you is visible in none of these official tools.
That last gap is the one your buyers feel. A prospect can ask “what’s the best [category] for [use case]” and never see your name, and no official report will flag it. The way to close the gap is to run the questions yourself, across the surfaces, on a schedule, and score the answers — which is precisely the setup we document in our methodology: a query map of your money questions, a harness that runs them on a schedule, verbatim capture, and a month-over-month diff you can defend.
Try the free quick check
Before you build any of the manual parts: we will run your brand’s money queries across ChatGPT, Perplexity, Gemini and Claude and send back a snapshot scorecard — cited, mentioned, or absent, with what the answers actually said. No setup on your side, no meeting required. Get the free quick check.