Case study · Process
Case Results Audit
Reading a law firm's case results page, counting what's on it, adding up the amounts and writing the assessment took one to two hours per firm. The work that actually needed a person was about ten minutes of it.
Confidential mockup — all data fabricated
A pricing audit answers a narrow question for a prospective client: what proof of outcomes does this firm already publish, and what would it take to make it worth reading? Someone had to open the site, find the results page if one existed, count the entries, sum the amounts, group them by practice area, and write it up against a fixed template.
Almost none of that is judgment. The judgment is in the last two sections — what's working, and what to do about it — and everything before it is retrieval and arithmetic that a person does slowly and inconsistently.
The firm below is invented, and so is every figure in it. The structure, the two-stage split and the template are from the real workflow.
01 — What the audit covers
The template has eight sections and never changes: the header block, the entry count, the summed amount, the aggregate-claim check, the overview, the practice-area table, strengths, and areas for improvement. Two clients audited a month apart produce documents you can read side by side, which is the part hand-assembly never managed.
02 — Two prompts, not one
Stage one — ask the page
One question can't be answered by reasoning about a website: is an aggregate recovery figure actually displayed, and where. A firm can have recovered fifty million and say so nowhere.
So the first prompt drives the browser. It visits the homepage, the About page and the results page, and reports what it found on each — the wording, the placement, or its absence.
Stage two — write the audit
The second prompt takes that check block as established fact rather than something to infer, along with the firm details and the attached template.
Splitting them is the whole trick. A single prompt asked to both browse and assess will happily assert that a figure appears on a page it never confirmed.
There's a second field on the form: rough notes on how the firm presents its results. It's optional, and leaving it blank is usually the better call — the assessment comes back cleaner when it hasn't been told what I already think.
03 — One firm, start to finish
Six steps, from the intake form to the finished document. Nothing runs on a timer — step through it at your own pace.
- 01 Intake
- 02 Browser pass
- 03 Check block
- 04 Audit prompt
- 05 Extraction
- 06 Report
The whole run, start to finish. Six steps from a firm name to a finished audit document.
Generator — firm & context
Pulled from the Monday board
Stage one — prompt for Claude for Chrome
Pasted back into the generator
Stage two — assembled audit prompt
Attach _Template - Case Results Pricing Task.docx before sending. The generator shows this reminder every time. The template is attached rather than described, so the output structure is identical for every firm instead of being re-described — and re-drifted — each run.
Extraction
- 23 case result entries found on /case-results/
- All 23 carry a dollar amount — nothing qualitative to set aside
- Amounts summed: $25,830,000
- Grouped into 6 practice areas; counts reconcile to 23
- Homepage claim "$25 Million+" is consistent with the summed total
- Aggregate figure is absent from the results page itself — the one page a visitor opens to compare outcomes
- 9 of 23 entries carry no narrative beyond the case type
- No dates, attorney attribution, or venue on any entry
Saved to Drive — Halvorsen & Reid Injury Law — Case Results Pricing Task
Case Results Overview: Halvorsen & Reid Injury Law
Case Results Overview
Halvorsen & Reid Injury Law publishes 23 individual case results at halvorsenreid.example/case-results/, presented as a card grid sorted highest to lowest. Every entry carries a dollar figure, and the published results total $25,830,000. An aggregate claim of "$25 Million+ Recovered" appears in the homepage hero, which the summed total supports — but that claim does not appear on the results page or the About page.
Results by Practice Area
| Practice area | Cases | Total amount |
|---|---|---|
| Trucking collisions | 6 | $12,450,000 |
| Traumatic brain injury | 3 | $6,900,000 |
| Premises liability | 4 | $2,315,000 |
| Motor vehicle collisions | 7 | $1,890,000 |
| Wrongful death | 2 | $1,750,000 |
| Motorcycle collisions | 1 | $525,000 |
| Total | 23 | $25,830,000 |
Strengths
- The dollar amount is the visual anchor on every card, with the case type beneath it, so a visitor can scan for a situation resembling their own without reading dense text.
- Descending order front-loads the strongest outcomes — the top three entries are all seven figures.
- Breadth across six injury types signals a general serious-injury practice rather than a single niche.
- The homepage aggregate claim is substantiated by the published entries, which is not true of every firm making one.
Areas for Improvement
- The aggregate figure is on the wrong page. It runs in the homepage hero and nowhere else. The results page is where a prospect goes to compare firms, and it is the one page that never states the total.
- Nine of 23 entries have no narrative. The entries with a sentence of context are noticeably more persuasive; the bare ones read as filler by comparison.
- No verdict versus settlement distinction. Willingness to try a case is a meaningful signal and the page currently flattens it.
- No dates, attorney attribution, or venue. A prospect cannot tell whether these are recent wins or a decade old, or who handled them.
- No filtering across 23 entries. Someone looking for a trucking result scrolls the whole page to find one.
Firm, figures and findings are fabricated. The two-stage structure and the template are from the real workflow.
04 — What the audits taught
Ask the page, not the model
Whether a recovery figure is displayed is a fact about a rendered page, not something to reason toward. The browser pass answers it and hands the answer to the second prompt as given. A model asked to browse and assess in one breath will assert a figure it never actually saw.
Withhold your own read
The notes field exists, and most of the time it stays empty. Telling the audit what I already think produces a document that agrees with me, which is worth nothing to the person pricing the work. Leaving it blank is the difference between a second opinion and an echo.
A missing results page is the finding
Some firms report outcomes qualitatively because dollar figures don't apply, and some publish no results at all. Neither is a failed audit — the instruction is to report the gap rather than force a total, because "there is nothing here to sum" is what the client is paying to learn.