GEO · AEO · AI search
GEO for law firms: what you can actually measure
Generative engine optimization — GEO, also sold as AEO (answer engine optimization) or “AI SEO” — is the work of making a firm more likely to appear, and to be recommended, when an AI assistant answers a prospective client. Much of what is sold under the name is ordinary good practice; some of it is folklore; and almost none of it is measured. This page separates the three.
What GEO means
The term comes from a 2024 research paper that built a test generative engine and measured how changes to a page’s content changed how visibly it was used in the engine’s answers (Aggarwal et al., KDD 2024 (arXiv), read September 3, 2026). Its results were measured on that constructed engine and a benchmark of queries, not on ChatGPT, Google or Copilot, and the effects varied by subject. It grounds a hypothesis worth testing, not a recipe.
In the market, GEO now means any effort to be named in AI answers, and AEO is the same effort under an older name. For a law firm the useful version is narrower: when a prospective client describes their situation to an assistant and asks who to hire, is the firm in the answer, and is it the one recommended?
How GEO differs from SEO — and where it does not
The foundation is shared. Assistants that search the web draw on search indexes. Google states there are no additional requirements to appear in AI Overviews or AI Mode beyond ordinary eligibility for Search (Google Search Central, read September 3, 2026); Microsoft documents that Copilot sends a query to Bing to ground its answers (Microsoft Learn, read September 3, 2026); OpenAI documents a separate crawler, OAI-SearchBot, for surfacing sites in ChatGPT’s search features (OpenAI, read September 3, 2026). A page that cannot be crawled or indexed cannot be read by the assistant searching for it.
The output is different. A results page lists ten links and lets the client decide. An assistant writes one answer that names a few firms and often recommends one. That creates outcomes a results page does not have: a firm can be named and passed over, a firm’s page can be cited as a source while another firm is recommended, and the same question can get a different answer tomorrow.
Mentions, citations and recommendations are three different things
Most confusion about AI visibility comes from counting one of these and calling it another.
| In the answer | What it tells you | What it does not |
|---|---|---|
| Your firm is named | The assistant found you for this question. | That it would send the client to you. |
| Your page is cited | The assistant read your site while answering. | That it named you — a cited page can support a rival’s recommendation. |
| Your firm is recommended first | It is the one firm the answer tells the client to contact. | What the client did next. |
Vendors report these under many names — share of voice, visibility score, position, mention rate. Before comparing two numbers, ask which of the three each one counts and what it was divided by. AgentSignal reports named and recommended first separately, each with its count, and records every cited source beside the answer that returned it (the definitions).
What a law firm can change
None of these is a guaranteed lever; each is a hypothesis about why an assistant could not find, verify or prefer the firm, and each can be tested against the same questions before and after.
- Be reachable. Do not block the crawlers the assistants search with — Googlebot and Bingbot, and the AI search crawlers such as OAI-SearchBot, Claude-SearchBot and PerplexityBot — and serve your text in the HTML rather than only after JavaScript runs.
- Be specific about the work. A page for each kind of matter you actually handle, answering what clients ask first: cost and consultation terms, who handles the case, languages, how quickly you respond.
- Be verifiable. Credentials, bar admissions and results stated so they can be checked — results only where your jurisdiction’s advertising rules permit, with the disclaimers they require (North Carolina State Bar (adopting ABA Model Rule 7.1 and comment), read September 3, 2026).
- Be corroborated elsewhere. Accurate, consistent listings in the directories and review platforms clients use — and never purchased or fabricated reviews (Federal Trade Commission, read September 3, 2026).
Some techniques are sold as GEO with no evidence behind them — an llms.txt file, “AI schema”, hundreds of near-identical city pages. The methodology lists each one we refuse to recommend, with its reasons and dated sources.
What nobody can promise
No one controls which firm an assistant recommends, and the answers vary between models and between runs of the same model. Treat any promise of placement in AI answers, and any percentage improvement quoted without its starting count and denominator, as a claim to check rather than a result.
Measure before you pay for GEO
If you are about to hire an agency for GEO or AEO work, the most useful thing you can hold is an independent baseline: the same client questions, put to the same kinds of AI model, with the answers recorded — before the work starts. Afterwards, the same questions asked again show what changed, with the counts, and what the comparison cannot prove.
That is what AgentSignal does. The AI visibility audit records whether each answer named your firm, recommended you first or recommended someone else, and a private share link lets you hand the evidence to whoever is doing the work.