Measuring AI visibility for a law firm

Before you go off buying AI tracking tools, bear in mind that you haven't sized up the opportunity yet.

AI visibility is a part of a firm's search presence, but it is not IMHO the primary source of new cases. Keep that in proportion report by report, budget by budget, and you'll be fine.

What Google reports, which is less than you would like

Search Console has a very limited, per-URL AI features report with AI Mode and AI Overview impressions and clicks.

There are two distinct limits in play here¹:

• No sign of which question triggered the appearance
• AI feature traffic folded into the overall web totals, with no separate line

What Bing reports, which is genuinely more

Bing Webmaster Tools publishes an AI performance report that provides per-URL citation counts and the grounding queries behind them to site owners².

A grounding query is a behind-the-scenes search with somebody's original question at its root. Few tools can show that, and nobody is telling.

Bing makes up a relatively small portion of search, yet this grounding data really only comes from Bing at this point. It takes a bit of reading, but the picture that emerges of how an answer engine uses a firm's site is quite clear. It costs virtually nothing to check this. Everyone I speak with who sells law firm SEO almost never checks their Bing data, so I made a point to bring the Bing numbers to clients.

What the prompt trackers add, and what they do not

Tools like Peec, Otterly, Profound, Gumshoe and Ahrefs Brand Radar offer scheduled runs and multi-engine checks to tell you who gets named for a fixed set of questions².

So for example (just an example), for each practice area in each market that you serve, you write the question a client would ask, plus one with the firm's name.

The trick is to watch the trend across lots of questions. The problem is that you perceive trackers as a scoreboard. A tracker puts your firm in 11 of 40 questions this month against 9 of 40 last month, which is useful, but it doesn't tell us much about any one answer. Position is fleeting, first place in one answer is just a flash in the pan.

Why a single check proves nothing

AI answers vary based on the run, the user, the date, etc., so your firm might not be at the very top for everybody. If you do not insist that your tracker (or agency) give you repeated samples and run each question across engines, then you have no way to know if your numbers are meaningful³.

The same applies to the platforms' citation data.

A monthly figure from one query on one engine is just noise.

Three things that cannot be measured at all

1. Answers given without any search.

Only about 18% of ChatGPT conversations end up in a web search⁵. The other 82% comes from training data, both unseen and unreported. What shapes those answers can change slowly, without warning in three-to-six-month retraining cycles⁷.

2. Retrieval without citation.

Discarded content is 85% of the material ChatGPT retrieves⁶. Your best page can simply be read in full and discarded just moments later. The firm has no way of knowing if its page was read.

3. Which fan-out query produced a Google appearance.

The AI features report does not cover fan-out queries or grounding queries behind an impression¹. On Bing that data is given, on Google it is not given.

The monthly report I would actually send

1. Share of a fixed question set where the firm is named, across at least two engines and with the question count stated. "Named in 11 of 40 questions, up from 9" certainly beats the usual visibility scores.
2. Bing's citation counts and grounding queries for the main practice pages. The only free hard citation data, it seems, is the Bing report².
3. Search Console AI feature impressions, labelled as directional, with the web total caveat written on the page and not just said once in a meeting¹.
4. Then the numbers that pay the bills. Calls, forms, consultations and signed cases, from the intake and measurement page. For law firms especially, call tracking matters more than just about anything. 56.3% of legal conversions are phone calls, and most reports often can't see them⁹.
5. One line on what cannot be measured, repeated in every monthly report so nobody forgets it.

It's a plain, honest report, and long term, it works better than any dashboard with false precision.

The same honesty goes for your average intake desk or front office. I would add that 39% of firms take more than two hours to answer an online enquiry, or never answer it¹⁰. No amount of AI visibility, or Bing citations or mentions in the prompt trackers can change that. Keep that number handy and include it whenever you report.

What to do when the numbers move

Treat a citation drop like a ranking drop: let the checks below guide you to a likely cause.

1. Check the page still ranks. It is often the case that the cited pages you lost in the AI answers are no longer ranking.
2. Check the page is still eligible. A release with a stray snippet restriction or a blocked AI crawler would easily cause this problem. Take a look at the AI Overviews page and check accordingly.
3. Check what changed on the page. I favour keeping the answer near the top, since that is what the engines extract.
4. Check whether the engine changed its sourcing generally. It's fairly common that an engine reshuffles its sources within weeks⁴. If you look at three unrelated clients dropping in the same week, it is definitely not their pages.

The two biggest factors in the citation literature are accessibility and search rank⁸, which is often woefully neglected in AI visibility advice. This list is front loaded with those two checks for reasons given above. Most citation problems are pretty much indexing or ranking problems.