← back to blog
    Berner SetterwallJune 3, 20266 min read

    Search Console Finally Shows AI Overviews & AI Mode Data — Here's What's In It (and What's Not)

    For two years the single most asked question in SEO has been some version of: "Is my content showing up in AI answers, and is it costing me clicks?" Until now the honest answer was "we can infer it, but Google won't tell us directly."

    That changed this week. Google announced a new Generative AI performance report in Search Console that, for the first time, shows how often your URLs appear inside Google's generative AI surfaces — AI Overviews and AI Mode.

    This is genuinely good news. It's also narrower than the headlines suggest, and there's one detail that matters a lot if you pull Search Console data programmatically the way we do. Here's the full picture.

    What the new report actually shows

    The report is a dedicated view inside the Search Console Performance section, scoped to Google's AI features. At launch it gives you:

    • Impressions — how many times links to your site appeared inside AI Overviews and AI Mode.
    • Pages — which of your URLs surfaced, grouped by canonical/final URL.
    • Countries — where the searches originated.
    • Devices — desktop, tablet, mobile.
    • Dates — with day, week, and month granularity (Pacific Time), same as the standard report.

    You get the familiar chart-plus-table layout, and you can export both.

    According to Google's help documentation, it covers AI Overviews and AI Mode and explicitly excludes Search Labs experiments, since those are still in flux. It's rolling out to a subset of property owners first (early reports point to the UK) and expanding globally over time. Alongside it, Google is also testing a control to opt your content out of AI Mode and AI Overviews — worth knowing exists, even if most sites won't touch it.

    What it leaves out — read this part

    Three limitations define how useful this report is in practice:

    1. Impressions only. No clicks. At launch there is no click or CTR data for the AI surfaces. So you can see that you were cited in an AI answer, but not whether anyone clicked through. For a feature whose entire controversy is "AI answers eat your clicks," the click half of that equation is exactly what's still missing. Use it to measure visibility, not traffic.

    2. These impressions aren't new — they were already in your "Web" totals. This is the one that trips people up. The new report doesn't add a separate search type; it surfaces a slice of data that was already counted inside the existing "Web" results in your standard Performance report. AI Overviews and AI Mode impressions have been folded into your Web numbers all along. The new view just lets you see that slice — it does not let you subtract it cleanly, and there's no "exclude AI" toggle on the main report. If you've been treating your Web impressions as "classic ten blue links," they never were.

    3. It's a report, not a data feed (yet). Which brings us to the part most coverage skips.

    The API gap: it's not in searchanalytics.query

    We monitor over 130 million Search Console impressions a month across our customer base, and every one of those flows through the Search Console Search Analytics API (searchanalytics.query) — the programmatic endpoint that powers BigQuery pipelines, dashboards, and our Search Console MCP server.

    So the first thing we did when this launched was check the API. The verdict, as of this writing: the Generative AI data is not in the API.

    The type field on searchanalytics.query still accepts only the long-standing values — web, image, video, news, discover, googleNews. There is no aiOverview or aiMode type. There's no dedicated searchAppearance value for it either, and no separate endpoint. The report lives entirely in the Search Console UI.

    That means today you cannot pull your AI Overviews / AI Mode impressions programmatically — not into BigQuery, not into a scheduled report, not through any MCP. You can eyeball it in the dashboard and export a CSV by hand, and that's it.

    This is a normal pattern for Google: the UI report ships first, the API catches up months later (the same thing happened with the Discover and News reports). But it's worth being precise about, because a lot of tools will quietly imply they're tracking your "AI visibility" via the API when the underlying data simply isn't exposed there.

    What this means for the Cogny MCP — and when it lands

    We'd love to ship a get_generative_ai_performance tool today so you could just ask Claude "how did my AI Overview visibility trend last month?" and have it query your own data. We can't, honestly, until Google exposes the dimension. Shipping a tool that returns nothing — or worse, silently returns blended Web numbers dressed up as "AI data" — is exactly the kind of fake-feature behavior we refuse to put in front of a customer.

    So here's the plan, stated plainly so we're held to it:

    • We've flagged the exact extension point in the MCP server (get_search_analytics in our cogny-mcp-proxy Search Console server). The moment Google ships an API surface for this data, wiring it in is small:
      • if it arrives as a new type enum value (e.g. aiOverview), it's effectively a one-line addition to the existing tool;
      • if it arrives as a new searchAppearance dimension value, it's a documentation update — the tool already passes dimension_filter_groups straight through;
      • if it arrives as a dedicated endpoint, it's a small new read-only tool following the same pattern as get_search_analytics.
    • When that ships, every Cogny user with a connected Search Console property gets it through Claude immediately — no dashboard-hopping, no CSV exports.

    Until then, the most useful thing our agents can do is what they already do well: read your full Performance data, correlate Search Console with Bing Webmaster Tools, GA4, and ad spend, and flag the Search Console patterns that actually predict AI citations — high-impression, low-CTR queries being the classic fingerprint of an AI answer absorbing your click.

    A welcome nudge — and Bing has been here a while

    Credit where it's due: this is a real improvement in transparency, and it's the most concrete signal Google has given site owners about generative-AI visibility to date.

    It also brings Google a step closer to what Bing Webmaster Tools has been offering. Bing has been more forthcoming about AI- and Copilot-influenced search data, and our MCP already pulls Bing's traffic and keyword stats today — which is why we keep arguing that serious GEO measurement is a cross-engine exercise, not a Google-only one. Google catching up here is good for everyone; it just confirms the direction Bing pointed at first.

    The honest state of play in June 2026: you can finally see your AI Overviews and AI Mode impressions in Google's dashboard, you still can't see the clicks, and you can't yet pull any of it through the API. We'll close that last gap in the MCP the day Google opens it — and we'll write the follow-up the moment we do.

    Want your Search Console and Bing data read, correlated, and explained by an AI agent instead of a dashboard? Talk to us — we'll do it live in your first session.