On June 3, 2026, Google launched dedicated generative AI performance reports in Search Console — one for Search, one for Discover (Google Search Central, June 3, 2026: https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports/) — and on August 31, 2026, both the reports and a companion opt-out control were rolled out to all websites worldwide. If you manage a B2B SaaS site, you have two practical questions now: what do the numbers in that report actually mean, and what should you do with the new toggle? This piece answers both, and then covers the part almost every “where is the button” guide skips — why this report’s numbers will not match the AI numbers you get from anywhere else, and the gap none of the official dashboards close.
This is the companion to our full AI traffic measurement setup (GA4 + GSC + Bing, step by step). That article is the setup guide; this one is the interpretation guide. They are not duplicates — read the setup one first if the three sources are not configured yet.
What the report actually measures
Per Google’s help center documentation (verified 2026-10-19: https://support.google.com/webmasters/answer/16984139), the Search report covers two features right now: AI Overviews and AI Mode, and Google says the list will grow over time. Experiments in Search Labs are explicitly excluded — those numbers live nowhere, and you should not expect to see them.
The metric is impressions, and Google defines it precisely: “how many times links to your site were shown to a user in a generative AI feature on Google Search.” Three properties of that definition matter when you read your own numbers:
- An impression is an appearance, not an engagement. The user may have seen the Overview, skimmed it, or closed it. The report has no clicks, no positions, no citations, and no way to distinguish a link the user read from one rendered off-screen.
- The chart counts per-user appearances, not links. Google’s docs note that if two results from the same site appeared in one generative AI feature, the chart’s total counts them as a single impression. The table’s per-page aggregation works differently. This is why Google’s docs warn that chart totals and table totals can differ — same data, different aggregation. Do not treat the gap as a data error, and do not build anything on one view without checking the other.
- Data has the usual Search Console constraints. The 1,000-row limit, time ranges and Pacific Time date handling from the standard performance report all apply, and the newest data is preliminary.
The dimensions — pages (assigned to the canonical URL, after redirects), countries, dates, devices — plus a search-type filter (text web vs. multimodal) are the same machinery as the performance report, which is why it is fast to read once you know these caveats.
What an impression is for, operationally: it tells you which of your pages Google’s AI features actually feature. For a B2B SaaS site that list is short — usually category pages, comparison pages, a handful of guides. That list is your AI-feature exposure map. Record the monthly total and the per-page split, and read single-month moves with care: Google changes which queries get AI Overviews independently of anything you do, so a page’s AI impressions can jump or crater without a content change.
The opt-out toggle, decided
The control sits at Settings → Search generative AI in Search Console, and rolled out globally alongside the reports on August 31, 2026 (Google Search Console Help: https://support.google.com/webmasters/answer/16908024). Per Google’s documentation:
What it covers. The toggle manages your site’s inclusion in exactly three places: AI Overviews, AI Mode, and generative AI features in Google Discover. Google says the list may be updated over time. Notably, the Gemini app is not in the covered set — TechTimes’ coverage of the control’s June 2026 release states that the toggle “explicitly excludes the Gemini app” and frames that exclusion as the central trade-off publishers weigh (TechTimes, June 8, 2026: https://www.techtimes.com/articles/317979/20260608/google-ai-overviews-opt-out-hits-search-console-gemini-excluded-publishers-weigh-trade-offs.htm). Google’s own scope statement — the control covers “generative AI features on Google Search” — is consistent with that reading, but Google does not publish a separate document spelling out Gemini-app exclusion, so treat the coverage’s wording as the specific claim.
What opting out actually does. Your site’s content is prevented from being visible in those three features — including being linked — and, per the same doc, “content crawled from your site won’t be eligible to be used as an input to generate an AI response or preview” in them. You will receive no impressions or traffic from those features. It generally takes one to two days to propagate, longer for cached content.
What opting out does NOT do. Four boundaries, all from Google’s doc, that separate this toggle from the broader “keep my content out of AI” wish list:
- It does not affect the rest of Google Search. You keep normal rankings; Google states the control “isn’t used as a ranking or inclusion signal affecting other parts of Search.”
- It does not stop AI training. To limit model training you control the Google-Extended crawler in robots.txt — a separate, documented mechanism.
- It does not stop your content from helping Google Search as a whole or from powering the Gemini app, which sits outside the control’s scope.
- It does not override other participation choices (Merchant Center, Google Ads) or block content from Google Search entirely — that is
noindex’s job.
There is also a property-inheritance mechanic worth knowing before you touch the setting: URL-prefix properties inherit the control from their closest parent property that set it manually, up to the top-level domain. If you run multiple properties, changing the toggle at one level may be overridden at another.
The decision, for a B2B SaaS site: in almost all cases, do not opt out. The data-backed argument runs through the AI Overviews CTR evidence we broke down in Ahrefs says −58%, Seer says −61%: those studies show that where AI features appear, clicks to the top organic results fall sharply — so opting out does not protect clicks you were getting. But the same evidence set contains the counterweight: Seer’s observational finding that brands cited in the AI Overviews earned 35% more organic clicks than brands that were not (observational, not causal — the authors say so). Being absent from the feature removes you from the cited set entirely. The rational default for a site whose buyers use Google is to stay in, earn the impressions, and work to be cited in them. The honest cases for opting out are narrow — for example, a site whose content would be mangled out of context in an Overview, or an explicit brand-policy decision — and those deserve a documented decision, not an accidental one.
Why your three dashboards disagree
Once the reports exist, the predictable failure mode is reconciliation: the GSC generative AI report, the Bing Webmaster Tools AI Performance report, and your GA4 AI-referral rows will not agree. That is not a bug — the three measure three different things, and conflating them is how founders reach wrong conclusions about AI visibility.
| Dashboard | Whose surface | What it counts | Window/caveat |
|---|---|---|---|
| GSC generative AI report | Google Search’s AI features (AI Overviews, AI Mode; Discover is a separate report) | Impressions only — appearances, per Google’s definition above | Preliminary data at the edge; chart vs. table aggregation differs |
| Bing WMT AI Performance | Bing’s own AI features (public preview since Feb 2026, Topics Portfolio beta Oct 2026: https://www.bing.com/webmasters/help/ai-performance-9f8e7d6c) | Bing’s own AI feature performance metrics | It is not a ChatGPT report. ChatGPT runs on Microsoft’s models but is not a Bing surface and does not report into this. |
| GA4 AI-referral rows | Whatever surface the user’s browser reports as referrer | Sessions that arrived — the tail of the funnel, after a citation, a displayed link, and a click | Referrer parameters are frequently dropped; this is a floor, not a census |
Three concrete mismatches to expect:
- Impressions vs. sessions will look like a scandal. Google can show your site in an AI Overview a million times while GA4 records double-digit AI referral sessions from the same month. Neither is wrong: the first counts appearances, the second counts arrivals, and the published evidence says the arrival rate off AI features is low. If you treat impressions as a traffic forecast, you will be permanently surprised.
- GSC and Bing cover non-overlapping surfaces. Your GSC numbers say nothing about Bing, and your Bing numbers say nothing about Google or about ChatGPT. The only overlap between the two official reports is the small slice of users who search on both engines.
- Definition drift. “Generative AI feature” is a moving list on both platforms — Google has said it will update its list, and Bing’s report is in preview with beta components. A number that looks comparable this month can be measuring a different feature set next quarter. Compare each dashboard against itself, month over month, and stop comparing across dashboards.
The setup and the per-source detail are in the measurement guide; the point here is that disagreement is structural, and the “which number is real” question has no single answer because the three tools are not measuring the same quantity.
What’s still invisible
Walk through what the official stack does not show, in the order a B2B founder notices:
- Citations in ChatGPT, Perplexity, Gemini and Claude answers. The GSC report covers Google Search’s AI features. None of the official tools measures whether your brand appears when a buyer asks ChatGPT “best [category] for [use case]” — which is where a meaningful share of B2B evaluation now happens. The Ahrefs data on what ChatGPT cites (our analysis of the most-cited domains) is cross-brand and cross-category, not per-brand and per-query.
- Mentions without links. A brand can be named in an AI answer with no URL — an impression the GSC report cannot count, because nothing was shown on your site, and a GA4 session it will never produce.
- The queries themselves. The report groups by page and dimensions; it does not hand you the query list that triggered the features the way the standard performance report does. You are seeing where, not exactly for what.
- Outcome quality. Cited, mentioned, or absent; favorable or not — no official tool scores the content of the answer.
For a B2B SaaS company, the first and fourth of those are the ones that cost deals. The measurement gap is what our methodology exists to close: run your money queries across ChatGPT, Perplexity, Gemini and Claude on a schedule, capture the answers verbatim, score citation presence and wording, and diff month over month — a per-brand, per-query, per-surface number that no official dashboard can produce.
Try the free quick check
If you want to see what lives outside the dashboards before you decide anything: we run your brand’s money queries across ChatGPT, Perplexity, Gemini and Claude and send back a snapshot scorecard — cited, mentioned, or absent, with what each answer actually said. No setup on your side, no meeting required. Get the free quick check.