· via dev.to (home feed)
How SEOs estimate the clicks Google's Search Console AI report doesn't show
Google's Search Console report for AI Overviews and AI Mode lists impressions but no clicks, CTR or queries. A dev.to walkthrough joins that export with API data to estimate which pages are losing clicks.

What Google actually ships
In June 2026 Google added a Search generative AI performance report to Search Console, fulfilling a request SEOs had made ever since AI Overviews arrived. According to a technical walkthrough published on dev.to, the report covers impressions in AI Overviews, AI Mode and generative AI features in Discover, segmented by page, country, device and date. What it does not contain is the number most site owners want: clicks. CTR and query data are missing too.
The post adds that, as practitioners most recently re-checked in August 2026, the report is not exposed through the Search Analytics API or the BigQuery bulk export either. It lives only in the Search Console interface. Google has said further metrics are coming, but has given no timeline.
The core idea
The workaround depends on one behavior: impressions from AI features are also counted inside the ordinary Web search type. Each page therefore has two knowable numbers: total Web impressions from the API, and AI feature impressions from the interface export. Dividing the second by the first yields a per-page ratio showing how much of a page's search presence involves generative AI.
The method then asks a comparative question: over the same window, did pages with high AI exposure lose CTR faster than pages with low exposure? That proves nothing on its own, the post concedes, but it is currently the only quantitative angle on the problem, and it produces a ranked shortlist of pages worth examining by hand.
The workflow
Step one is the UI export. In Search Console, open Performance, choose the generative AI report (the label varies slightly by account), go to the Pages tab, set the date range and export a CSV. The post flags two cautions: data reportedly starts around May 18, 2026, and nothing earlier was backfilled, so ranges should begin after that date. The same date range must then be used in step two, because mismatched windows are, in the author's experience, the most frequent mistake when joining two exports.
Step two pulls Web performance data for the same property through the Search Analytics API, using a service account with read-only access and a short Python script built on google-api-python-client, google-auth and pandas. The script fetches page-level clicks, impressions and average position for both a current period and a prior one, and the post recommends comparing against the same dates a year earlier to strip out most seasonality.
The script merges the three datasets and applies filters: pages need at least 500 impressions in both periods, and any page whose average position moved by more than one spot is dropped. Pages are grouped into exposure bands — under 10%, 10–40%, 40–70%, and above 70% — and median CTR change is compared across them. A rough estimate of lost clicks is also computed per page: the gap between the clicks it received and what it would have received at the previous period's CTR on current impressions.
Reading the results
In the post's illustrative output, median CTR declines grow steeper as exposure rises — from roughly 0.1 percentage points in the under-10% band to nearly 1.4 points above 70%, with position held stable. That gradient, where bigger drops land on more-exposed pages, is the shape that points toward AI features as the likely explanation.
The filters carry most of the weight. Removing pages whose ranking shifted isolates changes that are not simply about position; the impression floor keeps statistically noisy small pages out; and using medians rather than means stops a single viral page from dominating the result.
What the method cannot tell you
The post is blunt about the limits. Clicks on links inside an AI answer are already counted within normal click totals and cannot be isolated. The AI export contains no queries, so there is no way to see which searches triggered an AI feature. And a year is a long time — SERP layouts, competitors and brand demand all change — so the output is framed as a list of pages to investigate rather than a defensible traffic-loss figure for the finance team.
Why it matters
Google's report confirms that generative AI features now touch a measurable share of search impressions, but it stops short of the metric that matters commercially. Until click data arrives, teams asked to quantify AI's effect on organic traffic must derive their own signals, and this join of two exports is a repeatable, month-over-month way to do it. The follow-up work is where the value sits: for each flagged page, search its top queries, read what the AI answer says and whether it links to you, and if a summary fully satisfies the query, consider retargeting the page toward deeper intents — comparisons, implementation detail, pricing — that a summary cannot serve. Watching the exposure ratio itself over time matters too, since a rising ratio means Google is attaching AI features to more of that page's queries.
- #seo
- #google-search-console
- #analytics
- #python
- #ai-overviews