What Is Case Recommendation Share?
Case Recommendation Share (CRS) is the percentage of AI assistants' answers to client-style lawyer-hiring questions that recommend a specific law firm — put it forward as one to contact, not merely mention or cite it.
If a potential client asks ChatGPT, Gemini, Perplexity, Google AI Overviews, or Google AI Mode a question like "who should I hire for a custody case in Columbus?", the answer usually names two or three firms and moves on. Case Recommendation Share measures how often your firm is one of them.
Case Recommendation Share is the headline metric Briefly computes. It is not an industry standard, and we don't claim it is one. We define it here in full (the formula, the classification rules, and the limitations) so that anyone can compute it themselves and check ours.
The formula
Case Recommendation Share = recommended answers ÷ successful tracked answers
Where:
- Recommended answers — answers in which the firm is classified recommended or primary recommendation (defined below).
- Successful tracked answers — every answer an assistant actually returned during the measurement window. Failed scans (timeouts, refusals, errors, empty responses) are excluded from the denominator entirely.
That denominator rule matters. If an assistant is down for a day and 40 scans fail, a metric that counts failures as "not recommended" reports a drop that never happened. CRS only counts answers that exist.
What counts as "recommended": the classification ladder
Every successful answer is classified into exactly one of five states for each tracked firm. In plain language:
- Absent: the firm does not appear in the answer at all.
- Mentioned: the firm's name appears, but not as a suggestion. "Unlike larger firms such as X…" is a mention.
- Cited: the firm's site or profile appears as a source, but the answer text doesn't put the firm forward.
- Recommended: the answer puts the firm forward as an option to contact. "Firms like X and Y handle these cases."
- Primary recommendation: the answer's lead suggestion: named first, singled out, or explicitly preferred.
Only states 4 and 5 count toward Case Recommendation Share. "Your firm appeared in an AI answer" and "the AI told someone to call you" are different events, and only one of them wins cases.
A worked example
The numbers below are fictional. The firms are fictional. The domains use .example so no real firm is implied.
Say Briefly tracks 40 client-style hiring questions for Harlow & Voss Injury Law (harlowvoss.example) in one market for one week. 38 answers come back successfully; 2 scans fail and are excluded.
| Classification | Answers | Counts toward CRS? |
|---|---|---|
| Absent | 17 | No |
| Mentioned | 4 | No |
| Cited | 4 | No |
| Recommended | 9 | Yes |
| Primary recommendation | 4 | Yes |
| Successful answers | 38 | — |
Case Recommendation Share = (9 + 4) ÷ 38 = 34.2%.
A competitor, Kessler Injury Group (kesslerinjury.example), appears in 21 of the same 38 answers, but 16 of those are mentions and citations. Its CRS is 5 ÷ 38 = 13.2%. A raw "visibility" count would rank Kessler ahead of Harlow & Voss. The recommendation count reverses it. That reversal is the point of the metric.
Why not just use a "visibility score"?
Most AI visibility tools report a blended score: mentions, citations, and recommendations averaged by an undisclosed formula. Case Recommendation Share differs in two deliberate ways:
- It only counts answers that send business. A mention in a sentence about your competitor inflates a visibility score. It does not make a phone ring.
- It excludes failed scans from the denominator. An outage or a refusal says nothing about your firm; treating it as a zero manufactures fake volatility.
The cost of that strictness is a smaller, harsher number. A firm with a 60% visibility score may have an 18% Case Recommendation Share. The harsher number is the one that predicts phone calls.
How CRS relates to the other metrics
- Recommendation share of voice — your recommendations divided by all firm recommendations in the same answers; CRS asks "how often am I on the list," share of voice asks "how much of the list is mine."
- Average recommendation position — among answers where you're recommended, whether you're the first name or the "also consider"; CRS counts the wins, position grades them.
- Platform coverage — CRS computed per assistant; a 40% overall CRS can hide a platform where you're at zero and a competitor owns every answer.
Measurement notes: why one spot-check misleads
CRS is only meaningful as an aggregate over many questions, many runs, and multiple assistants. Single spot-checks (pasting one question into ChatGPT) are close to noise:
- Asking the same assistant the same question twice does not reliably return the same firms. A 2026 audit of retrieval-augmented commercial recommendations found that even identical reruns of the same prompt produced recommendation sets with only ~50–61% overlap (arXiv, 2026).
- Rewording matters more than it should. The same audit found cosmetic paraphrases of a prompt dropped recommendation overlap to ~29%. Carnegie Mellon-led research found the same brittleness at the brand level: ordinary paraphrases shifted how often a model mentioned a target brand from never to always, and in the study's adversarial setting, attacker-optimized synonym substitutions moved mention likelihood by up to 78.3 percentage points (LLM Whisperer, CHI 2025; arXiv 2406.04755). The headline figure is an attack result, not everyday variance. But the everyday paraphrase effect alone is enough to sink spot-checks.
- Platforms disagree. Only 11% of cited domains appeared across more than one of four platforms tested (ChatGPT, Perplexity, Google AI Mode, Claude) in Whitehat SEO's 118,000-answer study (2026), so one platform's answer tells you little about the other four.
Briefly therefore computes CRS from repeated scheduled runs of a fixed question set across five assistants, and reports it as a trend, not a snapshot. Every underlying answer is stored and readable; each number traces back to real answers you can open yourself. Full details live on our methodology page.
Honest limitations
- CRS measures AI outputs, not lawyer quality. It is not a ranking of firms, and a high CRS is not an endorsement of anyone's lawyering.
- It depends on the question set. Track questions a real client would never ask and you get a precise answer to the wrong question. Question realism is a judgment call; we document ours.
- Classification has edge cases. "You might look at firms such as X" sits near the mentioned/recommended border. We publish our rules so disagreements are about the rules, not hidden behavior.
- Assistants change without notice. A model update can move every firm's CRS overnight. That makes trends more trustworthy than any absolute value.
- Answers vary by user context. Assistants personalize by location and history. Tracked answers approximate a clean-context user, which no real client exactly is.
FAQ
Is Case Recommendation Share an industry-standard metric? No. It is the metric Briefly defines and measures. The definition, formula, and classification rules are published openly on this page so anyone, including competitors, can compute it the same way.
What's a good Case Recommendation Share? There's no universal benchmark yet; any tool that gives you one without publishing its data is guessing. What matters is your number against named competitors in your markets, and whether the gap is closing.
How is CRS different from a mention rate? Mention rate counts every appearance of your firm's name, including unflattering comparisons and stray citations. CRS counts only answers that put your firm forward as one to contact.
Why exclude failed scans instead of counting them as misses? A failed scan contains no information about your firm. Counting a ChatGPT outage as "not recommended" shows a drop that reverses when the outage ends: fake movement that trains you to ignore the metric.
Can I compute my own Case Recommendation Share without Briefly? Yes. Write 25–50 questions a real client would ask, run each several times across the assistants your clients use, classify every answer on the ladder above, and divide recommendations by successful answers. It's tedious by hand, which is why we built the tracker, or start with the free check to see a one-time snapshot.
About the author
John Rice builds and operates the scan engine behind Briefly, which runs client-style lawyer-hiring questions across ChatGPT, Gemini, Perplexity, Google AI Overviews, and Google AI Mode on a recurring schedule and stores every answer it collects. He is not a lawyer; he measures what AI assistants say, with receipts. The definitions on this page are the ones the engine implements: the documentation of a measurement, not a certification of anyone's lawyering.
Briefly measures what AI assistants say. It does not rank, rate, or endorse attorneys.
Sources
- Paraphrase Brittleness in Production Retrieval-Augmented Commercial Recommendation (arXiv, 2026): https://arxiv.org/abs/2605.27440
- LLM Whisperer, Lin, Gerchanovsky, Akgul, Bauer, Fredrikson, Wang (Carnegie Mellon-led), CHI 2025 (arXiv 2406.04755): https://arxiv.org/abs/2406.04755
- Perplexity vs ChatGPT vs Gemini: AI Citations, Whitehat SEO (2026): https://whitehat-seo.co.uk/blog/ai-engines-comparison-citations
- Related on this site: AI Search Statistics for Law Firms · How to track when ChatGPT recommends your law firm