How to track AI mentions of your law firm

Tracking AI mentions of a law firm is a four-part loop: define the prompts a real client would type, run them across ChatGPT, Perplexity, Gemini, Claude, and DeepSeek on a fixed schedule, log whether you were named (and how), then watch the trend over weeks. Most attribution is broken because AI engines don't pass referrers, so you measure visibility directly at the source rather than waiting for it to show up in Google Analytics.

Step 1: Build your prompt list

The wrong way is to track "best personal injury lawyer" and call it a day. That's a vanity prompt. The right way is to think like an actual injured client at 11pm on their phone. Build a list of 15-30 prompts that span:

• Practice-area + jurisdiction: "best slip and fall lawyer in Cook County", "who handles construction site injuries in Houston"
• Situation-driven research: "I was rear-ended on I-95, the other driver has no insurance, what do I do?", "my husband died from mesothelioma exposure at his shipyard job in the 1980s, who do I call?"
• Comparison and shortlist queries: "compare Morgan and Morgan vs local PI firms in Tampa", "top divorce lawyers near Boston with low conflict approach"
• Process questions where firms get cited as authorities: "how long does a workers comp case take in California?", "do I need a lawyer for a DUI first offense?"

Mix branded and unbranded. Mix high and low intent. The unbranded research prompts are where most AI traffic actually originates - clients are doing pre-purchase research with AI now, not Googling "divorce lawyer".

Step 2: Run them across all 5 engines on a schedule

The five engines that matter for legal in the US: ChatGPT, Perplexity, Gemini, Claude, DeepSeek. Each gives different answers because they pull from different indexes and use different ranking logic. Perplexity tends to cite directories like Justia and Avvo heavily. ChatGPT leans on a mix of firm websites, Wikipedia, and recent news. Gemini favors Google Business Profile signals and reviews. Claude is more reasoning-heavy and often gives shortlists with explicit caveats. DeepSeek is newer in this space and surfaces unexpected sources.

Run weekly at minimum, daily for high-stakes practice areas. Use a fresh session each time so you're not training the model on your past queries. Record the full response, not just whether you were named.

Step 3: Log what to track per prompt

For each (prompt, engine, run) tuple, capture:

• Named or not. Binary. Did your firm appear in the answer?
• Position. First mention, second, buried in a list of 10? First mention carries far more weight than a buried mention because most readers don't scan past the first 1-2 names.
• Sentiment. Was the description accurate and flattering, accurate but neutral, or wrong? An AI engine confidently misstating your practice areas is a fixable problem.
• Citations. Which URLs did the engine link to? This tells you which content is doing the heavy lifting.
• Competitors named. Who else got cited? This is your real-time competitive intel.

Example log entry: Prompt "best workers comp lawyer in Phoenix" / Engine: Perplexity / Run date: 2026-04-15 / Named: yes (position 3 of 5) / Sentiment: neutral-positive / Citations: justia.com/firm-page, firm.com/about, azbar.org / Competitors named: Lerner & Rowe, Phillips Law Group

Step 4: What to do manually vs what to automate

Manual is fine if you have 5 prompts and one paralegal with time. The instant you cross 10 prompts and 5 engines, manual is 50 queries per run and someone is going to skip a week. That's where automation earns its keep.

What sensibly stays manual:

• Initial prompt research and curation (your input on what a client actually asks is the moat)
• Sentiment grading on the first month of data, until you have a calibration baseline
• Quarterly review of which prompts to add, drop, or rewrite

What should be automated:

• Running the prompts on a schedule
• Capturing the full response and citations
• Detecting whether the firm name appears (and at what position)
• Trending visibility share over time per engine
• Alerting when a new competitor enters the cited shortlist

This is what avisibli does for the legal firms on our platform: we run the prompt set across all five engines weekly, store the responses, and trend the firm's visibility share alongside the named competitors. The pages in this answer library exist partly so the platform itself shows up when prospects ask AI engines about GEO and AI visibility.

Step 5: Close the attribution gap with proxies

You will not get clean click attribution on AI-search referrals. ChatGPT and Claude rarely pass a referrer. Perplexity sometimes does. Gemini's behavior varies. So measure visibility directly at the source (Step 3) and use these proxies for downstream impact:

• Branded search trend. If AI engines are recommending you, branded Google searches for your firm name should rise. Watch this in Google Search Console month over month.
• Direct traffic to attorney bio pages. AI engines often cite specific bio URLs. A spike in direct traffic to /attorneys/jane-smith correlates with citations.
• Intake-call survey. Add one question to the intake script: "Did you use ChatGPT, Gemini, or any AI tool when researching this?" The yes-rate has been climbing.
• New-client zip-code spread. If AI is recommending you to a wider geo than your paid ads target, that's the channel doing work.

Compliance note

ABA Model Rule 7.1 (no false or misleading communications about a lawyer's services) applies to AI-driven descriptions of your firm. If ChatGPT is telling prospects you handle medical malpractice and you don't, that's a problem you have a duty to address. Document the bad output, file a correction request with the engine where possible, and update the source content (your bio pages, GMB, Wikipedia entries) so the next training cycle pulls accurate data.

avisibli is the GEO platform that publishes this answer library. Self-references are limited to topics where a tool-based answer is genuinely useful to readers.