Google Now Tells You Whether AI Sees Your Firm. Here's What It Doesn't Tell You.

For the past two years, most of what law firms have known about their visibility in AI answers has come from two places: screenshots, and vendors. Both have problems. A screenshot is one observation of one prompt on one day. A vendor's number depends on a methodology you usually can't see.

So it matters that Google now reports this directly. On 3 June 2026 Google announced generative AI performance reports in Search Console, initially for "a subset of websites" (Google Search Central Blog). According to a note on that post and on the help documentation, the reports reached "all websites worldwide" as of 31 August 2026 (Search Console Help). On 24 September Google added a filter for image-based searches (Google Search Central Blog).

If your firm's website is verified in Search Console, the data is probably already waiting for you.

My view is that this is a genuinely useful development, and that it will be misread. Here's why, keeping four things separate: what Google's documentation says the report contains (fact), what follows from that (inference), what we're testing (hypothesis), and what I think (opinion).

What the report actually contains

Fact, from Google's documentation. The Search report covers two features: AI Overviews and AI Mode. Google says it expects "to update this list over time". There is a separate generative AI report for Discover, and data from Search Labs experiments isn't included.

The report has one metric: impressions. Google defines these as "how many times links to your site were shown to a user in a generative AI feature on Google Search".

You can group those impressions four ways:

• Pages: the final URL the AI feature linked to, after redirects, mostly attributed to the canonical URL.
• Countries: where the search came from.
• Devices: desktop, tablet or mobile.
• Dates: daily, weekly or monthly, in Pacific Time. (The June announcement also mentions hourly granularity; the current help page doesn't. Check what your account shows.)

A search type filter splits "Web: text-based" from "Web: multimodal". The multimodal data, added in September, covers searches using Lens, Circle to Search on Android, image uploads to Google Search and Chrome's "Search this image".

Three mechanical details matter more than they look:

1. The chart counts your site once per appearance. Google's example: if two results from the same site appear in one AI feature, "they count as a single impression in the chart total". The table, grouped by page, counts per page. So chart and table totals can differ, and that's expected.
2. The usual limits apply. The table shows up to 1,000 rows, and the newest data is preliminary (a dotted line) and can change.
3. Exports turn gaps into zeros. Values shown as "~" or "-" in the interface become zeros in the downloaded file. A zero in your spreadsheet may mean "not available", not "none".

Some sites won't see the report at all: Google says this can happen if a property hasn't received enough generative AI impressions, or has excluded itself from those features.

What it doesn't tell you

Fact. The report lists no metric other than impressions and no dimension other than page, country, device and date. In particular, it doesn't let you see, within this report:

• Clicks from AI features. Google says this AI data is also included in the overall performance report, which does show clicks, but there it isn't separated out.
• Queries. You can see which page was linked, not what the person asked.
• Position or prominence within the answer.
• How your firm was described, whether it was recommended, merely listed, or cited as a source for a general point.
• Who else appeared. Search Console only ever shows your own properties, so there is no competitor view.

Inference. Put those together and the report answers one question: do links to our site appear in Google's AI features, on which pages, where and on what devices? It can't answer: for what questions, against which competitors, described how, and with what effect on enquiries?

For a law firm, that gap is significant. An impression on your "what happens at a first mediation" page is a very different commercial event from an impression on your "family lawyers in Leeds" page. The first may mean an AI answer used your explainer as a supporting link. The second is closer to being put in front of someone choosing a firm. The report counts both the same way.

That's the distinction I keep returning to in this series: an AI mention and an AI recommendation are not the same thing. An impression is a mention of a URL, not a verdict on the firm.

Why it's still worth having

Opinion. None of that makes the report weak. It makes it a baseline, and a good one.

It's first-party: it comes from the system producing the answers, not from someone sampling it from outside. It's free. It covers real searches from real people rather than a prompt list someone designed. And it's longitudinal from the day you start watching, which is the thing most firms lack.

In an earlier piece I argued that one AI answer is an observation, not a ranking, because answers vary between identical requests. Independent research keeps reinforcing that. A September 2026 preprint studying citation decisions in AI search found that when the same frozen search sessions were re-run, 15% of individual cite-or-don't-cite decisions changed (CITECHOICE, arXiv:2609.15164). That study is small (129 transcripts), general-purpose rather than legal, and not about Google's products specifically. But it illustrates why an aggregate count across many real searches is more trustworthy than any single check.

That is what Google has given you: aggregation. What it hasn't given you is context.

Pull your baseline this week

1. Confirm access. Check your firm's site is verified in Search Console and that you can open the Generative AI performance report. If it isn't there, note why: access, low volume, or an exclusion setting.
2. Export the longest range available, by page and by date, for text-based and multimodal separately. Store the file with the date you pulled it.
3. Tag your pages by purpose. Mark each URL as practice-area, location, lawyer profile, explainer or other. This is the single most useful thing you can add, because the report won't.
4. Compare against your normal Search performance for the same pages and period. Which pages appear in AI features at all, and which don't?
5. Record what changes, and when. Site changes, new content, Google announcements. Search Console lets you add annotations. Without them, next quarter's movement will be uninterpretable.

Then leave it alone for a month before drawing conclusions. The AI visibility checklist covers the wider audit.

What to avoid

Don't let anyone sell the number back to you as an "AI ranking". I'd expect this report to appear in agency dashboards quickly, which is fine as long as it's labelled honestly. It's an impressions count. It has no position, no query and no competitor, so it cannot be a rank. If a report shows "AI visibility score: 72" derived from it, ask how.

Don't read page impressions as practice-area endorsement. Your employment law page appearing often may mean AI answers link to it as background, not that your firm is being put forward for employment matters.

Don't treat a small or zero number as a verdict. Many law-firm sites have modest search traffic. Low volume, the export's zero-for-unavailable rule and the preliminary-data window can all make the picture look thinner than it is.

Don't rebuild pages in reaction. Google's own position is that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" (Google Search Central). A baseline is for understanding before you act. That's the principle I call measurement before optimisation.

Don't assume it describes other AI systems. This is Google Search only. People also ask ChatGPT, Claude, Perplexity and Copilot about legal problems, and there's no reason to assume those systems treat your firm the same way.

The gap this leaves

This is where FirmRanker's work sits, so let me be clear about my interest. FirmRanker was founded by me. Practice Proof, which I also founded, provides digital and AI visibility services to law firms.

The questions the report can't answer are the ones FirmRanker is designed to test: which prompts a firm appears for, including the ones clients actually ask that don't contain the word "lawyer"; which competitors appear alongside or instead of it; how it's presented, as a recommendation, a listed option or a source; and how that varies across repeated runs and across different AI systems.

This article reports no FirmRanker findings. We're testing whether Search Console's impression data and prompt-level observation tell a consistent story for the same firms, which would need firms willing to share their exports. We don't know yet whether they will line up. If they don't, that would be interesting too. Any results will be published with the methodology.

My view is that the two are complements. Google tells you, reliably, that you appear. Prompt-level measurement is needed to understand for what, against whom, and how.

Confidence summary

The better question

The report answers "do we appear?" That's worth knowing, and it's worth knowing from Google. But the question a managing partner actually needs answered is different: when people ask the questions our clients ask, are we the firm they're pointed to, and is that changing?

Google has made the first question easy. The second still takes deliberate measurement. For the broader picture, see AEO for Law Firms.

FirmRanker was founded by Dan Toombs. Practice Proof, also founded by Dan, provides digital and AI visibility services to law firms.