Real Numbers, Finally: How to Read Google’s Generative AI Report in Search Console

We ran the first URL-level comparison of AI impressions against traditional search impressions on DocuSign, Sage, and other global brands. Here’s the workflow, what the data showed, and what it can’t tell you yet.

For two years, almost every claim about AI search has been an estimate. Visibility scores, share-of-voice indexes, prompt-sampling tools – useful, but all of it is lab data. It’s a proxy someone built because the real thing wasn’t available.

Google’s generative AI report inside Search Console now supports a path toward real user data. It is the second report that shows real data about an AI discovery feature: actual impressions, tied to actual URLs, from an LLM operating inside Google Search. Not a simulation of what an AI might cite. What it did cite. 

The challenge for brands, and even many industry experts, is turning this new data into something actionable. We’ve developed a practical workflow that helps teams move beyond raw impressions to identify meaningful patterns, prioritize optimization opportunities, and make more informed decisions about where to invest in AI visibility.

Why Google’s Generative AI report is worth your attention

Two things make it different from the AI visibility tooling most teams already pay for.

It’s observed, not modeled. Every visibility tool on the market runs prompts and infers presence. This is Google reporting on its own surface. When you look at a saturation rate (the share of a URL’s impressions that came from generative AI results) you’re looking at something that happened, not something a vendor estimated. This idea of a saturation ratio gives us the first view into Google’s preference for using content in an AI response. 

It’s URL-level data and that’s the unlock: Because the report is tied to individual URLs, teams can connect it with traditional Search Console data, Google Analytics, and other reporting sources to uncover meaningful patterns: which page types appear in AI responses, which templates perform best, which topics gain visibility, and which markets or locales are over- or underperforming.

There is also a broader strategic benefit. Until now, much of the conversation around AI visibility has relied on modeled or proxy metrics. Actual impression data gives SEO and leadership teams a tangible reality check. It shows where AI responses are contributing to non-branded awareness and, for many brands, reveals that AI visibility is already more significant than expected. That recalibration alone makes the analysis worthwhile.

The Generative AI Impression workflow

For now, accessing this data still requires some manual work. Google has not yet made generative AI reporting available through the Search Console API, which means teams cannot automatically pull it into dashboards or recurring reporting workflows. Instead, the analysis starts with a manual export from the Search Console interface and a separate export of the corresponding traditional search data.

The good news is that the process is relatively straightforward. With the right date ranges, filters, and URL-level exports, most teams can complete the initial analysis quickly. The value comes not from the export itself, but from how the datasets are combined and interpreted.

1. Export the generative AI report. In Search Console, open the generative AI report and export at the URL level. You’re capped at 1,000 URLs — a real constraint on large sites, and the first thing to disclose when you share results.

2. Export the equivalent traditional data. Pull the search appearance total report for the same date range and the same property.

3. Segment before you export. Run one pull with no filter for a global view, and one with a country filter applied. For Docusign we did global and US, because that’s where the investment concentrates.

4. Match on URL. The URL is your peg. A VLOOKUP joins AI impressions to overall impressions for each page.

5. Bring clicks, CTR, and position over from the traditional report. The AI report gives you impressions and nothing else. No clicks, no CTR, no position. Everything downstream of an impression has to come from the standard report.

6. Calculate AI share per URL, then roll it up by page type. The single-URL number is noise. The pattern by template, directory, or content type is the finding.

The final step is to turn the generative AI export into a connected dataset rather than treating it as a standalone report. By matching each URL to traditional Search Console metrics and, where useful, Google Analytics, conversion data, page-type classifications, or other internal reporting, you can evaluate AI visibility alongside traffic, engagement, and business performance.

That matching process is where the analysis becomes useful. AI impressions alone tell you where visibility exists; the connected dataset helps explain which pages earn it, whether those pages also drive clicks or conversions, and which content types, templates, topics, or markets deserve further investment.

What DocuSign, Sage, and other brands’ data showed

Content-heavy pages get cited. The strongest AI-share pages are still overwhelmingly informational: explainers, FAQs, how-to content, and definition-led pages. This held even for templates that read as product pages. If the template carries substantial information below the fold, it gets pulled. The logic is unremarkable once you say it out loud: a generative answer needs material to synthesize and a source to attribute, and thin pages supply neither.

The top-cited page was an FAQ nobody had touched. DocuSign’s most cited and mentioned URL was their e-signature FAQ page — a page that hadn’t been optimized in a long time. What it has is a clean question-and-answer structure with additional FAQs at the bottom. It wasn’t built for AI. It just happens to be shaped the way a generative system wants to consume information.

Commercial pages lag behind. Product-oriented pages remain important search destinations, but they generally show lower AI share than top informational pages. These pages are built to convert a user, not to answer a question a generative system can lift and cite. That gap is also the opportunity: adding the kind of product information a generative system can cite is what closes it. 

Clicks didn’t move. We looked for a correlation between high AI impression share and click or CTR performance. For DocuSign, it was flat, with a slight negative lean. That may differ on your site — it’s worth checking — but nobody should walk into this expecting AI citation to show up as traffic.

The findings confirmed insights from other AI analytics tools. This matters. The report largely validated what other AI visibility or tracking tools were already indicating. That’s not a disappointing result — it’s a triangulation result, and it’s the reason we’d tell you use available AI reporting tools, like Bing’s Webmaster AI Performance report, to get a broader understanding of your AI presence.

What to actually do with the findings

The export is the starting point, not the deliverable. Three moves follow.

Reverse-engineer your top AI drivers. Take the URLs with the highest AI share and run a structural analysis across them. What’s the common thread, a content pattern, a template element, a way of phrasing? This is closer to a landing-page or feature audit than to keyword work.

Then look off-page. Are those same URLs heavily linked or discussed on the home page? Cited in editorial roundups, on G2 or other review sites? The pattern that explains AI citation often isn’t contained in your own content.

Benchmark locales against the global average. If you’ve invested in localization and a market is underperforming the global AI share, that’s a concrete signal worth chasing, particularly as Google keeps expanding AI features market by market, most recently in France.

On cadence: run it quarterly. There isn’t enough movement to justify weekly or monthly manual pulls, and without API access the effort compounds fast. Once the API arrives, the right build is a dashboard with annotations — so when you roll out an FAQ block or a template change, you can see whether AI impressions actually respond.

Use the Generative AI Report

If Google follows Bing’s path, grounding queries are the most plausible next addition: the queries that triggered your citation. That would be a substantial upgrade. Clicks are the thing everyone actually wants, and we wouldn’t expect them soon… if ever! 

Until then, this is what exists: a partial, slightly inconsistent, genuinely real window into how an LLM uses your content inside the world’s largest search engine. That’s more than we had last quarter. Pull it, read it carefully, state the caveats, and don’t let it be the only thing you look at.

If you’d like the spreadsheet template we used to build this comparison, it’s linked below — download your own Search Console exports, and it will do the rest.

[→ Download the Generative AI vs. Traditional Search comparison template]