AI Visibility for Law Firms: What Clients Ask and What AI Gets Wrong

Prospective clients rarely ask an AI assistant for "a lawyer". They ask for a practice area in a place, often with a constraint attached. That pairing of practice area and jurisdiction is what an engine has to match, and it's exactly where firm facts go wrong.

Here are realistic prompts, written as labelled hypotheticals:

• "Who are the best employment lawyers in Denver for a wrongful termination case?"
• "Find me a divorce attorney licensed in both New York and New Jersey."
• "Which law firms in Austin handle startup fundraising and Series A term sheets?"
• "What does an estate planning lawyer in Phoenix usually charge for a will and trust?"
• "Does [attorney name] still practice at [firm name], or did she move firms?"
• "Compare [firm A] and [firm B] for a trademark dispute."
• "Is there an immigration lawyer near downtown Chicago who speaks Spanish?"

Each prompt tests a different fact: admission in a state, an office in a city, a named attorney's current firm, a language, a fee range. If any one is missing or wrong in the sources an engine reads, the firm can drop out of the shortlist. Consumer-heavy practices face the sharpest version of this, which is why AI visibility for personal injury lawyers gets its own treatment.

The next move: write your own list of 10 to 20 prompts by practice area and city, in the words a client would use.

Each engine reaches your site through a different door, so check them separately.

• ChatGPT. Search answers depend on OAI-SearchBot. OpenAI's crawler page says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Search is model-discretionary, so some answers come from what the model learned in training.
• Google AI Overviews and AI Mode. Google states a page must be indexed and eligible to be shown with a snippet to appear as a supporting link. There are no additional technical requirements.
• Perplexity and Claude. Both run their own retrieval bots (PerplexityBot, Claude-SearchBot). Claude.ai search is a user toggle.
• Microsoft Copilot. It grounds on the Bing index, so Bing Webmaster Tools status matters on its own.

Two things nobody controls: whether an engine searches the web for a given question, and what a model learned before its knowledge cutoff. As of October 2026, OpenAI's model documentation lists a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path ChatGPT also uses. A partner who joined after that date doesn't exist in an ungrounded answer.

That is why the same prompt can return different firms. The page on how AI search differs from Google search explains the mechanics, and the search on vs off test shows how far your grounded and ungrounded answers diverge.

Law firms change faster than the web records them. Lawyers move laterally, offices open and close, practice groups merge, and admissions get added. Old copies of those facts live on in training data and on pages nobody updates.

The one on-page step with a positive experimental signal is plain: state the specific fact in a sentence on a page engines can fetch. "Jane Doe is admitted in Texas and Oklahoma" gives an engine something it can quote. A vague "licensed in multiple states" does not.

Check for contradictions inside your own site too. A roster page, an old news post and a footer that disagree about who works where give every engine a reason to get it wrong.

Schema markup won't fix stale facts. Ahrefs' matched-control test found AI Overviews citations fell 4.6% on pages after schema was added, and Google says structured data isn't required for generative AI search.

When engines cite other sites instead of yours, being named correctly on those sites is the route into the answer. The evidence shows a strong association there, not a guaranteed effect. Treat these as places to check, not levers to pull.

• Legal directories. Confirm each attorney's firm, admissions, office and practice areas. Request corrections for departed lawyers.
• State bar member listings. These are the record of admission. Make sure names and firm affiliations match what your site says.
• Review and map listings. Closed offices and old phone numbers live here longest.
• News and press coverage. Old announcements of a lateral hire can outrank the later departure.
• Former firms' sites. A partner's old bio still live on a previous firm's site is a common source of wrong answers.

Rankings and "best lawyers" lists come up often in these answers. Whether appearing on best-of lists helps AI recommend you is a separate question with a careful answer. So is whether you can pay to be recommended in ChatGPT answers.

The next move: list the five third-party pages that show up most often when you run your prompts, and audit each one against your roster.

If a crawler can't fetch your page, nothing on it reaches the answer. These checks are mechanical, and they're where the strongest evidence sits.

robots.txt, per bot

Rules are per bot and independent. Blocking GPTBot stops training use only; OAI-SearchBot controls ChatGPT search. A blanket AI block often removes search eligibility by accident. A firm that wants to appear in ChatGPT search but stay out of training could write:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Check PerplexityBot, Claude-SearchBot and Googlebot the same way. There is no robots.txt token that blocks only AI Overviews.

Firewall and CDN

A firewall or CDN can return a 403 to a crawler that robots.txt allows. Nothing in a dashboard warns you. Cloudflare changed its defaults on 2026-09-15 to block Training and Agent crawlers on ad-displaying pages for new domains and free-tier customers, so re-check even if nobody edited anything. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Rendering

Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Attorney directories that load bios through a JavaScript filter may be invisible to them. Make sure each attorney's name, admissions and office appear in the raw HTML.

Law firms publish things they may not want summarised: case results with client details, matter descriptions, or pages kept live for a specific audience. Google and Bing document on-page controls that keep content out of AI answers.

• noindex keeps a page out of the index entirely.
• nosnippet and max-snippet:0 stop the page's text from being shown in snippets and AI features.
• data-nosnippet on an HTML element excludes just that section. Bing says content marked this way is still indexed normally but excluded from snippets and AI summaries.

These controls only subtract. They hide content; no markup makes a section more likely to be used. They also get copied by accident. A nosnippet added to a template to hide one disclaimer can suppress every attorney bio built from it.

For Google's AI Overviews and AI Mode, Search Console also has an inclusion control that is on by default and inherits to child properties. Confirm no property has been opted out by inheritance.

The next move: search your templates for snippet directives and confirm each one was intentional.

Some pages hide instructions aimed at AI systems: white text, comments or cloaked content telling a model to praise one firm or disparage another. It exists, and it is illegitimate. Never use it on your own site or pay anyone who offers it.

For a law firm, the risk runs the other way. A page that mentions your firm could carry text designed to misrepresent you. A 2026 study of LLM search agents (arXiv:2606.16821) found that vulnerability to manipulated web content varied widely across model backends, measured in a mock search environment. Those rates shift with every safety update, so don't treat any single figure as a fixed property of a model.

What to do: when an AI answer describes your firm in a way you can't trace to a real source, check the pages it cites. Look for hidden text or instruction-like phrasing. Document it and report it to the host, much as you'd handle a defamatory post.

1. Days 1 to 7: baseline. Run your prompt list in ChatGPT, Google, Perplexity, Claude and Copilot, logged out. Record which firms appear and every fact stated about yours. The free checker on this site runs up to three buyer questions as a quick snapshot. Answers through an API can differ from the apps, as API answers vs consumer app answers explains.
2. Days 8 to 14: access. Check robots.txt per bot, test firewall reachability for each AI crawler, and confirm attorney bios are in the raw HTML. Check Google Search Console and Bing Webmaster Tools separately.
3. Days 15 to 21: your own facts. Make one current roster the source of truth. State each attorney's admissions, office and practice areas in a plain sentence. Remove departed lawyers and retired practices, and redirect their pages.
4. Days 22 to 30: third parties and re-check. Request corrections on directories, bar listings and map profiles. Check pages for accidental snippet directives. Re-run the same prompts and compare.

Some errors will persist until models retrain or engines re-crawl. That part is out of your hands. What you control is whether every source an engine reads is current and reachable. For ChatGPT specifically, see what you control when getting cited by ChatGPT.