Search Console Handed You an AI Impressions Report and a Switch to Delete Yourself From AI Answers

On 3 June 2026, the Google Search Central blog announced Search Generative AI performance reports in Search Console. Around 11 August the report spread widely, and by 13 August far more site owners were seeing it — while Google's official line still described it as a test with a subset of websites. Data starts on 18 May 2026 and there is exactly one metric: impressions. No clicks, no CTR, no queries, no position, and no way to tell whether an impression came from AI Overviews or AI Mode.

The same release ships a second thing: a switch in Settings, labelled "Search generative AI", that lets you keep your content out of AI Overviews, AI Mode and generative features in Discover. Google says turning it off does not affect ranking outside those features — but it does mean forfeiting whatever traffic and visibility they were sending you. The timing is not accidental. AI Mode passed one billion monthly active users at Google I/O 2026 and now reaches close to 200 countries. Semrush analysed more than 600,000 US desktop keywords and found that between November 2025 and April 2026, the share of commercial-intent SERPs carrying an AI Overview grew 71%, with finance up 231%. Generative surfaces are the front door now, not a garnish.

Set it next to the competing category and the picture sharpens. For the past year, Profound, Peec AI, Scrunch and Otterly have sold one core promise — how often your brand shows up inside AI answers — while Ahrefs and Semrush chased them with Brand Radar and the AI Toolkit. Google just commoditised the entry tier of that market by giving away first-party impression counts. But it stopped at the point most favourable to itself: you can see that you were cited, and you cannot calculate what the citation was worth.

For small and mid-sized businesses in Taiwan this is a familiar trap — an irreversible decision demanded without a denominator. This piece covers three things: what the report is actually counting, what the switch really costs you, and how to instrument the gap between AI impressions and on-site behaviour without buying a single SaaS subscription.

What shipped, with the full numbers

Announced 3 June 2026, Search Console Beta, with separate reports for Search and for Discover.
Data begins 18 May 2026; no backfill to earlier periods.
Available dimensions: page URL, country, device, and a time series that can be broken down to the hour.
Missing dimensions: clicks, CTR, query strings, average position, and any split between AI Overviews and AI Mode.
Impression definition: one impression each time a link to your site is displayed to a user inside a generative AI feature; links hidden behind an expandable element count only once the user expands it.
Totals do not reconcile: if two URLs from the same site appear in one generated answer, the property-level chart counts one impression while the page table credits both URLs — so summing page rows will not reproduce the property total.
The control lives in Search Console Settings under "Search generative AI" and is a binary, property-level choice.
Effective dates: the control began applying in Google Search on 17 June 2026, and changes take roughly one to two days to propagate, subject to caching.
What the switch does not do: it does not affect ranking outside generative features, does not stop model training, and does not cover summaries in the Gemini app.
Market context: SE Ranking studied 50,032 commercial keywords and found 14,733 (29.45%) returned a text ad in AI Mode, rising to 53.56% for keywords with CPCs of $10 or more.
Ahrefs compared 150,000 AI Overview keywords against 150,000 without in February 2026 and reported a 58% click decline for top-ranking pages, up from 34.5% measured in April 2025 — though the non-AIO control group also fell over the same window, making AI Overviews an accelerant rather than the sole cause.

What to do this week, by role

Brand owners and decision makers

Do not make the on-or-off call this quarter. You hold impressions and no clicks, so any estimate of what you would lose is guesswork. Ask marketing for a baseline instead: AI impressions, organic clicks for the same pages, and GA4 sessions and conversions for the same pages, plotted on one chart across three months. If you sit in a category with heavy ad penetration in the SE Ranking data — pets hit 72.38% — assume organic citation value is already being diluted.

Marketers and SEO practitioners

Treat the report as a citation list, not a traffic report. Its most useful column is the page URL: that is Google telling you which of your content it considers good enough to ground an answer on, which is a direct instruction on how to restructure the rest. Accept the corollary too — this impression count cannot serve as the denominator for a CTR calculation.

Developers and agencies

This is a shippable engagement. Pull the generative AI report daily via the Search Console API, land it in a warehouse, join it to GA4 and server logs at page level, and deliver a dashboard a client can actually read. A Laravel scheduler plus BigQuery or plain Postgres covers it, and it is worth considerably more than pressing the switch on their behalf.

☐ Check whether the generative AI report has appeared in your Search Console property
☐ Export impressions from 18 May 2026 as your baseline
☐ Overlay AI impressions against GA4 sessions for the same pages
☐ Leave the switch alone until the baseline exists

Tooling compared

Tool	Entry price (2026)	What it measures	Best fit
Search Console generative AI report	Free	First-party impressions in Google generative features, by page, country and device	Baseline for every site; nothing else substitutes for it
Otterly	From ~US$29/mo	Brand mentions and citations inside AI answers	Single brand, smallest viable monitoring budget
Peec AI	~EUR 89–199/mo	Cross-model mentions, competitor comparison, source analysis	Mid-market teams; best depth-to-price ratio
Scrunch AI	From US$250/mo	AI visibility plus agent traffic analytics	Acquired by Sitecore in June 2026; suits existing DXP shops
Profound	Growth ~US$399/mo, enterprise quoted	Large-scale prompt sampling and conversation-level insight	Large brands with dedicated GEO headcount
Semrush AI Toolkit	US$99/mo add-on (Semrush One from ~US$199/mo)	AI visibility bundled with classic SEO	Teams already on Semrush who do not want a second system
Ahrefs Brand Radar	US$199/platform/mo, or US$699 bundled, on top of a base plan	Brand presence tracked across multiple AI platforms	Well-funded teams needing coverage beyond Google

What nobody is telling you

First, the impression number is soft, and it cannot be a denominator. Google states that links behind expandable elements only count once expanded, yet AI Overview citations frequently require a "show more" click to surface at all. Add the aggregation rule — two URLs from one site in a single answer counts once at property level but twice in the page table — and two things follow: your page rows will never sum to your total, and any ratio you compute by dividing GSC AI impressions into GA4 sessions is statistically meaningless.

Second, this looks like groundwork for paid AI placement. SE Ranking already measures ads on 29.45% of commercial AI Mode queries. Making organic citation unmeasurable while reserving the measurable, biddable slot for advertising is a coherent commercial sequence. The switch framed as "giving control back" has the practical effect of persuading a cohort of small site owners — the ones who cannot quantify the loss — to voluntarily exit the organic citation pool.

The no-subscription SMB alternative

The goal is not to replace Profound. It is to close the gap between AI impressions and on-site behaviour. The route:

Schedule a daily pull of the generative AI report through the Search Console API into BigQuery or Google Sheets, so you outlive the 16-month data window.
Use GA4 landing page plus source/medium reporting to watch whether direct traffic and branded search rise on the pages that get cited — AI citations carry no UTM, so this is the only workable proxy.
Filter user agents in Cloudflare Logs or raw server access logs to separate Google-Extended, GPTBot and PerplexityBot crawl frequency and see who is actually reading you.
Build a Looker Studio view that overlays three lines per page: AI impressions, organic clicks, GA4 sessions.
Run a weekly n8n job that fires a fixed set of brand and product prompts and records which pages get cited — a low-cost GEO monitor you own.
Understand that Google-Extended in robots.txt is not the same lever: it governs Gemini-related use, while the Search Console switch governs appearance in AI Overviews, AI Mode and Discover.

FAQ

Does turning the switch off hurt my normal search rankings?

According to Google, no. The control is not used as a ranking signal for results outside generative AI features, and your pages continue to appear in standard blue links and the regular Discover feed. What you lose is impressions and clicks from AI Overviews and AI Mode.

With no click data, how do I estimate traffic from AI impressions?

There is no precise method today. The workable approximation is to isolate your highest-impression pages in the report and compare their GA4 direct traffic and branded search movement, judging by trend correlation rather than absolute numbers. Any tool claiming an exact AI CTR is extrapolating from its own sampling, not from Google first-party data.

Is blocking Google-Extended in robots.txt the same as this switch?

No. Google-Extended governs Gemini-related data use, while the Search Console control governs whether your content appears in AI Overviews, AI Mode and generative Discover features. The switch also does not stop model training and does not cover summaries in the Gemini app.

Why do GSC AI impressions not reconcile with GA4?

Because they measure different events. GSC counts your link being displayed, including cases where the user may never have expanded the citation; GA4 counts someone actually arriving. Layer on the differing aggregation rules for multiple URLs in one answer and the two numbers are structurally incapable of matching.

Should a Taiwanese SMB switch it off right now?

In almost every case, no. Unless you are a publisher monetising through a paywall or licensing and can already demonstrate that AI citation is directly eroding revenue, opting out without click data trades a growing visibility channel for a guess.

My take

The switch is packaged as handing control back to site owners, but control presupposes symmetric information. Google released the impressions, withheld the clicks, and then asked you to make a property-level, all-or-nothing decision with no per-page and no per-feature granularity. You cannot exclude the twenty blog posts that get cited constantly and convert never while keeping your product pages in. It is all in or all out. Without a denominator, that is not agency — it is a transfer of blame.

The pragmatic read is that this report's value is not in the number but in the page list. It tells you which of your content Google considers fit to answer a question with, a signal that used to cost hundreds of dollars a month in SaaS and is now free. For a studio like ScriptWalker, the opportunity is not flipping switches for clients. It is wiring Search Console API, GA4 and server logs into one durable measurement pipeline, so that when the client genuinely has to decide in three months, they are holding data rather than a toggle.

Email：[email protected]
電話：0916-224-047
LINE：@ufv9089p