Search Console AI report: what it actually measures
An impression report is useful when its scope and missing values are preserved. Use this reading method and tracking template.
Last updated:
The new AI report in Search Console lets you track impressions in Google’s generative answers. On its own, it does not measure visits or demonstrate an effect from an llms.txt file. To support a useful decision, preserve the period, page and availability status alongside each value.
What changes in Search Console
Google announces worldwide availability as of 31 August 2026. The report covers AI Overviews and AI Mode, excluding Search Labs experiments. Dimensions include pages, countries, dates and devices. Recent data can still change. Google report help.
The help page also retains a gradual-rollout notice and says insufficient impressions can explain a missing report. General availability therefore does not prove that your property displays it.
Our recommendation: save your first observation with its extraction date. Do not treat it as a complete history of your AI visibility.
An impression does not prove a visit
An impression and a click describe different events. Their counting depends on the displayed result; Google explains these conventions in its guide to impressions and clicks.
Keep Web clicks in a separate column. Dividing those clicks by impressions from the AI report would mix different scopes, rather than establish an AI click-through rate. A visit from another assistant also belongs in a separate analysis.
The same distinction applies to our file examples: observing a public file does not establish that a search engine used it. Our methodology separates observation, validity and demonstrated use.
The zero-value trap in exports
Google’s help says unavailable values shown as “~” or “-” become zeros in downloaded data. An exported zero therefore cannot distinguish no impressions from no measurement by itself. Totals can also differ when aggregated by property or by page. Export and aggregation rules.
These are three teaching cases constructed for this article, with no real traffic data. Use them to test your spreadsheet before importing your own figures.
| Visible value | Exported value | Status to preserve | Supported reading |
|---|---|---|---|
| 120 | 120 | Available | 120 impressions in the selected scope |
| ~ or - | 0 | Unavailable | No conclusion about volume |
| 0 | 0 | Available | Zero measured impressions in this scope |
We recommend adding an availability field before comparing periods. If you only have the export and cannot check the original state, mark it “unknown”. Do not turn that uncertainty into an estimate. The dated teaching dataset makes these cases reusable.
A reproducible record for each update
Use this simple procedure and keep it consistent between observations:
- Record the property, dates, filters and selected dimension.
- Preserve the raw export and extraction date; work on a copy.
- Associate each row with the same canonical page and record availability.
- Separately log site changes: content, translations, redirects or an llms.txt file.
- Compare periods of equal duration and scope; document differences before interpreting them.
The blank CSV template contains the fields for this procedure. Stable field names simplify imports; downloading it requires no Google connection or personal data.
An increase following a change remains a temporal coincidence until alternative explanations have been examined. Do not claim that llms.txt increased visibility based on that timeline alone.
Where does llms.txt fit in this analysis?
Google says Search does not use llms.txt as a special visibility mechanism. Other systems may use the file, but its presence is not a requirement for appearing in Search’s AI features. Official Google guide.
Keep two questions separate: is your file usable by the tools you intend it for, and do your pages receive Google impressions? The validator answers a structural question. It does not certify a search performance outcome.
Does the report prove Google read our llms.txt?
No. Page impressions do not trace retrieval of this file. Documenting a request requires separate evidence, and establishing how the response was used requires further evidence.
Should we change the site if the report is missing?
Start by checking what is available in the correct property. A missing report does not identify a site defect on its own. Avoid a redesign based solely on that signal.
When should we compare two observations?
When their scopes are comparable and you have recorded any provisional data. Choose a cadence that suits your work; a daily check adds little if it cannot inform a decision.