Your rankings held but your traffic fell: what AI Overviews did to law firm websites

What the click data actually shows

Clicks fell where AI Overviews appeared, and the effect grew through 2025. Pew Research Center tracked 900 US adults across 68,879 Google searches in March 2025. When a search produced an AI summary, users clicked a traditional result in 8% of visits; without one, 15%. Users clicked a link inside the summary itself in 1% of visits. And users ended their browsing session on the results page 26% of the time when a summary appeared, against 16% without.

Ahrefs measured the same effect from the site side, using Google Search Console data for 300,000 keywords. In April 2025 the average click-through rate for the first organic position was 34.5% lower when an AI Overview was present. Re-run on December 2025 data, the gap had widened to 58%, with position-one CTR on AI Overview keywords falling from 0.073 to 0.016. A page that held position one across the year could lose more than half its clicks from that position without a single ranking change.

How many legal queries trigger an AI Overview? The four numbers, reconciled

Somewhere between under 1% and 78%, depending on what you count, and the figure most often quoted in legal marketing is misattributed. The “78% of legal queries trigger AI Overviews” statistic circulates with Semrush credited as the source. Semrush’s 10-million-keyword study reports overall AI Overview share (rising from 6.49% in January 2025 to a peak of 24.61% in July) and top industries, but no legal figure. The 77.67% number comes from SE Ranking’s YMYL study, which sampled 1,200 “your money or your life” queries and found the legal niche triggered an Overview 77.67% of the time, the highest of any YMYL category.

Other samples give very different answers because they sample different query mixes. SE Ranking’s June 2024 broad study of 100,013 keywords put legal at 0.34%. Ahrefs, in data reported through 5WPR’s April 2026 index, found 23.6% of legal queries triggered an Overview overall and 57.9% of question-style legal queries did. Whitespark’s Q2 2025 local study of 540 queries in Houston, Phoenix and Denver found Overviews on 15% of local-intent queries against a local pack on 93%, while hybrid queries such as “best car accident lawyer in Houston for a rideshare crash” showed an Overview 97% of the time and a local pack only 17%.

Why some firms felt it and others did not

Because the traffic that disappeared was informational, and firms differ in how much of that they had. A consumer practice that built its organic footprint on “what is the statute of limitations in Texas” and “how much is my case worth” articles saw those pages keep their rankings while Google answered the question above them. A commercial litigation firm whose traffic was mostly branded searches and referrals barely moved. LawFuel’s September 2026 review of the data called the split explicitly: consumer practices lost informational traffic; commercial and branded traffic held.

Great Jakes’ May 2026 analysis of its law firm client portfolio is the sceptic’s dataset and it is worth reading straight. AI referrals were 0.47% of sessions, up 220% year over year. LinkedIn sent 187% more sessions than all AI engines combined. Google still delivered 48% of sessions. For a firm whose traffic was never informational, the honest reading is that AI referral traffic is small and growing fast, and the traffic loss story belongs to a different kind of firm.

The local pack is the other exception. Whitespark found the local pack on 93% of local-intent queries and Overviews on only 15%. A firm that wins “DUI lawyer near me” through Google Business Profile has not lost that placement. What it has lost is the top of the funnel, where a person asks a question before they know they need a lawyer.

The traffic that arrives now is worth more per visit

Fewer clicks, higher intent. Semrush’s July 2025 AI search traffic study found AI search visitors 4.4 times as valuable as traditional organic visitors, measured by conversion rate, and projected AI visitors would surpass traditional search visitors by early 2028. Seer Interactive’s seven-month case study found ChatGPT sessions converting at 15.9% against 1.76% for Google organic. Seer also found that brands cited inside an AI Overview earned about 35% more organic clicks and about 91% more paid clicks on the same queries, which is the mechanism a firm actually wants: being the cited source turns the Overview from a competitor into a referral.

The person who reads the Overview, sees a firm named in it and then clicks through has already been told the answer and chosen to contact that firm anyway. That is a different visitor from the one who landed on a statute-of-limitations article in 2023 and left. Sessions fell; the share of sessions that were ever going to become a matter did not fall as much, and for the firms named in the answers it rose.

What to measure instead of sessions

Appearance rate, cited-source share and matters, in that order. Sessions was always a proxy for something else, and it has stopped being a reliable proxy. The measures that survive the change are the ones tied to being named and to the matters that follow.

• •Appearance rate: how often the firm is named across a fixed set of client questions, run repeatedly each month across ChatGPT, Perplexity and Google AI Overviews. This is Firm AEO’s primary metric and the one its guarantee is written against, because it is the only number that describes visibility inside the answer rather than beside it.
• •Cited-source share: of the sources the engines cite for those questions, how many are the firm’s own pages or profiles the firm controls. This is the leading indicator; appearance rate follows it.
• •Impressions in Google Search Console, separated from clicks. A page whose impressions hold while clicks fall is being read by the Overview; a page whose impressions fall has lost retrieval. They need different fixes.
• •AI-referred sessions in GA4, with the caveat that much AI traffic arrives without a referrer and is logged as direct. Intake-form and booking attribution (“how did you find us”) is the reliable backstop.
• •Matters opened, by source. The number the firm actually runs on.

What to do about the pages that lost clicks

Make them the source the Overview cites, rather than the page it summarises past. An informational page that answers “how long do I have to file a car accident claim in Florida” in a single, self-contained, jurisdiction-qualified sentence near the top is the page an engine lifts and attributes. The same content buried in the fourth paragraph after two hundred words of empathy is the page an engine reads, uses and does not credit. Ahrefs’ March 2026 finding that only about 38% of Overview citations come from top-ten pages means the rewrite competes on extractability, not on rank.

The second move is to connect the informational page to a named attorney and a specific matter type. Citorian’s five-engine test found engines named a specific firm in only 18% of situational prompts, the kind that begins “I was rear-ended yesterday”. The other 82% of those answers give advice and tell the person to consult a lawyer, without naming one. A firm whose informational content is written by an identifiable attorney, corroborated on the directories, and linked to a page for that exact matter type is positioned to be the name in that 18% and to widen it. Every one of these pages has to stay inside ABA Model Rule 7.1 and the state’s advertising rules, which for informational content mostly means jurisdiction qualifiers, no outcome implications and the disclaimers the state requires.

A short word on Canada

The mechanism is the same and the data is thinner. Every published figure on how people find lawyers in Canada that Firm AEO could locate is American, recycled by Canadian agencies; Clio’s Legal Trends Report, from a Canadian company with a North American sample, is the closest citable source. Canadian firms have two further measurement problems: Quebec’s Law 25 requires consent for non-essential analytics, so Quebec firms under-count everything, and CASL constrains how an AI-referred lead may be followed up by email. Attribution has to move into the intake form. The advertising rules differ by province too; a Canadian-tagged post on this blog covers them.