The situational prompt problem: why AI names a law firm only 18% of the time

What is the situational prompt problem?

The situational prompt problem is the gap between how people describe a legal problem and how answer engines decide to name a law firm. A situational prompt describes an event (a crash, a firing, an arrest, a death in the family) with no request for a lawyer in it. Answer engines respond to that phrasing with steps and general information, and name a specific firm only occasionally.

Citorian’s five-engine personal injury study, run June 13 to 15, 2026 across ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews in five US metros, measured the spread directly. Vetting and direct prompts (“top PI firms in [metro]”, “best PI lawyer in [metro]”) named a firm 94% of the time. Specific-injury prompts (“truck accident lawyer [metro]”) named one 79% of the time. Question-led prompts (“how do I choose a PI lawyer”) fell to 23%, and situational prompts to 18%.

The overall figure hides that spread. Across all 359 logged answers, the engines named at least one firm 62% of the time, a number that sounds healthy until the prompt mix is examined. A firm whose visibility report averages across phrasings is mostly measuring how many direct requests are in its prompt set.

Why does an AI engine give steps instead of naming a lawyer?

Because the prompt asks what to do, and the safe answer to what to do after an injury is not a firm name. Citorian reported that for situational prompts the engines prioritized guidance: call 911, seek medical attention, document the scene, and contact the insurer. Firm names, where they appeared at all, came later in the answer or only after the question was rephrased.

That ordering is a design choice by the engines rather than a ranking judgement against any firm. A situational question is often an urgent one, and every major engine is built to put safety and general information ahead of commercial recommendations when a user describes harm. No amount of directory work changes the fact that the first paragraph of an answer to “I was rear-ended and my neck hurts” is going to be about medical care.

The second reason is that situational phrasing carries no location and no practice label the engine can match against a firm’s record. “Truck accident lawyer Houston” gives retrieval two hard entities to resolve. “I got hit by a semi on the way to work” gives it an event. Retrieval systems match entities to entities, so the prompt with no entity in it retrieves general explanatory pages, government pages and forum threads, and the firm pages that would support a recommendation never enter the candidate set.

How many people ask legal questions this way?

Nobody has published the share of legal prompts that are situational, and Firm AEO will not invent one. What the published data does show is that people use AI chat mainly to ask for guidance, and that the queries they type into AI search are much longer than keyword searches. Both point the same way: the situational phrasing is likely the common one, not the edge case.

OpenAI’s own usage research, published through the National Bureau of Economic Research in September 2025 by Chatterji and colleagues, classified ChatGPT messages by intent and found about 49% were Asking (seeking information or advice to make a decision), 40% Doing and 11% Expressing. The same paper found non-work messages had grown from 53% to more than 70% of all usage, and that Practical Guidance held steady at roughly 29% of conversations. A person describing a car accident to ChatGPT is an Asking, non-work, Practical Guidance message on all three counts.

Google reported on May 19, 2026 that the average AI Mode search is triple the length of a traditional Search query. Longer queries are, almost by definition, more descriptive and less like a two-word keyword. And the audience is not small: in iLawyerMarketing’s August 17, 2026 survey of 1,110 US adults, 41.9% said they would use ChatGPT to research which lawyer to hire, up from 28.1% a year earlier.

Is 18% a ceiling, or only the first turn of the conversation?

It is the first turn. Citorian scored single prompts in logged-out sessions, which is the right design for a comparable study but not how a real conversation runs. The study itself noted that firm names came later or on rephrasing. A person who gets safety steps from the first answer usually asks a second question, and the second question is where hiring intent appears.

The practical model is a short conversation, not a single query. The table below sets out a typical four-turn path for a personal injury situation. It is illustrative: no published study has scored naming rates turn by turn, and the likelihood column reads across from Citorian’s single-prompt bands rather than from a multi-turn measurement.

What does a law firm win when it is cited but not named?

A citation in a situational answer puts the firm’s page, and often its name, in front of the person at the moment of highest anxiety, even when the engine does not recommend it. That is a smaller prize than a recommendation, and it should be measured separately, but it is the only prize available on the first turn for most firms.

The link itself is rarely clicked. Pew Research Center’s July 2025 study of 68,879 Google searches found users clicked a link inside an AI summary in about 1% of visits. The value of the citation is exposure and credibility, not traffic, and a firm that reports citation clicks as its success metric will conclude the work is failing.

The value shows up at turn two and three. Clio’s 2025 Legal Trends Report found that of consumers who had used AI for a legal question, 28% were directed to contact a lawyer. That handoff sentence is where the conversation turns toward hiring. A firm whose page was the cited source for the steps has already been seen, and a firm whose page answers the threshold question for its state is the obvious candidate for retrieval when the next prompt names a city.

Which situational questions can a law firm realistically own?

The ones that carry a jurisdiction, a deadline or a threshold the generic answer cannot supply. A bare description of an accident is answered from general knowledge that no firm owns. A situation that turns on a state’s filing deadline, a notice requirement or a dollar threshold forces the engine to retrieve a source that states the local rule, and that source can be a law firm page.

The table below maps typical situational prompts by practice area to the local hook that makes a firm page retrievable. It is illustrative content planning, not a measurement, and the specific rule for any state belongs to the firm’s attorneys to state and verify.

How should a law firm write a page that answers a situational question?

Lead with the safe, general steps the engine is going to give anyway, then add the local rule the engine cannot give without a source, then state the point at which the situation needs a lawyer. That three-part shape matches how the engine composes its answer, which is what makes individual sentences from the page liftable into it.

The Princeton GEO study (Aggarwal and colleagues, KDD 2024) found that adding statistics lifted visibility in generative engine answers by roughly 22% to 41%, and adding quotations and citations produced gains of a similar order. For a situational page that means a sourced deadline, a named statute and a dated agency figure, not an adjective. The checklist Firm AEO applies to these pages:

• •Open with the steps a reasonable person takes in the first 24 hours, in plain language, with safety first.
• •State the jurisdiction in the sentence that carries the rule (“In Arizona, ...”), so the qualifier survives when the sentence is quoted alone.
• •Give one threshold sentence that says when the matter stops being a do-it-yourself situation, without predicting an outcome.
• •Put a named attorney on the page as author or reviewer, with bar admission, so the page resolves to a person and a firm.
• •Cite the statute, court rule or agency page for every deadline, and date the page.
• •Keep it general information rather than advice to the reader, and have the firm’s ethics counsel approve the threshold language under Model Rule 7.1.

How do you measure a prompt that rarely returns a firm name?

By scoring the conversation, not the prompt, and by counting citations separately from names. A single situational prompt run three times will return no firm name most of the time for almost every firm, which tells the firm nothing. A scripted three-turn conversation, run in logged-out sessions, shows where in the conversation the firm first appears and in what role.

The script is fixed: the situational prompt, then a threshold follow-up (“do I need a lawyer for this in [state]?”), then a hiring follow-up (“who handles this in [city]?”). Each turn is logged with two outcomes, whether the firm was cited as a source and whether the firm was named. This extends the single-prompt ledger described in Firm AEO’s monthly tracking protocol with one new field, the turn at which the firm first appeared.

A firm that is cited at turn one and named at turn three is in good shape even if its turn-one naming rate is zero. A firm that never appears until turn three is relying entirely on the directory layer. A firm that is cited at turn one but disappears at turn three has content that engines trust and a corroboration record that they do not, which points at directory and review work rather than more pages. The figures a firm publishes from this testing fall under Model Rule 7.1: state the prompts, engines, run count and date, and never present a citation as a ranking or an endorsement.