Learn · How-To

How to Read the Search Console AI Report Without Fooling Yourself

Google's generative AI performance report is genuinely useful and routinely misread. It is the first first-party signal any of us have had about AI surfaces, which makes it valuable. It is also impressions-heavy, already counted inside your existing web totals, and it suffered a logging fault in August 2026 that produced a drop lots of people are still explaining as a performance problem. This guide covers how to read it without drawing conclusions it cannot support.

1. 1

   Find the report and check whether you have it

   Open Search Console, go to the Performance section and look for the generative AI experiences reporting. Availability has been rolling out progressively rather than appearing for everyone at once, so if it is absent that is a rollout question, not a problem with your site.

2. 2

   Understand that these numbers are already in your totals

   The AI figures are not additive. They are a slice of your existing Search performance data, not extra impressions layered on top. Anyone presenting AI impressions as incremental visibility has misunderstood the report, and the resulting total is double counting.

3. 3

   Treat impressions as presence, not performance

   An impression means your page was shown in an AI surface. It does not mean you were cited prominently, quoted, recommended, or clicked. Presence is worth knowing and it is not the same as influence. Say which one you are reporting.

4. 4

   Exclude 12 to 20 August 2026 from your comparisons

   A logging error caused sharp, artificial impression declines in the generative AI report during that window, and it was corrected around 20 August. If your chart shows a dip there, check the dates before you investigate anything else. Any month-on-month comparison that includes those days is comparing against a fault.

5. 5

   Compare the trend, not the absolute number

   The absolute impression count tells you very little on its own. The direction over eight to twelve weeks, and which pages carry it, tells you considerably more. Export monthly and keep your own history, because report definitions change and the interface will not warn you when they do.

6. 6

   Pair it with something the report cannot see

   Add two things beside it: your server logs, which show which AI crawlers actually fetched which URLs, and a self-reported "how did you hear about us" field on your forms. Between them they cover most of what an impressions-only report structurally cannot.

7. 7

   Write down what you still cannot prove

   Finish the exercise by listing what remains unknown — whether an impression led to a mention, what an AI actually said about you, and what any of it produced commercially. A measurement practice that names its own blind spots is worth more than a dashboard that implies it has none.