How to Check Whether ChatGPT Recommends Your Law Firm Locally

Law firms increasingly ask a simple question: does ChatGPT recommend the firm when a prospective client asks for a lawyer? One prompt cannot measure that reliably because wording, market, conversation context, source availability, or the product itself can change the answer.

Measurement therefore needs a repeatable process instead of a screenshot, with fixed prompts, recorded results, separate mention and citation fields, and scheduled reruns. The method should also distinguish ChatGPT from Google AI, Perplexity, Gemini, and other systems because their answers can differ sharply.

This guide explains how to create that baseline for a law firm without inventing a metric called an AI rank. It covers prompt design, market controls, source capture, repeat testing, referral tracking, crawler checks, ethics safeguards, and a 90-day measurement cycle. It also gives marketing teams a shared baseline for later content and authority work. The goal is evidence that can support decisions about content, local presence, attorney profiles, and outside authority.

One ChatGPT Screenshot Is Not a Visibility Test

ChatGPT Search can recommend businesses, but OpenAI does not publish a fixed local ranking position for law firms. Its Search guidance says results use multiple factors, and OpenAI does not guarantee placement.

Current legal-industry research also shows why a single answer is weak evidence, because a 2026 law-firm study recorded 748 ChatGPT prompt runs across 332 United States markets. In 231 family-law markets tested with two prompt phrasings, 31.2% of the answer pairs shared no firms.

The same study found that only 28.1% of those markets kept the same first-named firm across both phrasings. Those numbers describe that specific July 2026 sweep rather than a permanent ChatGPT rule, but they demonstrate why wording variation belongs in serious testing.

A second 2026 study reinforces that separation: Forward Push research analyzed 38,173 clean answers across four AI tools and reported about 6% overlap between ChatGPT and Perplexity firm lists. The study observed that result in its sample; OpenAI does not disclose it as a platform rule.

FindLaw also tested more than 300 legal-service queries across Google and major AI tools. Its 2026 white paper reported that ChatGPT recommendations overlapped with Google page-one results in less than 25% of its transactional and commercial test set.

Together, these studies point to a practical standard: measure a disclosed prompt set, not an undefined AI visibility score. The baseline should be reproducible enough that another person at the firm could run the same test.

Build a Prompt Set Before Opening ChatGPT

A law firm should decide what it wants to measure before running the first test. Otherwise, prompts tend to drift toward whatever wording produces the preferred answer, making the baseline difficult to repeat and easy to interpret selectively.

Start with the firm’s actual revenue-driving services and markets; a personal injury practice might test car accidents, truck crashes, wrongful death, and premises liability. A family-law firm might test divorce, custody, support, and high-asset matters.

Each topic should then include several intent classes: recommendation prompts ask for a provider, while comparison prompts ask what separates firms. Trust prompts ask whether a particular lawyer or firm appears credible, while problem prompts begin with the client’s situation.

Avoid prompts that force a desired answer, such as asking why the firm is the best; the test should resemble a neutral client question. It should also avoid invented circumstances that make one firm uniquely eligible.

Two phrasing variants are useful for priority prompts because small wording changes can expose instability. For example, compare “divorce lawyer” with “family law attorney,” while keeping the city and underlying intent the same.

Do not create dozens of near-duplicate phrasings simply to inflate the sample, because a compact prompt set is easier to rerun and audit. Twenty well-chosen prompts can be more useful than hundreds of undocumented variations.

Include the exact geographic term in the prompt when the test concerns a market. Google says local results depend partly on relevance, distance, and prominence, according to its local ranking guidance. ChatGPT does not publish the same local ranking model, so explicit market wording makes the test easier to interpret.

Control the Test Conditions and Record Them

Perfect laboratory control is not possible with a consumer AI product, but firms can remove avoidable variation. Run each observation in a fresh conversation so previous messages do not change the context of the next prompt.

Record the date, product, prompt, stated market, and any visible search mode or web-search state. If the test uses a signed-in account, keep that condition consistent across the baseline whenever practical. The goal is not to reverse-engineer ChatGPT; it is to make later comparisons meaningful.

Do not silently change a prompt after the first month; if a new client question deserves testing, add it as a new prompt with a new identifier. Keep the original prompt history intact.

OpenAI’s publisher FAQ also gives the website side of the test. A public site can appear in ChatGPT Search, and publishers should avoid blocking OAI-SearchBot when they want content available for discovery and citation.

The same FAQ distinguishes OAI-SearchBot from GPTBot. The former relates to ChatGPT Search discovery, while GPTBot relates to potential training use. A firm’s measurement sheet should not treat a GPTBot visit as evidence of search visibility.

Test crawler access separately with server logs, robots rules, and the site’s CDN or firewall. A prompt result cannot reveal every technical reason that a particular page was unavailable to retrieval.

Measure Mentions, Citations, Accuracy, and Stability Separately

The first useful metric is presence rate: how many recorded runs named the firm at least once? Divide appearances by completed runs in that prompt group. Keep the denominator visible so a percentage never floats without context.

The second metric is citation rate. ChatGPT can name a firm while leaving its website out of the supporting links. That distinction matters because outside directories, news coverage, business profiles, and other sources may be carrying the firm into the answer.

Forward Push’s 2026 study illustrates that distinction. In its Perplexity sample, directories appeared heavily among cited sources, while law-firm websites appeared less often. The exact percentages belong to that study and platform, but the measurement lesson applies broadly: capture the source, not only the name.

Third, score factual accuracy. Note whether ChatGPT gets the firm’s practice areas, attorneys, locations, telephone number, office status, and other important facts right. A recommendation containing stale or incorrect facts is not the same outcome as an accurate recommendation.

Fourth, measure stability. If a firm appears in three of four variants one month and one of four the next, the change is visible even without a fictional position number. Report stability separately by practice area and market.

These metrics should stay descriptive. OpenAI does not promise that Search will place a business in its results, and an observed position does not reveal an internal ranking weight.

Ethics also matter when a firm acts on the findings. Model Rule 7.1 prohibits false or misleading communications about a lawyer or legal services. Its official comment explains that even truthful statements can mislead when they create an unsupported conclusion.

That means a firm should not turn a favorable AI observation into “ChatGPT ranks the firm number one” unless that statement can be supported accurately in context. State advertising rules may impose additional requirements.

Separate ChatGPT From Google AI and Other Recommendation Systems

Law firms should resist the urge to merge all AI visibility into one score. Google says AI Overviews and AI Mode remain rooted in its core Search systems, according to its AI optimization guide. ChatGPT Search uses OpenAI’s own retrieval process.

Google also now gives site owners dedicated data for its generative Search features. Its Search Console reports show impressions and pages surfaced in features such as AI Overviews and AI Mode. Google says those reports reached all websites worldwide by August 31, 2026.

Use that Google data for Google. Do not treat it as a proxy for ChatGPT. Likewise, a ChatGPT prompt observation should not serve as evidence that a firm appears in Google AI Mode.

The same separation applies to Perplexity, Gemini, and Claude. Forward Push reported about 6% list overlap between ChatGPT and Perplexity in its 2026 legal study. FindLaw also reported materially different overlap with Google page-one results across the AI tools it tested.

Those studies do not prove a permanent relationship among the platforms. They demonstrate why a firm should identify the engine and date whenever reporting visibility.

Google Business Profile data also deserves separate treatment. Google’s practitioner rules cover lawyers as individual practitioners and distinguish solo from multi-practitioner locations. Its business-data guidance says local information can come from owners, public web content, licensed data, and user contributions.

That public record can help explain inconsistencies found during an AI audit. It does not prove that ChatGPT uses Google Business Profile as a fixed recommendation factor.

Use the Results to Diagnose the Public Record

A measurement baseline becomes useful when it points to a specific gap. If the firm never appears and OAI-SearchBot cannot reach key pages, access is a reasonable first investigation. If the site is accessible but firm facts are wrong, identity needs attention.

If ChatGPT names the firm for the brand but not for a priority practice area, inspect practice-area coverage and outside corroboration. Attorney bios should connect the right lawyers to the work, offices, admissions, and current firm.

Google’s practitioner rules are useful during that audit because lawyers can have individual Business Profiles under defined conditions. Do not represent a multi-practitioner office the same way as a solo practitioner. Real-world profile structure matters more than creating extra profiles for keywords.

Review outside sources when competitors appear repeatedly. The legal studies cited earlier found substantial directory visibility in AI answers, although each study used its own design and platform set. Check whether major profiles use the correct firm name, lawyers, location, practice areas, and website.

Review claims, too. The American Bar Association’s 2026 website guidance recommends checking experience, fees, results, testimonials, awards, specialization language, contact forms, and other website elements against applicable ethics rules.

Build reviews lawfully rather than manufacturing them to chase an AI benchmark. The FTC review rule restricts fake reviews and certain incentives tied to review sentiment.

Structured data can clarify visible facts, but it should not become an invisible claims layer. Google’s LocalBusiness guidance requires markup to represent the page accurately. Schema.org also defines LegalService for providers of legal services.

The diagnostic question is always specific: what evidence is missing, conflicting, inaccessible, or weak for this prompt group? That approach is more useful than adding a generic “AI optimized” badge to the site.

Run a 90-Day Law Firm ChatGPT Measurement Cycle

Days 1–30: establish the baseline. Choose priority practice areas and markets, create stable prompt IDs, run the initial tests in fresh conversations, and capture every firm name, citation, source, and factual error. Check OAI-SearchBot access and create the first visibility ledger.

Days 31–60: repair the highest-confidence gaps. Fix stale attorney or office facts, crawler blocks, inconsistent profiles, thin priority pages, and unsupported marketing claims. Do not change several unrelated variables merely to chase one prompt result.

Days 61–90: rerun the same prompt set. Compare presence, own-site citations, accuracy, and prompt stability against the baseline. Keep new prompts separate so the original cohort remains comparable.

Referral analytics should run beside the prompt ledger. OpenAI says ChatGPT Search links can include utm_source=chatgpt.com, which allows analytics tools to identify many direct referrals. Server logs can also show OAI-SearchBot requests.

Neither signal captures every influenced prospect. Someone can see a firm in ChatGPT, search the brand later, then call from Google or type the URL directly. Intake should therefore ask how the prospect first heard about the firm, with AI search available as a response.

A simple monthly report can show the number of completed prompt runs, presence rate by practice area, own-site citation rate, factual errors, ChatGPT referrals, and AI-attributed inquiries. The report should retain the prompt definitions and dates.

Do not declare victory because one favorable answer appears after a site edit. The meaningful question is whether the same market and practice-area prompt set becomes more consistently accurate and visible over repeated measurement.

Frequently Asked Questions

A law firm can measure ChatGPT visibility without pretending the system behaves like a traditional rank tracker. Define the questions first, capture the evidence, repeat the test, and connect the observations to real client acquisition. That creates a baseline the firm can use again after every meaningful improvement.