The question moved. Most firms didn’t follow it.
A buyer researching who to hire doesn’t only search anymore. They ask. “Who’s a good personal injury lawyer in Phoenix.” “How do I find a law firm that handles mass tort cases.” The answer comes back as a short, generated list of sources, not ten blue links.
AI visibility is whether a firm is one of those sources. It’s a different target than a Google ranking, and most firms are still building for the old one. Roughly 78% of legal search queries already trigger a Google AI Overview. The generated answer isn’t a future problem. It’s the one showing up on legal searches right now.
We ran our own count to find out who actually wins that answer today. We pulled 236 grounding citations from AI answers to 26 lawyer-hiring queries across 8 metros, one engine, one dated snapshot. Then we classified every cited source: a firm’s own website, a legal directory, or earned press.
That result inverts the standard advice. The cross-industry assumption says AI answers run on press coverage and third-party listings, so firms should chase placements first. In legal, earned press was the smallest real class we measured, at 11%. The source the engine reached for most was the firm’s own page.
Read that carefully, because it’s better news than it sounds. The asset that wins the citation is the one thing a firm fully controls. No pitch, no placement fee, no editor’s calendar between the firm and the answer. The catch: not every firm’s page qualifies. The 67% went to the specific pages an engine could trust and quote, and the firms that publish nothing citable were invisible in the same answers.
What “AI visibility” actually means
Ranking answers one question: does this page show up in a list of results. AI visibility answers a harder one: does the model trust this source enough to name it out loud in a two-sentence answer.
That splits into two states worth tracking separately.
Cited. The AI names the source and usually links to it. This is the rare, high-value state.
Mentioned. The AI references the brand by name inside the answer with no link attached. This happens far more often than a citation does, and it still shapes what the model associates with the brand the next time someone asks.
A firm chasing only the “did we get a link” question misses most of what’s actually happening. Both states get counted separately in a real AI-visibility read, never blended into one number.
Why there’s no single AI-visibility score
The instinct is to want one number, the way a Google rank is one number. AI visibility doesn’t work that way, because the engines aren’t reading the same web.
Fractl’s analysis of 8,090 keywords across 25 verticals and 22,410 unique domains found that Google’s AI Overviews and standalone LLM chatbot answers cited the same domain for the same query only 7.2% of the time.
Winning one surface doesn’t win the others. A firm optimized only for Google’s AI Overviews is optimizing for one of at least three places a buyer might be asking. Measuring visibility per engine, not as one blended score, is the only version of the metric that means anything.
What actually earns the citation
If firms’ own pages take 67% of the citations, the real competition is firm page versus firm page. The question stops being “how do we get covered” and becomes “why would the model quote our page instead of the one across town.”
None of this replaces the fundamentals a real SEO build already requires: a fast site, real content, real authority. What changes is what gets rewarded on top of that.
A defined, consistent entity. The same firm name, address, attorneys, and credentials across the firm’s own site, its directory listings, and any press mentions. A firm that reads as three different entities across the web is harder for a model to confirm as real, let alone cite.
Original, dated findings. A generated answer needs something to point at that isn’t a restatement of the same five articles everyone already wrote. A firm publishing its own numbers, dated and sourced, gives the model something to cite that a competitor’s rewritten “best of” post can’t offer. That’s the exact mechanism behind the 236-citation study above: it’s Taqtics running this play on itself, in public.
Freshness on the pages that matter. A stale page on a fast-moving topic risks falling out of the set a model even considers. This isn’t a minor ranking signal the way it is for classic SEO. For a generated answer pulling live sources, it can be the difference between a page getting read at all and getting skipped before the model ever gets to judge its content.
How to actually measure it
Skip the single dashboard number. Measure two things, on a schedule, per engine.
Cited versus mentioned. Run the real questions a buyer would ask, not keyword fragments, across ChatGPT, AI Overviews, and Perplexity. Record which sources get a named citation with a link, which get mentioned with no link, and which don’t show up at all.
Share of the answer in a market. Once a quarter, map which sources are winning the citation for a given practice area in a given market, and which ones are still unclaimed. An unclaimed position in that answer is next month’s content plan, not a mystery.
That’s the build behind Get Cited by AI (AEO): the owned data study that earns the citation, the audit that shows what each engine says today, and the share-of-voice read that shows what’s still open. The click isn’t disappearing. It’s moving into an answer. The only question is whether a firm’s name is in it.
References
- Taqtics legal AI-citation study. n=236 grounding citations, 26 lawyer-hiring queries across 8 metros, one grounded answer engine, June 2026.
- Search Engine Land. "How AI media partnerships influence your brand visibility in genAI: Research." Fractl analysis, October 20, 2025.
- SE Ranking. "AI Overviews Industry Analysis: Legal Queries Trigger Rate." 2026.