What Google's new AI reports actually show (and who can see)
Two Google reports on AI visibility that no one's talking about.
Google AI Mode passed one billion monthly users in May.
For most of that, a Shopify catalog is either getting read into an answer or it isn't, and until recently there was no way to check that.
This fact changed twice in the last two months.
People who follow this might have heard about at least one of the two below.
Merchant Center's new AI performance insights
Google announced this on May 27. The pilot only went live for a limited set of US-based Merchant Center accounts around last week.
It's product-level, not page-level - worth pointing out.
1/ Share of voice against a competitor set inside AI-driven shopping results.
2/ Where products are surfacing across the discovery, evaluation, and purchase stages of a shopping journey.
3/ An attribute completeness score that names which structured fields - color, material, style - are missing on which products.
The last one is the same gap our audit tool has been helping merchants find in their Shopify catalog. Google just started showing it back to merchants directly, inside the same account that manages the Shopping feed.
It needs a live Merchant Center connection to work. If the Google & YouTube sales channel is installed and synced, that box is already checked.
What Search Console's AI report shows
This is the one from my post back in June, for anyone catching up.
Google split generative AI impressions into a dedicated Search Console report on June 3, announced by the Search Console team. Separate views for AI Overviews, AI Mode, and the generative parts of Discover. Before this, all of it sat mixed into the standard Web report with classic blue links, with no way to isolate it.
Search Console's Generative AI report
What it shows now: impressions by page, country, device, and date - back to about May.
Same release added a toggle to block a site from these features entirely. Almost no brand should use it, but it exists now.
What it doesn't show (yet): clicks
Neither one has fully rolled out
Search Console's version began with a subset of UK sites, in part due to regulatory pressure from the UK's Competition and Markets Authority.
A Google engineer described the pace as deliberately incremental, testing and reading feedback rather than flipping a switch. US accounts might still not have it.
Merchant Center's version is narrower still: pilot, US only - expanding to Canada, Australia, India and New Zealand in the coming months, per Google's documentation.
What the Search Console report doesn't tell you
It doesn't appear to be in the Search Console API yet.
The query endpoint developers already use for everything else still only recognizes web, image, video, news, and Discover as result types, nothing for AI Overviews or AI Mode.
So this can't be pulled into Claude, a dashboard, or BigQuery the way the rest of Search Console data can, at least for now. Google has a track record of shipping the UI report before the API catches up, so this may be a matter of time.
And GA4's AI Assistant channel, the natural pairing for the click side, has a gap of its own. It auto-tags clicks from ChatGPT, Gemini, and Claude with no setup.
But a click from Google's own AI Mode or AI Overviews gets filed under Organic Search, not AI Assistant. The one AI surface that the Search Console report measures is the one GA4 mislabels. Pairing the two gets you close, not exact.
What to check either way
1/ Confirm access first. Search Console → Performance → look for a Generative AI tab. Merchant Center → Analytics → Products → AI performance.
2/ If you have either, look for zero-impression pages or products first. That's the clearest signal something isn't being retrieved at all.
3/ Read the country breakdown before concluding. A flat line in a market can mean the rollout hasn't reached that market yet, not that the catalog failed there.
4/ Treat GA4 as directional, not exact, for the reason above.
Different report, same underlying question - does the structured data behind a product exist, and is it complete enough for AI to use?
The tool audits that same gap today.
And if you want to solve for it, we are building Atomz exactly for that.
- Ankit
If this was useful, the next one will be too.
Weekly. Free.
Unsubscribe with one click.
