AI citations for law firms: audit the sources behind an answer
An AI citation is a source link shown with an assistant's answer. For a law firm, that link might point to its own website, an attorney profile, a directory listing, or a third-party article. A citation tells you what source the answer displays. It does not, by itself, establish that the answer recommends the firm or that the source caused the recommendation.
The useful task is to inspect the answer and its links together. Does the assistant put your firm forward as an option? Which source supports the location, practice area, or attorney description? Is that source accurate? This guide provides a repeatable audit and a worksheet you can copy into your own spreadsheet.
Separate the answer from the source
Use two columns rather than a single “AI visibility” count. The first records what the answer says about the firm; the second records where its displayed links go.
| Observation | What to record | What it does not establish |
|---|---|---|
| A link to your firm's practice page | Direct website citation and exact destination URL | That the firm was suggested for hire |
| A link to your firm's directory profile | Profile citation, directory name, exact URL | That the directory's paid placement caused the citation |
| A link to a directory category page | Third-party source; check whether your firm actually appears there | A citation to every firm listed on that page |
| Your firm is named as an option to contact | Recommendation, with the supporting sentence saved | That your own website was cited |
| Your firm appears only in passing | Mention, with the surrounding sentence saved | A recommendation |
| No readable answer was returned | Unavailable observation and failure reason | That the firm was absent from a successful answer |
These are reporting distinctions, not claims about an assistant's internal ranking system. Briefly's Case Recommendation Share definition separates a cited source from a recommendation. A source audit adds detail about the links without changing that definition.
A worked example: one answer, three different findings
Consider this fictional answer to “Which firms handle estate planning in Cedar Harbor?”:
“Northbank Law and Lakefield Legal are options to contact. Northbank describes estate planning services on its website. Check appointment availability directly with each firm.”
Suppose its displayed links lead to Northbank's estate-planning page, a directory profile for Lakefield, and a general article about choosing an attorney.
The audit would record:
- Northbank: recommended in the answer; its website is directly cited. Check that the linked page describes the relevant service and market.
- Lakefield: recommended in the answer; a third-party profile is cited. Check the profile's address and practice information before proposing a website rewrite.
- The general article: a displayed supporting source. Its inclusion does not make it a recommendation of either firm.
Now suppose the directory profile shows an outdated office address. The concrete task is to correct that profile and save evidence of the correction. “Buy more directory listings” is a different proposal, with no support from this example. The example is illustrative; it is not a Briefly scan result or evidence about how often any directory appears.
Run a source audit your team can repeat
1. Fix the question and market before collecting
Choose a small, written set of real hiring questions relevant to your practice. Include the city or service area in the question when location matters. Use the hiring-question library as a starting point, then keep the selected wording stable for comparisons.
Save the exact question, assistant, date, and search surface. Google AI Overviews and Google AI Mode should have different labels. Record any explicit location, account, or session context that might affect the observation. Do not quietly combine branded questions such as “Does Northbank handle estate planning?” with unbranded questions asking which firm to contact.
2. Preserve the answer before interpreting it
Save the answer text or a screenshot and all displayed source destinations. If the interface has a source panel, inspect that panel too. Keep the answer's wording separate from your notes. A screenshot of a firm name alone loses the context needed to distinguish a passing mention from a recommendation.
For repeat collection and outcome classification, use the existing recommendation-tracking workflow. The audit here starts with those saved answers; it does not establish a new minimum sample size or promise statistical certainty.
3. Open each source and check the specific claim
Ask three questions: Is this the correct firm? Does the page actually support the nearby statement? Is the information still accurate?
Record missing support as “not verified.” A source can be inaccessible, out of date, or too general to support the answer's wording. Do not fill the gap with an assumption. A link near a paragraph is useful evidence to inspect, not proof that every sentence in that paragraph came from that page.
4. Keep one row per answer–source pair
Copy these fields into a spreadsheet. Use the same answer ID for every source attached to one answer.
| Field | What belongs in it |
|---|---|
| Answer ID and collection date | Stable ID plus date and time |
| Question and surface | Exact question; assistant and mode |
| Collection status | Successful or unavailable; retain the failure reason |
| Firm outcome | Absent, mentioned, cited, recommended, or primary recommendation |
| Source URL and type | Exact destination; firm website, firm profile, directory category, or other |
| Claim being checked | The answer's specific wording about location, service, or identity |
| Verification | Supported, contradicted, or not verified; explanation |
| Correction owner | Firm editor, web vendor, directory owner, or no action |
| Next check | Due date and evidence needed to close the task |
Keep an answer-level record even when there are no citations. Otherwise your worksheet will silently drop uncited answers. When reporting counts, state whether you counted answers, source links, unique URLs, or domains. Five links in one answer are not five independent recommendations.
Turn findings into a short work queue
Fix an incorrect phone number, address, attorney affiliation, or service description first. Prefer the authoritative record your team can substantiate. Do not make a profile agree with an AI answer when the answer itself is wrong.
Next, inspect important pages whose content is unclear or incomplete. Google's content guidance emphasizes useful original work and clear authorship. For a law firm, a practical response is to have the responsible attorney review the page's substantive claims and have staff verify its actual consultation process. That application is our editorial recommendation, not a documented recipe for winning citations. Google's helpful-content guidance
If the website itself sends conflicting identity or location information, the local entity check can help inspect what selected pages declare. It cannot verify a directory account or prove which data an assistant used.
Treat paid directory proposals separately from factual corrections. Ask which observed source and reader need the purchase addresses. Google's spam policies prohibit links created primarily to manipulate rankings, and its generative AI guide cautions against inauthentic mentions. Neither a paid listing nor a larger citation count establishes recommendation quality. Google's spam policies, generative AI guide
What to report after making a correction
Record the changed URL, the exact change, and the date it became visible. Repeat the same question set later and keep unsuccessful observations. Report “the corrected address appeared in these saved answers” when that is what happened. Do not turn a before-and-after observation into “this edit caused the assistant to recommend us.”
For Google-specific eligibility and reporting, use the law-firm AI Overviews checklist. For recommendation outcomes across assistants, use the free Answer Log alongside your source worksheet. You classify each saved answer in the log; it calculates recommendation share from your entries. The separate worksheet holds the source-by-source investigation.
Briefly measures what AI assistants say. It does not rank, rate, or endorse attorneys. This is a marketing measurement workflow, not legal advice.