Diagnose in order, ruling out the simpler and more easily verified explanations before attributing a decline to AI Overviews, because AI Overviews are the most attention-grabbing explanation right now and it’s easy to assume causation before actually checking the alternatives. The correct diagnostic order is seasonality first, other SERP feature changes second, and AI Overview presence third, cross-referenced against actual query-level data rather than a general industry impression.
Step one: rule out seasonality with a year-over-year comparison
Before attributing any CTR decline to a SERP feature, compare the same period against the prior year rather than only looking at recent trend lines. Seasonal demand shifts (a topic that’s naturally less searched at this time of year, a shift in the mix of query types users are entering) can produce a CTR decline that has nothing to do with any SERP layout change. This is standard measurement discipline that predates AI Overviews entirely, and skipping it is the most common way practitioners misattribute a decline. If the decline shows up consistently year-over-year at the same time, and last year showed the same pattern, seasonality is a live explanation before you look any further.
Step two: check whether the decline concentrates on queries with AI Overview presence, or is broad and query-agnostic
If seasonality doesn’t explain the pattern, segment the affected queries by whether an AI Overview is actually present on those SERPs, via manual spot-checks or a tool that tracks AI Overview presence specifically. A decline that concentrates on queries where an Overview is observed to appear is meaningfully more likely to be AI Overview-related than a broad, undifferentiated decline hitting queries regardless of whether an Overview shows up. A query-agnostic, site-wide CTR decline is more consistent with something else entirely: a different SERP feature (an expanded “People Also Ask” block, added image or video carousels, an ads layout change), a title/meta description rendering issue, or even a broader ranking or indexing problem showing up as a CTR symptom rather than its actual cause.
Google Search Console is the right tool for this segmentation. Filter by query and check search-appearance data (including the AI Overviews filter where available in your account) rather than relying solely on a third-party rank tracker’s general “SERP feature present” flag, since Search Console reflects your actual measured impressions and clicks rather than a sampled estimate.
Step three: check for average position stability
If the CTR decline is concentrated on AI Overview-present queries and average organic position for those queries has stayed roughly stable, that’s consistent with the scroll-depth/pixel-displacement mechanism (the Overview pushes the organic block down the page, reducing the share of users who scroll to it, even though the underlying rank hasn’t changed). If average position has also dropped meaningfully for the same queries, that points to an actual ranking change happening concurrently, a separate cause that needs its own diagnosis, rather than assuming the AI Overview explains the whole effect.
A worked example showing why skipping steps produces the wrong conclusion
Suppose a set of tracked queries shows a 15% CTR decline over the past quarter. A team that jumps straight to “AI Overviews are responsible” without doing the year-over-year check might be missing that the same 15% decline occurred at the same point last year, for a topic with predictable seasonal demand fluctuation unrelated to any SERP feature. Conversely, a team that does check seasonality, rules it out correctly, but then skips the feature-presence segmentation and assumes AI Overviews anyway, might miss that the actual driver is a newly expanded “People Also Ask” block appearing on the same queries, an entirely different SERP feature with a similar displacement mechanism but a different underlying cause and, potentially, a different available response. Only the full three-step sequence, seasonality, then feature-presence segmentation, then position-stability check, actually isolates which explanation the data supports, and shortcuts at any step risk landing on a plausible-sounding but unverified conclusion.
Common misdiagnosis: confusing correlation with AI Overview presence for causation
Even after confirming a CTR decline concentrates on queries where AI Overviews appear, it’s worth checking one more thing before concluding the Overview itself is the driver: whether those same queries also experienced any other simultaneous change, a title tag update, a Search Console-flagged manual action, a Core Update during the same window, that could independently explain declining performance and merely coincides with AI Overview presence rather than being caused by it. Two things being true at once (AI Overview present, CTR down) is suggestive but not definitive proof of the causal link, particularly on sites that also made other changes during the same period; cross-check your own change log and Google’s publicly announced update timeline against the exact window of decline before finalizing the attribution.
A note on sample size before drawing conclusions
Query-level CTR data, especially when segmented down into AI-Overview-present versus absent buckets, can involve small enough impression volumes per query that normal statistical noise looks like a meaningful pattern. Before concluding a decline is real and attributable to any of the three candidate causes, check that the queries you’re analyzing have enough impression volume that the swings you’re seeing aren’t just noise around a small sample. A handful of queries with low impression counts showing an apparent CTR drop is weaker evidence than a broad, consistent pattern across a large set of higher-volume queries, and treating the former as if it had the same evidentiary weight as the latter is a common way this diagnostic process goes wrong even when every step above is followed correctly.
Handling mixed-cause cases where more than one factor is present
Real diagnostic situations often don’t resolve cleanly into a single cause, it’s entirely possible for seasonality and an AI Overview to both be genuinely present and both contributing to the same observed decline. If the year-over-year comparison shows a partial seasonal effect (say, accounting for roughly half the magnitude of the current decline) while the same queries also show a newly-appeared AI Overview, don’t force the diagnosis into a single winner when the evidence supports a shared cause. The practical approach is decomposing the decline as far as the data allows: use the year-over-year seasonal baseline to estimate what portion of the current drop would have happened anyway, then attribute the remaining, unexplained portion to the AI Overview-present segment specifically, cross-checked against the position-stability test described above. Where the data genuinely can’t be cleanly separated, for example when seasonality and AI Overview appearance began in the same week, making it impossible to isolate their individual contributions with confidence, the honest response is reporting a combined, uncertain attribution rather than picking whichever single cause is more convenient to act on. Treating a mixed-cause decline as if it had one clean explanation risks both overcorrecting (assuming an AI Overview mitigation will recover the full decline when only part of it is attributable) and undercorrecting (dismissing a real AI Overview effect because seasonality partially explains the same window).
Practical implication
Don’t skip straight to “AI Overviews are killing our traffic” as an explanation without doing this segmentation, because it’s an easy narrative to reach for and not always the correct one. The diagnostic sequence, seasonality, then feature-presence segmentation, then position stability, protects against misattributing a decline to the wrong cause and either wasting effort on an AI Overview-specific mitigation strategy when the real cause was a ranking drop, or failing to investigate an actual ranking problem because AI Overviews provided a convenient, plausible-sounding explanation that wasn’t actually verified against the query-level data.
