AI SEO for law firms is two games. Game one is classic SEO: rank in Google’s blue links. Game two is AEO, getting your firm quoted inside ChatGPT, Claude, Perplexity, and Google’s AI Overviews. The same page can win both, but only if you format it so an AI can lift a clean answer out of it.
I run marketing for 10 law firms. My lemon law firm ranks nationally on organic search in all 50 states. Over the last year I watched something change in the traffic data that nobody warned us about: rankings held, clicks fell. The answer was showing up above the links, and the searcher never scrolled.
That is the whole problem in one sentence. If you only play game one, you lose traffic you never see leave.
Why this hits lawyers harder than most businesses
Legal questions are exactly the kind of question AI answers well. “Does my car qualify as a lemon in Texas?” “How long do I have to sue after a car accident in Florida?” These are informational, jurisdiction-flavored, and low-risk for the AI to summarize. So they get summarized.
The client still needs a lawyer at the end of that. The question is whether the answer names your firm or the one down the street. Being the source that AI cites is the new first page, and almost nobody in legal is optimizing for it yet. That gap is temporary.
The AEO playbook
Seven things. Do them in this order on every page you publish.
- Answer the question in the first two sentences. Write the answer so it makes sense pulled out of the page with no context. No throat-clearing intro, no “in this article we will explore.” If someone quoted only your first 40 words, they should have the real answer.
- Use the actual question as an H1 or H2. People search “how long does a lemon law case take,” not “case timelines.” Match the phrasing. AI models chunk pages by heading, so a question-shaped heading with its answer directly beneath is the cleanest unit to extract.
- Add an FAQ section with real search phrasings. Four to six questions, written the way a nervous client types them at 11pm, each answered in 40 to 80 words. This section gets cited more than anything else on my pages.
- Put facts in tables and lists. Deadlines by state, fee structures, what documents you need, timelines. Tables extract cleanly. Prose facts get paraphrased and lose your name in the process.
- Add schema markup. Article, FAQPage, and BreadcrumbList at minimum. Attorney and LegalService schema on your practice pages. It takes 20 minutes with Claude and it tells crawlers what your content actually is.
- Publish an llms.txt file. A plain markdown file at yoursite.com/llms.txt listing your important pages with one-line descriptions. Adoption is early and the payoff is unproven, but it costs an hour and it is the kind of bet worth taking while it is cheap.
- Be specific enough to be worth citing. Real numbers, real jurisdictions, real names, real dollar figures. An AI choosing between your page and a generic competitor picks the one with facts in it. Generic prose is unciteable by definition.
Classic SEO signals vs AEO signals
| Signal | Classic SEO | AEO | What to do |
|---|---|---|---|
| Keywords | Exact-match phrase in title and H1 | Full natural question | Use both: keyword title, question H2s |
| Content depth | Long pages tend to rank | Answer density beats length | Front-load the answer, then go deep |
| Structure | Headings help skimmers | Headings define extractable chunks | One question per heading, answer underneath |
| Facts | Supporting detail | The thing being quoted | Tables, numbers, named jurisdictions |
| Authority | Backlinks and domain age | Brand mentions across the web | Podcasts, directories, press, quotes |
| Freshness | Helps in competitive niches | Models favor recently updated pages | Update dates and re-publish quarterly |
| Reviews | Local pack rankings | Cited as reputation evidence | Push Google reviews relentlessly |
The AI content trap
Here is where most firms get this wrong. They hear “AI SEO” and produce 40 blog posts a month about “what to do after a car accident.” Those posts do not fail because AI wrote them. They fail because they say nothing that ten thousand other pages have not already said.
What wins is the content only your firm can produce. Your case outcomes. Your state’s weird procedural quirk that nobody blogs about. Your actual intake process. Real settlement ranges from real matters, anonymized properly. The manufacturer that drags every case out an extra 90 days. That material does not exist anywhere else, which is exactly why an AI has to cite you for it.
Use AI to write. Do not use AI to think of what to say.
The workflow I use
- Find the real question. Pull queries from Search Console, run the topic through Perplexity and AlsoAsked, and skim what clients actually asked your intake team last month. That last source is the best one and everyone ignores it.
- Outline answer-first. One-paragraph direct answer, then H2s that are questions, then the table, then the FAQ.
- Draft with your own facts as source material. Paste your case data, your state rules, your numbers into Claude and have it write from those, not from its own knowledge.
- Edit hard for voice. Cut every sentence that could appear on any firm’s site. Usually that is a third of the draft.
- Ship the technical layer. Schema, internal links, updated date, llms.txt entry.
Here is the prompt I use for step three. Fill the brackets and paste.
You are an SEO and AEO editor for a law firm. Write an article optimized
both to rank in Google and to be quoted by ChatGPT, Claude, and Perplexity.
INPUTS
Practice area: [e.g. lemon law]
Jurisdiction: [e.g. Texas]
Exact question to answer: [paste the search query verbatim]
Firm facts to use (use ONLY these, do not invent):
[paste real numbers, timelines, statutes, case outcomes, process steps]
OUTPUT RULES
1. Open with a direct answer to the exact question in 2 sentences,
written so it makes sense quoted with zero context.
2. H2 headings must be phrased as questions real people search.
3. Include one comparison table of concrete facts (deadlines, costs,
options, or timelines).
4. Include 4 FAQs phrased the way a client would type them, each
answered in 40 to 80 words.
5. Plain language at an 8th grade reading level. Short sentences.
6. No filler, no "in today's world," no generic advice that would apply
to any firm in any state. Every paragraph must contain a fact from
the inputs above.
7. Flag anything you could not support with the inputs as [NEEDS FACT].Measuring it
Classic SEO is easy: Search Console for impressions, position, and clicks by query. Watch for the pattern I described, flat rankings with falling clicks. That means AI Overviews took the answer.
For AEO, there is no dashboard yet, so build one by hand. Once a month, ask ChatGPT, Claude, and Perplexity your 10 target questions exactly as a client would phrase them. Log which firms get named, which pages get linked, and whether you appear at all. Ten minutes, tracked in a spreadsheet. When your name starts showing up on questions where it did not before, the playbook is working. My tracker took four months to move on the first questions. See more prompt patterns here.
I am running this in public
This site is built on exactly these rules. Every page opens with a standalone answer, uses question headings, carries a table and an FAQ block, ships Article and FAQPage schema, and publishes an llms.txt. It is a live experiment and I will publish what the citation log shows.
Start here: open Search Console, sort by impressions, and find one query where you rank in the top 10 but almost nobody clicks. Rewrite that page answer-first this week.