By The FirmCited team, BizLegal AI Last updated
General information and software only - not legal advice, no attorney-client relationship. Consult a licensed attorney.
Reports are generated with AI-assisted software and may contain errors. Review before relying on them.
AI citation tracking is a dated record of whether a law firm name appears when someone asks an AI answer engine a client-style question. You keep the question list fixed, you name the engine, and you save what that answer did with the firm name. The record describes that sitting. It does not say the firm will be named the next time you ask.
A mention check asks one question and stores four facts: the date, the engine, the exact question, and whether the firm name appears. Other firm names and any linked URL go on the same row. If the answer names no firm, write "not named". You are filing what one product returned in one session. Two people can see different wording on the same day, and a later model update can change the names even when the site did not change. Read the saved answer before you mark the row. Models misspell firm names and merge two practices in the same city.
For a US consumer practice, a workable set is ChatGPT, Claude, Perplexity and Google AI Overviews. A short pass can use one engine. A comparison pass uses the same questions on each engine you can actually open, and a cell you did not run should say "not checked". Google AI Overviews sit above some Google results and do not appear for every query. When none is shown, record "no overview". Do not copy the ordinary blue links into the AI column. Those links are a search result, a different record. Write the model name when the screen shows one, and "name not shown" when it does not.
Use the words a person uses before they hire, and include the practice and the place. "Who handles estate planning in Portland, Oregon?" is a checkable question. A line from the firm's ads is not. Keep a small mix stable: a "who should I call" question, a cost question, a process question such as "what should I bring to an initial meeting", and one question that names your firm. Ten questions will show a pattern. Twenty-five is the panel size FirmCited uses on a paid scan. Changing a word makes a new question, so do not tidy the wording between months.
A chat remembers earlier turns, and that memory can steer the next answer toward a firm name you just typed. Start a new chat for each question, use a temporary chat when the product offers one, and note whether you were signed in. Paste the full answer or save a screenshot, then mark the name as present or absent. A name in a list, a name in a quoted headline, and a name the model says it cannot confirm are three different facts. Say which one you saw. Four engines and ten to twenty-five questions take a few hours by hand. The value is the file you can open next month.
Use the same columns every time: date, engine, question, firm named (yes or no), other firm names, and where you saved the answer. "Named on 6 of 25 questions in October" describes October. It is not a forecast, and it is not a count of consultations or fees. Monthly is a practical gap for a fixed list. Weekly rows often differ because the model drew a different page, not because the firm published anything new. A yearly check can hide a long stretch in which the name was absent. Put the date in the filename and read across months. One row is a snapshot, not a cause.
The sheet does not explain why a name appeared. It does not measure consultations, fees or case volume, and it does not say an engine will name the firm after you publish a page, add schema, or edit a biography. Answer products are outside the firm's control. If software drafted a row, a person at the firm should read the saved answer before treating the row as fact. Label the sheet as a record of answers on specific dates, not as a rank, a market share, or a pipeline.
Open last month's list and do not edit it. Ask each question on each engine in a fresh session, mark yes or no, list the other names, and file the sheet with the date. Write "no overview" or "refused" instead of leaving a cell empty. For a separate look at the site, the free AI Citation Scorecard is a 20-point self-check scored out of 40. You score it, and FirmCited does not run prompts on that page. FirmCited Monitor repeats a 25-prompt list every month on up to 4 engines, sends a report when at least 2 engines answer, and emails a private portal link. The pricing page states the fee in US dollars. FirmCited is not a law firm. Neither the hand sheet nor the report assigns a rank or commits an engine to a future mention.
A free 20-point self-check you score yourself. FirmCited does not run prompts on that page.
A monthly 25-prompt list on up to 4 engines, with a private portal link by email, when at least 2 engines answer.
Current fees, in US dollars, for the audit, Monitor and the other plans.
What the llms.txt file is, what it does not do, and a short example.
Ask a client-style question in a fresh chat, save the answer, and note whether your firm name appears. Repeat the same question later. One chat is a snapshot, and a later chat can name someone else.
Run the same question list on each product, in a new session, with a row per engine. Record the AI Overview when Google shows one, and write "no overview" when it does not. Leave the ordinary blue links out of the AI column.
Once a month, with the wording held fixed, is a practical cadence. Weekly checks move for reasons that may have nothing to do with a new page. The series of monthly sheets is the record.
No. Tracking stores what an answer already said. It does not create a ranking, and it does not cause an engine to cite the firm. The next answer can still omit the name.
Keep the date, the engine, the exact question, a yes or no for your firm, the other firm names, and a copy of the answer. If nobody has read that copy, do not treat the yes or no as settled.
25 prompts across up to 4 AI engines, competitor share of voice and a prioritised fix list, for USD 490, credited toward any install.