Skip to content

Cookbook

AI Overview traffic drop: how to isolate it in Search Console

Search Console never tells you an AI Overview appeared — it gives you a signal pattern you can reason from, if you segment properly.

Updated ·

When organic traffic falls, an AI Overview is the most comfortable explanation available: external, not our fault, and impossible to disprove. That is exactly why it deserves suspicion. Search Console data does not expose whether an AI Overview appeared — impressions are not flagged according to whether a generated summary sat above the results. What the data does expose is the PATTERN of the signal, and from that you can reason responsibly. The key distinction is this: if your position is unchanged and your impressions are unchanged but clicks fall, that points to a presentation change, which is what an AI Overview is. If position slipped or impressions fell, the problem is in ranking and the AI Overview is a distraction. This recipe works that separation through, and states plainly where provable ends.

The procedure

  1. Fix a clean comparison window

    Pick two windows of equal length on the same weekday rhythm, avoiding seasonal breaks, holidays and your own publishing spikes. If you deleted, redirected or rewrote pages in between, the comparison is void — rule that out first. The most common mistake is not bad analysis; it is a bad window.

  2. Segment by query type

    Split queries into at least three groups: branded, informational (what, how, why), and transactional. They behave differently. Generated summaries appear predominantly on informational and definitional queries, which is where a drop would be expected. If your drop is concentrated in branded queries, this is almost certainly not the cause.

  3. Read CTR alongside position, not instead of it

    This step is the core of the diagnosis. Build a view with average position, impressions and clicks side by side, per query group. The presentation-change pattern looks like this: position roughly flat, impressions flat or rising, clicks falling — CTR degrading without you having moved down the ranking. If position also slipped, you are looking at a ranking change, and an AI Overview is at most an aggravating factor.

  4. Check the top queries by hand

    Take the ten or twenty most affected queries and look at them live, signed out, in your target market's language and region. Does a generated summary appear, and if so, who does it cite? This is the only direct evidence obtainable — the data gives you the pattern, the manual check gives you the fact. Record the date: these surfaces change often, so the observation is perishable too.

  5. Rule out the usual alternatives

    Before concluding it was the AI Overview, eliminate the more common causes: an algorithm update in the window, a technical change of your own (template, speed, internal linking), a new competitor in the top positions, seasonality, or a measurement change on your side. If any of those coincides with the drop, it is the likelier explanation, because it is more directly demonstrable.

  6. State what you cannot know

    If you get this far you have a supported hypothesis rather than proof, and it is worth writing that down. Search Console does not mark which impressions carried a generated summary, so no causal claim is responsible; what you can claim is that the pattern is consistent with a presentation change and that the manual check supported it. That is less than many reports assert, and it has the advantage of being true.

Frequently asked questions

Does Search Console report AI Overviews?
Not as a separate dimension. Impressions and clicks originating from a generated summary sit inside the normal performance data without being distinguished. That is why only pattern-based inference is possible, and why no responsible precise loss figure can be quoted.
How do I know it wasn't an algorithm update?
From timing and pattern. A ranking change usually moves position too and rarely affects all query types evenly. Stable position with falling CTR points more toward a presentation change. Certainty is not available, because both can happen at once.
Which of my pages are most affected?
Typically those answering short, definitional questions — precisely the content type a generated summary can substitute for. Deeper decision-support or procedural content is less substitutable, because the answer does not fit in three sentences.
Should I noindex the affected pages?
Almost never. An impression is not harm in itself, and noindex costs you the remaining clicks too. If a page now produces impressions without clicks, that is a content question: either it answers something already answered for you, or it is not deep enough to be worth opening.
How long before I can call a drop persistent?
At least several full weeks, compared on the same rhythm — daily variation and weekly patterns are large on their own. Acting on one or two bad days is the most common error, because the normal rebound afterwards then gets credited to whatever you changed.

Sources

  1. Google — Search Console performance report (what it contains). https://support.google.com/webmasters/answer/7042828
  2. Google — AI features and your website. https://developers.google.com/search/docs/appearance/ai-features
  3. Google — Google crawlers (Google-Extended vs Search). https://developers.google.com/search/docs/crawling-indexing/google-common-crawlers