AI demand letter tools draft settlement demands from your case facts in minutes instead of hours, and the right pick comes down to volume: buy a tool like EvenUp at moderate personal injury volume, build your own generator at hundreds of demands a month, and run a $20 Claude Project below 10 a month.
My lemon law firm sends demand letters at a volume that would break a normal drafting process. Every one of them starts as an AI draft. An attorney reviews and signs every one before it leaves the building. That system took me one build to set up, and it’s the highest-ROI piece of AI in the entire firm.
Here’s why demand letters are where you should point AI first, what the tools cost, and when building your own beats buying.
Why demand letters are the perfect AI use case
Most legal drafting is a bad fit for automation because every document is different. Demand letters are the opposite.
The structure never changes. Facts, liability, damages, demand, deadline. The bones of letter number 400 are identical to letter number 1. Only the facts move: the client, the vehicle or the injury, the dates, the numbers.
Volume is high. A plaintiff firm doing real intake sends demands weekly or daily, not quarterly. Anything you do that often is worth systematizing, and the same logic runs through legal document automation generally and contract drafting specifically.
And quality directly moves money. A demand letter that cites the right statute, itemizes damages cleanly, and lays out repair history in a tight chronology settles higher and faster than a sloppy one. Insurance adjusters and manufacturer counsel triage incoming demands. The organized ones get taken seriously. So this isn’t a cost-cutting play. It’s a revenue play that happens to also cut cost.
The buy options
Four tools dominate the conversation right now.
EvenUp is the market leader for personal injury. You upload medical records and case documents, and it produces a demand package with a medical chronology and damages analysis baked in. The medical record analysis is the real product; the letter is almost a byproduct. It fits PI firms where every case has a stack of treatment records someone would otherwise read manually. The catch is pricing: EvenUp charges per demand, and firms report fees in the hundreds of dollars per letter. At low volume that math works. At high volume it becomes a tax on your own caseload.
Clearbrief lives inside Microsoft Word and links every factual assertion in your draft to the underlying evidence. Hover over a sentence, see the exhibit. It’s less a demand letter generator than an evidence-grounded drafting layer, and it shines for litigators who already draft in Word and want receipts attached to every claim. Firm pricing runs per seat, per year.
Filevine Demands AI is the obvious pick if your firm already runs on Filevine. It pulls case data straight from your existing files, so there’s no re-keying. Don’t switch case management systems to get it, but if you’re on Filevine anyway, it’s the shortest path.
ProPlaintiff is the budget entrant, aimed at smaller PI shops that want AI demand drafting without EvenUp’s price tag. Less polished, cheaper, worth a demo if you send fewer than 20 demands a month.
Who should buy: a PI firm sending 10 to 40 demands a month with heavy medical records is EvenUp’s sweet spot. Litigation-heavy practices that live in Word should look at Clearbrief. Everyone on Filevine should turn on Demands AI and test it this week.
The build option, which is the path I took
At lemon law volume, per-demand pricing collapses. Pay $300 a letter on hundreds of letters a month and you’ve hired an invisible software vendor at a salary your best paralegal would envy. So I built my own generator with Claude.
The architecture is simpler than it sounds. Four pieces:
A template with variables. Our standard demand letter with slots where the case-specific content goes: client name, vehicle, purchase date, repair visits, damages, statute.
The case facts. Intake data and repair history flow in from our systems, plumbing that MCP now makes far easier to wire up. This is the part that took real work, because the letter is only as good as the data feeding it.
The firm’s voice. Three of our best past demand letters ride along with every generation as tone examples. The AI matches them. Output sounds like us, not like a chatbot.
A pre-output checklist the AI runs on its own draft. Every repair date from the record appears in the letter. Damages math is shown and adds up. The statute cited matches the client’s state, which matters when you practice in all 50. If any check fails, it flags the draft instead of passing it along.
Then a human attorney reviews, edits if needed, and signs. Drafting time per letter went from more than an hour to minutes of review. That’s the whole pitch.
If your firm has the volume, this is buildable. I covered how non-programmers ship tools like this in the Claude Code guide. Mine was built by describing what I wanted in English.
The middle path for smaller volume
Under 10 demands a month? Don’t build anything and don’t buy anything. Use a Claude Project.
Create a Project in Claude. Upload your three best past demand letters, the ones that settled well. Upload your template. Add instructions: “Draft demand letters in the style of the examples, following the template structure. Never invent facts. Flag anything missing.” Then paste in your case facts and ask for a draft.
That’s 80% of what my custom system does, for the $20 you’re already paying. The drafting prompt matters more than people expect, and the prompt guide has a demand letter prompt you can lift directly.
Here’s the whole decision in one table:
| Path | Fits | Cost | Setup |
|---|---|---|---|
| Buy (EvenUp, Clearbrief, Filevine) | PI firms, 10 to 40 demands/mo | Per demand or per seat, often hundreds per letter | Days |
| Build your own | High volume in one practice area | One build project, then near zero per letter | Weeks |
| Claude Project | Under 10 demands/mo | $20/mo | 30 minutes |
Quality control, or the part that keeps your license
Non-negotiable rules, regardless of which path you pick:
The attorney always reviews. Every letter, every time. AI drafts, humans sign. No demand leaves the firm without a lawyer reading the whole thing.
AI never sends. There is no automation between “draft generated” and “letter mailed.” A human is the gate.
Damages math gets double-checked by a person. AI arithmetic has improved but a demand with wrong math costs you credibility with the exact adjuster you’ll see again next month.
Never let AI invent facts. This is the actual risk, not bad prose. If a repair visit isn’t in the record, it doesn’t go in the letter. If treatment didn’t happen, it didn’t happen. Build your prompt to say “flag missing information” instead of letting the model fill gaps, because filling gaps is what these models do by default.
Do this today
Pull your single best demand letter from the last year. The one that settled fast, at a number you were happy with. Load it into a Claude Project as the gold standard, add your template, and draft your next real demand against it. Compare the AI draft to what you would have written from scratch. That comparison, on a real case, tells you more than any vendor demo will.