Practice Areas
AI for Personal Injury Lawyers: Demand Letters, Medical Records and EvenUp
AI personal injury lawyers use EvenUp, Supio and Eve for medical chronologies and demand packages. Wadsworth and Lacey show why a QC checklist still decides.
On this page
- The plaintiff-side vendors by the numbers
- Medical records: from four thousand pages to a chronology
- The demand package and the number the model must not choose
- Wadsworth and Lacey: the tool is not the control
- The demand-letter QC checklist
- Intake, volume and the client on the other end
- Where this leaves a PI firm
EvenUp raised $150m at a valuation above $2bn in October 2025 and says it processes about 10,000 cases a week for more than 2,000 firms; Supio and Eve sell the same medical-chronology and demand-package work. Personal injury is the practice area where vertical AI is most mature and most bought. It is also where the best-known AI sanction fell on a firm using its own internal tool, which is why the value of these products depends on the quality-control step that follows them, not on the products themselves.
The plaintiff-side vendors by the numbers
| Vendor | Funding and scale (vendor-reported unless stated) | Core product | Note |
|---|---|---|---|
| EvenUp | $150m Series E at $2bn+ (October 2025); $385m raised; 2,000+ firms; ~10,000 cases/week; "$10bn+ damages claimed" | Demand packages; MedChrons medical chronologies; AI Playbooks that re-read files as they grow | Claims "+69% likelihood of hitting policy limit"; unaudited |
| Supio | $60m Series B (May 2025); $91m total; backed by Thomson Reuters Ventures | Medical-record review, chronologies, demand letters with Westlaw integration; an "expert-reviewed" demand tier with in-house QA | Claims "437 hours saved across six cases" |
| Eve | $103m Series B at $1bn+ (September 2025); $164m total; 1,200+ plaintiff firms | Broadest single-vendor workflow from intake to litigation | Category leader by adoption per Vaquill's 2026 comparison |
Two facts frame the buying decision. Vaquill's comparison cites a 2026 survey of more than 300 PI firms in which over half spend under $5,000 a year on AI, which does not fund several point tools; and the products are sold per matter, per case or per seat through sales, so a small firm should price a business-tier general model with a good prompt set against them. A six-lawyer San Francisco firm, Ad Astra Law Group, chose not to replace a departing eighth-year associate and instead built AI into its drafting; Above the Law reported in February 2026 that staffing costs fell 27% and profits rose. Katy Young's line on complaints: "It used to take two days to draft a complaint, now it takes me two and a half hours." The wider small-firm picture is in AI for small law firms and the pricing anchors in what legal AI really costs.
Medical records: from four thousand pages to a chronology
Medical records are the raw material of a PI claim and the task the models do best. Vals AI found AI ahead of lawyers on extraction and document question-answering as early as February 2025, and a medical chronology is extraction with a date column. The workflow:
- OCR before upload. Hospital records arrive as scans of mixed quality. Claude and ChatGPT read scans by vision, but for dates, dosages and codes a text layer is safer; Claude is the model most likely to write "illegible" rather than invent a value. The method is in extracting data from PDFs and scans.
- Chronology with page references. One row per encounter: date, provider, facility, complaint, diagnosis, treatment, medication, cost, source page.
- Gap and causation flags. Gaps in treatment longer than a set period; references to prior injuries or pre-existing conditions; inconsistencies between the patient's reported mechanism of injury across providers.
- Specials schedule. Billed amounts by provider, with page references, for the damages table.
- Human check. A paralegal or nurse consultant ticks every entry against the record. Nothing in the chronology is quoted in a demand until it has been checked.
You are assisting a personal injury lawyer. Work only from the uploaded
medical records. Do not infer diagnoses, dates or amounts not shown.
1. Chronology: one row per encounter with date; provider; facility;
presenting complaint (quoted); diagnosis or impression (quoted);
treatment and medication; billed amount if shown; source document
and page. Sort by date.
2. Flags, each with a page reference: gaps in treatment over 30 days;
any reference to a prior injury, prior claim or pre-existing
condition affecting the same body part; any difference between
providers in the description of how the injury occurred; any
record that is undated or illegible.
3. Specials: a table of billed amounts by provider with pages.
Do not assess causation or the value of the claim.
The demand package and the number the model must not choose
A demand package is liability facts, the chronology, the specials, the general-damages narrative and the number. The vertical products assemble the first four at scale; a general model does the same for a smaller docket. The rule that keeps the practice honest is that the lawyer sets the number. A model asked to "value this claim" will produce a confident figure from nothing, and a demand built around that figure is a demand the lawyer cannot defend in a mediation. The model's job is to make the record around the lawyer's number complete, consistent and readable.
Anthropic's open-source litigation plugin, part of Claude for Legal, shows how a guardrail can be built into the tool: its demand-letter skill will not send until a human confirms the settlement-privilege framing (a Rule 408 gate in US terms), and send gates apply to every demand and takedown. Anthropic also says Claude Sonnet 5 (30 June 2026), the default model on its Free and Pro plans, brought "legal research and analysis gains" on plaintiff-law tasks; that is a vendor claim, and the QC checklist below does not change because of it.
Wadsworth and Lacey: the tool is not the control
Two sanctions are quoted to every plaintiff lawyer, and both involved paid or internal tools rather than a free chatbot.
In Wadsworth v. Walmart (D. Wyo., 24 February 2025) eight of nine cases cited in a motion by Morgan & Morgan lawyers were fabricated. The citations came from the firm's own internal AI platform. Judge Rankin fined Rudwin Ayala $3,000 and revoked his pro hac vice admission, and fined T. Michael Morgan and Taly Goody $1,000 each, writing that "a fake opinion is not 'existing law'". The episode followed Morgan: in May 2026 a Massachusetts judge denied him pro hac vice admission, finding "no showing that he has learned from his mistakes".
In Lacey v. State Farm (C.D. Cal., 6 May 2025) a special master imposed $31,100 on Ellis George and K&L Gates after nine of 27 citations in a brief proved wrong; the tools were CoCounsel, Westlaw Precision and Google Gemini. A year later the Fifth Circuit fined counsel $2,500 in Fletcher v. Experian (18 February 2026) for 16 fabricated quotations traced to vLex and CoCounsel, and remarked that the problem "shows no sign of abating"; the sanctions tracker has the detail. The lesson for a firm buying EvenUp or building on a general model is identical: no product, internal or external, substitutes for a lawyer opening every case cited. The full record is in AI hallucination sanctions and the method in the citation-checking protocol.
The demand-letter QC checklist
Every demand, whether from a vendor platform or a general model, goes through this before it is sent.
Facts and record
- Every date, provider and diagnosis in the chronology checked against the record page cited.
- Every billed amount in the specials checked against the bill; totals recomputed by hand or spreadsheet, not by the model.
- Mechanism of injury stated consistently with the police report, the client's statement and the first medical record.
- Gaps in treatment and pre-existing conditions addressed, not omitted.
Law and citations
- Every case, statute or regulation opened and read by the signing lawyer; any
[verify]or unverified tag resolved. - Jurisdiction-specific rules (limitation, comparative fault, damages caps, policy-limit demand requirements) confirmed from primary sources, not from the model.
Numbers and negotiation
- The demand figure chosen and documented by the lawyer, with the reasoning on file.
- Policy-limit language and time-limited demand mechanics confirmed against the applicable state rule.
- Settlement-privilege framing present where required.
Confidentiality and client
- No client medical data has passed through a consumer-tier tool; vendor terms on training and retention checked.
- The client has been told AI was used in preparing the package, consistent with the firm's disclosure policy.
- Send decision made by a lawyer, recorded on the file.
Intake, volume and the client on the other end
PI firms run on volume, and intake is where an agent pays for itself first: qualifying the caller, running a conflict check, collecting the accident facts and the insurer details, and opening the matter. The design and the traps are in AI client intake. Two constraints apply. The Clio Legal Trends Report 2025 found 78% of clients want to be told when AI is used, 35% of lawyers rarely disclose, and 36% of clients would trust an AI-using lawyer less, so the disclosure sits in the engagement letter and the intake script; see telling clients about AI. And an intake agent that speaks to injured, often uninsured members of the public needs a hard stop before anything that reads as advice, and a human on the line for anyone in distress.
Fees are simpler than in hourly practice and still not free of rules. A contingency fee does not fall because the chronology took an hour instead of a day, and Virginia's LEO 1901 (November 2025) confirms a reasonable value-based fee need not be cut for AI time savings; but California's 2026 guidance treats general AI subscriptions as overhead that cannot be passed through, and no jurisdiction permits a mark-up on AI costs without informed written consent. The detail is in billing clients for AI.
Where this leaves a PI firm
- Decide the buy-or-build question on docket size: a steady demand flow justifies a vertical platform; a smaller docket runs on a business-tier model, the chronology prompt above and the QC checklist.
- Put the OCR step and the "every entry against the page" check into the file-opening procedure.
- Fix in writing that the lawyer sets the demand figure and a lawyer makes the send decision.
- Adopt the QC checklist as a form on every file, signed by the reviewer.
- Train the whole team, including intake staff, on the difference between a tool that drafts and a lawyer who decides. Module 4 of the AI Lab for Lawyers course builds an intake-and-triage agent with these approval gates live, using browser tools only.
The plaintiff bar adopted AI early because the economics were obvious: more demands, faster, from the same staff. The sanctions show what happens when the review step is assumed rather than performed. The checklist is the difference.
Frequently asked questions
Is EvenUp worth it for a personal injury firm?
For a firm running a steady flow of pre-litigation demands, the demand-package products (EvenUp, Supio, Eve) remove most of the assembly work and are priced per matter or per seat through sales. For a smaller docket, a business-tier general model with the chronology prompt and a QC checklist covers much of the same ground. A 2026 survey of 300-plus PI firms found more than half spend under $5,000 a year on AI, which rules out stacking several tools.
Can AI review medical records for a personal injury case?
Yes, and it is the strongest use case. After an OCR pass, a model can build a chronology of every visit, provider, diagnosis, treatment and cost with a page reference for each entry, flag gaps in treatment and note pre-existing conditions. The lawyer or a nurse consultant checks every entry against the record before it goes into a demand or an expert brief.
Can AI write a demand letter?
It can assemble a first draft from the chronology, liability facts and the damages figure the lawyer has set, and vertical products do this at scale. What it cannot do is decide the number, judge the liability picture or be trusted with case citations. Run every draft through a QC checklist, verify any authority cited, and keep the send decision with a lawyer.
Do the AI sanction cases apply to plaintiff firms using their own tools?
Yes. In Wadsworth v. Walmart the fabricated cases came from Morgan and Morgan's internal AI platform, and in Lacey v. State Farm the tools were CoCounsel, Westlaw Precision and Gemini. Courts hold the signing lawyer responsible regardless of the tool, and one of the Wadsworth lawyers was later refused admission in another state because the judge saw no sign he had learned from the episode.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Use Cases
AI Client Intake: Agents That Qualify, Conflict-Check and Onboard
AI client intake agents qualify leads, run conflict pre-checks, collect KYC documents and draft engagement letters. The flow, the fields and the guardrails.
8 min read
Playbooks
Extract Data from PDFs and Scans with AI: Contracts, Invoices and Court Files
How lawyers extract data from PDFs and scans with AI: OCR first, a fixed schema, a full lease-extraction prompt, batch runs in Cowork or n8n, and sampling.
11 min read
Regulation & Ethics
AI Hallucination Sanctions: The Cases Every Lawyer Should Know (2023–2026)
AI hallucination cases reached 2,041 in Charlotin's database by September 2026, 811 involving lawyers. Every sanction from Mata to Couvrette's $110,204, with lessons.
12 min read
Playbooks
How to Verify AI Legal Research: A Citation-Checking Protocol
Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.
11 min read
Guides
AI for Small Law Firms and Solo Practitioners: What to Use and What to Skip
AI for small law firms in 2026 means a business-tier model, one legal tool and simple automation for under €200 a month, plus the policy and privacy settings most solos skip.
9 min read
Regulation & Ethics
Billing Clients for AI: What You Can Charge and What You Cannot
Lawyers may bill actual time spent using and checking AI or a fixed fee agreed in advance, but not saved hours, subscriptions or marked-up AI costs. Worked numbers inside.
10 min read