Search Console AI data inflates rankings, masks real losses
Google's new Search Console filter flatters your rankings and buries the traffic losses that AI Overviews are already causing.
Key takeaways
- AI Overview impressions are logged at position one in Search Console, artificially improving average ranking metrics.
- Click-through rates fall and sessions erode even as reported rankings hold steady, masking real traffic losses.
- Blending AI Overview data into core performance reports makes the cost of AI answers to organic traffic invisible.
- The fix requires segmenting AI Overview impressions into a separate data stream and excluding them from baseline trend analysis.
- Brands in financial services and policy sectors face the sharpest exposure, as authoritative content is most likely to trigger AI Overviews.
Eight weeks of live data is enough to spot a pattern. Search Engine Journal's analysis of Google's new "Generative AI" Search Console filter reveals a measurement problem that is already distorting how marketers read their own performance: the filter logs AI Overviews impressions against position one by default, inflating average ranking metrics while the clicks those impressions represent never arrive.
The mechanics are straightforward and the consequences are not. When a query triggers an AI Overview, Search Console records an impression at position one regardless of where the brand's organic result actually sits below the fold. Blend those impressions into an unfiltered performance report and average position improves, click-through rate falls, and a marketeer with incomplete context concludes the audience is simply less inclined to click. The real explanation, that traffic is being absorbed by the AI answer above the results, disappears into a statistical artefact.
The position-one fiction
Position-one data has always been the vanity metric most prone to misreading. AI Overviews make it worse. A financial services firm tracking organic performance for, say, a query cluster around pension planning or regulatory compliance may see its average position hold steady at 1.2 while total sessions from organic search quietly erode. The ranking signal says nothing is wrong. The revenue signal, if anyone has connected those pipelines, says otherwise.
This matters disproportionately for institutions where search visibility feeds into reputation rather than direct revenue. A multilateral organisation cited in an AI Overview may never know whether that citation is driving users to its site or simply answering the query and closing the tab. Search Console's new filter, in its current form, cannot distinguish between the two outcomes with any precision. Impressions are logged; intent resolution is not.
Search Engine Journal's review of eight weeks of data also flags the averaging problem: because AI Overview impressions enter the dataset at position one, they pull the mean upward even when conventional rankings are slipping. Any team using average position as a proxy for SEO health is now working with a metric that flatters. The correct response is to segment ruthlessly: filter AI Overview traffic entirely out of core performance baselines and track it as a separate, experimental data stream until the click behaviour is better understood.
What the data is actually telling you
There is a harder implication behind the measurement quirk. Appearing inside an AI Overview is not the same commercial event as ranking first in the blue-link results. For a B2B brand selling a long-consideration product to a procurement committee, a citation in an AI answer may build diffuse awareness. It almost certainly does not replicate the intent signal of a deliberate click to the brand's site. Treating both as equivalent in a blended impression count is not just imprecise; it actively obscures which content is earning visits and which is merely being summarised.
Industrial groups and policy institutions face a particular version of this problem. Their content tends to rank on authoritative, technical queries where AI Overviews are most likely to trigger. The same depth that wins citations also makes a human visit less necessary. If Search Console's blended data makes that dynamic invisible, those organisations will keep investing in content that feeds AI answers without ever measuring whether those answers convert to engaged readers or institutional relationships.
The corrective is not complicated but it requires discipline. Strip the Generative AI filter out of any report used for baseline trend analysis. Build a parallel view that tracks AI Overview impressions and their (minimal) clicks separately. Then watch the gap between the two data streams widen, because it will. That gap is the actual cost of AI Overviews to organic traffic, and right now Google's own tooling is designed in a way that makes it easy to miss.
Brands that notice the gap first and reprice their content investment accordingly will be ahead. Those that let a stable average-position figure reassure them through the next two quarters will find themselves explaining a traffic cliff that their dashboards never predicted.