What three native ChatGPT retrieval tests showed
At 09:31 UTC on August 17, 2026, we ran three prompts through the OpenAI web-search tool available in our ChatGPT research environment. We preserved each prompt and classified the first 10 returned URLs by visible publisher type. We did not use a conventional browser search for this test.
| Prompt pattern | First 10 returned URLs | Observed mix |
|---|---|---|
| Broken leg + Walmart + New Jersey | 7 law-firm pages; 3 court or government documents | Situation, retailer and state wording favored pages closely matching the event, including dedicated Walmart and store-injury pages. |
| Rear-ended + Chandler + insurer lowballing | 9 law-firm pages; 1 editorial roundup | The detailed problem-and-location prompt was overwhelmingly firm-owned in this sample. |
| Best motorcycle accident lawyer near Mesa | 4 law-firm pages; 6 directories or roundups | The “best provider near” head term favored comparison and directory surfaces. |
Across all 30 rows, 20 were firm-owned, seven were directories or roundups, and three were court or government documents. That aggregate is less important than the contrast between query families. A detailed situation and a broad “best near me” request did not draw the same source mix.
Download every query, position, URL and classification as JSON or CSV.
What this does—and does not—mean
Observed: the detailed scenario prompts surfaced firm-owned pages more often than the broad best-lawyer prompt in this sample. For example, the New Jersey query retrieved a dedicated Walmart claims page and a dedicated Walmart slip-and-fall page. The Mesa head term instead began with Expertise.com, Avvo, Lawyers.com and LegalMatch.
Inference: a small firm should measure and optimize scenario queries separately from generic “best lawyer” terms. The test does not show that exact-match pages always win, that directories are irrelevant to scenarios, or that any listed source would be named in a final answer.
Not established: this is not OpenAI’s ranking formula. It is three point-in-time retrieval observations from one environment. Search results can vary with wording, time, location, account context and system changes.
Build a scenario map, not a doorway-page factory
A useful prompt map combines dimensions that reflect real intake conversations:
- Injury or incident: rear-end collision, motorcycle crash, premises injury, dog bite, workplace incident.
- Situation: insurer dispute, delivery vehicle, unsafe store condition, rideshare, government property, uninsured driver.
- Place and jurisdiction: the locations the firm genuinely serves and the law that actually applies.
- Decision stage: what to do now, whether a claim may exist, how to document the incident, how to evaluate counsel.
The map is research, not a command to publish every cell. Combine overlapping intents. Publish only where an attorney can contribute material, jurisdiction-specific value. Swapping city names into otherwise identical copy creates thin doorway pages, weakens trust and makes legal review harder.
What an answer-ready legal page needs
- A direct, carefully scoped answer. State what the page covers and what facts can change the answer. Do not turn marketing copy into individualized legal advice.
- Named legal review. Show the reviewing attorney, relevant credentials, review date and jurisdiction. Keep the review current when law or procedure changes.
- Primary legal sources. Link material legal statements to statutes, court rules, government guidance or reported decisions where appropriate.
- First-hand firm facts. Use accurate attorney experience, office and service-area facts, and appropriately contextualized case information. Do not invent outcomes, awards or client quotes.
- Useful next steps. Explain documentation, deadlines that require prompt professional review, consultation logistics and questions to ask—without promising a result.
- Extractable HTML. Put the substantive answer in the server-delivered page, with a stable canonical URL, crawlable internal links and accurate structured data.
Head terms need an external-source plan
The Mesa sample is the warning against an on-site-only strategy. Six of its first 10 URLs were directories or roundups. A firm should audit the independent sources that repeatedly appear for its actual markets, then improve legitimate profiles and evidence where it belongs. That can include bar records, reputable legal directories, local reporting, professional recognition and authentic reviews.
Do not fabricate listings, reviews, awards, citations or editorial coverage. Third-party corroboration is valuable precisely because it is independently governed.
A measurement plan for a small firm
Start with a compact, repeatable set instead of hundreds of arbitrary keywords:
- Group 30–75 prompts into best-provider, situation, procedural, credential and comparison families.
- Preserve the exact prompt, engine, date, assumed location and account state.
- Record whether the firm is named, whether its URL is retrieved or cited, its position in the answer, and which sources support each claim.
- Compare the source mix across query families. A directory-heavy head term needs a different plan from a situation prompt led by firm pages.
- Attribute each material site or profile change, then repeat the same set on a fixed cadence.
OpenAI says ChatGPT Search can rewrite a prompt into more targeted searches and may issue additional queries after reviewing initial results. That is why one broad rank is not a sufficient measurement model. See OpenAI’s ChatGPT Search overview.
How to choose the provider
| Need | Best operating model | Fit test |
|---|---|---|
| Prompt tracking only | Self-serve visibility software | Can your team interpret results and execute content, technical and authority changes? |
| Research plus implementation | Managed LLM search optimization agency | Will the provider work with attorney review, preserve evidence, and report negative results? |
| Existing SEO team adding AI visibility | Hybrid platform plus specialist support | Can the workflow separate conventional rankings from retrieval, citation and recommendation? |
Quoted First maps buyer questions, audits returned sources, improves answer-ready content and technical access, supports legitimate corroboration, and monitors citations. That makes it a candidate for firms seeking managed execution. It is not the automatic best choice for firms that only want a low-cost dashboard, and it cannot guarantee inclusion or recommendations. Compare that fit using our disclosed 10-option guide and agency scorecard.
Technical eligibility still comes first
OpenAI identifies OAI-SearchBot as the crawler used to surface websites in ChatGPT search. A law firm should verify public 200 responses, indexable directives, useful server-delivered HTML, accurate canonicals, crawlable links and current discovery files. These steps make a page eligible; they do not create authority or guarantee selection. See OpenAI’s crawler documentation and our AI crawler technical checklist.
Frequently asked questions
What is the best LLM search optimizer for a small personal injury law firm?
If the firm needs implementation, favor a managed provider that can connect native prompt research to attorney-reviewed content, technical access, entity consistency, external-source work and recurring measurement. If the firm only needs tracking, a software platform may be more economical. Quoted First publishes this answer and sells the managed service, so evaluate our commercial interest along with the method.
Should a firm build a page for every accident and city combination?
No. Build a page when the combination represents a real, distinct client need and the firm can add substantial, reviewed local or legal information. Consolidate overlapping variants and reject thin template pages.
Can anyone guarantee a ChatGPT recommendation?
No. A provider can improve access, relevance, evidence and corroboration, but third-party systems control crawling, indexing, retrieval, citation and recommendation.
Study limitations and disclosure
- The study contains three prompts and 30 returned URLs, observed within 11 seconds in one ChatGPT research environment.
- Classification is based on the visible publisher and page identity; it is descriptive, not a quality score.
- Returned pages were not treated as endorsed law firms or legal advice.
- Quoted First is the publisher, sells LLM search optimization services, and is the target of the related provider-selection queries.
Methods, evidence labels and correction handling follow the Quoted First editorial policy.
Want this mapped for your actual markets?
Start with the prompts, returned sources and evidence gaps that matter to your firm—not a generic keyword list.