When Google introduced generative AI performance reports in Search Console, they rolled out to a limited set of properties first. They are now rolling out far more broadly, which means most site owners can finally see whether their content is showing up in AI answers.
This is genuinely useful and it is worth looking at. It is also more limited than most of the coverage suggests, and the limitation is the part that changes what you can actually do with it.
The report tells you that a page appeared in an AI answer. It does not tell you what was asked. That single gap decides what you can and cannot learn from it.
Primo Collab
What the report actually gives you
The generative AI performance report covers two surfaces: AI Overviews and AI Mode, drawn from Web search only. Within that, you get:
- Impressions over time.
- Which of your pages are appearing.
- Country breakdown.
- Device breakdown.
For most small business owners, this is the first real evidence about whether AI surfaces are part of their traffic story rather than a thing they have been reading about.
The limitation that matters most
There is no query breakdown. You can filter by page, country, date, and device. You cannot see the searches that produced those impressions.
This is a significant difference from the ordinary performance report, where query data is the most useful thing in the tool. In the standard report you can find a question you rank for but have not written about properly, and that is a content plan. Here, you can see that a page is appearing, and you have to infer the rest.
Practically, that means the AI report answers “is this page working” but not “what is it working for.”
The other documented limits
Worth knowing before you draw conclusions:
| Limitation | What it means in practice |
|---|---|
| Still rolling out | Not every property has it yet. Absence is not evidence of zero impressions |
| Minimum impression threshold | Small sites may not appear at all, even with some presence |
| 1,000 row limit | Standard performance report constraint applies here too |
| Recent data is preliminary | The last few days can change. Do not react to them |
| Discover excluded | Reported separately |
| Search Labs excluded | Experimental surfaces are not counted |
| Chart and table totals may differ | Aggregation works differently between the two. Not a bug |
That second row deserves emphasis. If you look and see nothing, the honest reading is “either I have little presence, or I am below the threshold, or it has not rolled out to me.” Those are three different situations and the report does not distinguish them.
What you can genuinely do with it
Despite the gaps, there are three real uses.
Find out which pages are legible to AI systems. The pages appearing are usually not the ones people expect. That tells you something about which of your content is structured clearly enough to be extracted, which is a useful signal about your writing and markup rather than just your rankings.
Find the pages that never appear. This is arguably more actionable than the ones that do. A high-value service or product page with no AI impressions is worth examining: it is often buried behind a vague introduction, missing structured data, or written in a way that offers nothing quotable.
Establish a baseline now. The single most valuable thing you can do this month is record where you stand, so that in three months you can tell whether a change helped. Without a baseline, next quarter’s number is just a number.
How to use it alongside the normal report
The two reports answer different questions and neither replaces the other:
- Standard performance report: what people search, where you rank, what gets clicked. Still where your content plan comes from. If you have never worked through it properly, start with how to read Search Console without panicking.
- Generative AI report: whether your pages are surfacing in AI answers at all, and which ones.
The workflow that makes sense is to take your query insights from the standard report, and use the AI report to check whether the pages you built from those insights are actually appearing.
The mistake to avoid
Treating a low number as a verdict on your content. The report is still rolling out, has a minimum threshold, and covers only two of the many surfaces where AI systems encounter your content. A quiet report might mean your content is not being surfaced, or it might mean you are not eligible to see the data yet. Check back in a month before concluding anything, and in the meantime keep doing the thing that actually drives citations: answering real questions plainly, early in the page, with specifics worth quoting.