AI Demand Letters: What They Are and What They Are Worth to a PI Firm

An AI demand letter is not a new kind of legal document. It is a settlement demand letter, the same instrument personal injury attorneys have sent for decades, drafted with an AI assistant doing the heavy lifting on structure and first draft language while the attorney supplies the facts and owns the result. The phrase has caught on because something genuinely changed in the last two years: a competent assistant can now take a clean case file and return a well organized, persuasive draft in minutes instead of hours.

That is why firms are searching for it, and it is why the term is worth understanding precisely. This briefing answers the question a managing partner actually asks before adopting anything: what does it do, what does it cost us, what could go wrong, and do we build it or buy it.

The Problem: A Partner Level Document Built With Associate Level Hours

The demand letter is the most leveraged document in a personal injury practice. It frames liability, organizes the medical story, justifies the number, and sets the anchor that every subsequent negotiation moves from. A strong one moves the settlement; a weak one invites a lowball and a long fight.

Yet most of the work that goes into building one is not partner level judgment. It is assembling the medical chronology, summarizing treatment, tallying special damages, pulling the liability facts into order, and drafting thirteen to seventeen sections in a consistent voice. On a typical file that is six to ten hours between a paralegal and the supervising attorney, and a meaningful share of it is sorting, formatting, and first pass drafting. That is the exact clerical load that erodes the value of a litigator's day and slows down how fast a firm can move files.

How Attorneys Use Claude to Draft a Demand Safely

The model here is simple and it is the only one that holds up to ethics scrutiny. The attorney decides what the case is worth and what the facts are. Claude drafts and organizes the document from those facts. The attorney verifies every line before it leaves the office. Claude never sets the demand number, never invents a fact, and never sees raw client records unless the firm has the right controls in place.

In practice that means the AI receives a sanitized case file: a structured summary the firm prepared, with client identifiers removed or replaced and the medical detail reduced to the facts the demand needs. From that input Claude produces a sorted, sectioned draft. The attorney reads it against the file, fixes what is wrong, sharpens the argument, and signs the work that is theirs. The letter that goes out is a human attorney's letter. The AI shortened the path to it.

A Step by Step Claude Workflow

This is the short version of the workflow taught in depth inside the course. It moves from a clean input to a verified draft in four passes.

Step 1: Sanitize the inputs

Before anything enters the AI session, the firm builds a sanitized case file. Replace the client's name with a placeholder, strip direct identifiers, and reduce the medical records to a structured summary of providers, dates, findings, and treatment rather than pasting raw records. This keeps protected health information and privileged material out of a general purpose tool and protects work product. If the firm needs to send actual protected health information to any vendor, that requires a signed Business Associate Agreement under HIPAA, which is exactly the dependency a sanitized workflow avoids.

Step 2: Build the case file Claude will draft from

Give Claude the facts in one organized brief: parties, liability theory, the medical chronology, special damages with figures you calculated, lost wages, and the demand number you set. The number and the math are yours. Claude organizes and argues from them; it does not generate them.

You are a legal drafting assistant for a personal injury attorney. Below is
a sanitized case file. Draft a settlement demand letter organized into these
sections: introduction and representation, liability narrative, injury and
treatment summary, special damages, general damages, demand and basis,
conclusion. Use only the facts in the case file. Do not add any figure,
diagnosis, or fact that is not stated below. Where the file is missing
something a strong demand would include, list it under a heading called
GAPS rather than inventing it.

[PASTE SANITIZED CASE FILE HERE]

Step 3: Draft in passes, in your voice

Take the structured draft and refine it section by section. Ask Claude to tighten the liability narrative, to make the damages argument more concrete, or to rewrite the opening in a firmer register. The "GAPS" list tells you exactly what facts to go pull before the letter is real. This is where the attorney's judgment shapes the document.

Step 4: Stress test, then verify everything

Run the draft through the eyes of the adjuster who will receive it, then verify every factual claim against the file.

You are now a defense claims adjuster reviewing this demand. Identify the
three weakest points a defense would attack: causation, treatment gaps,
pre-existing conditions, or an overstated number. For each, quote the exact
sentence in the demand that creates the exposure and explain the attack.

Fix what the stress test surfaces, then do the non negotiable final pass: read every figure, date, diagnosis, and citation in the letter against the source file. An AI can produce a clean sentence that contains a wrong number. The attorney catches it, every time, before the demand is sent.

The ROI: Where the Hours Actually Go

The honest version of the math is this. AI does not cut the parts of a demand that require judgment, the valuation, the strategy, the negotiation read. It cuts the assembly and the first draft. On a file where building and polishing the demand runs six to ten hours, firms that have a repeatable workflow report taking several hours off that, mostly out of the sorting, summarizing, and drafting passes. These are self reported practice figures, not a published study, so treat them as a planning estimate and measure your own first ten files.

Across a caseload the compounding matters more than any single file. A firm that can produce a strong demand a few hours faster can either carry more files per attorney or send demands sooner, and in personal injury, time to demand affects time to settlement. That is the commercial case, and it is why the search term carries a $74 cost per click. The buyers know the document is worth money.

Accuracy, Privilege, and Verification Guardrails

Three rules keep an AI demand letter defensible. They are not optional, and a firm that skips them is taking on real exposure.

• Keep client and medical data out of public tools. Sanitize before anything enters the session. Raw protected health information and privileged material stay in your own systems. If you must route real data to a vendor, do it under a signed Business Associate Agreement, not by pasting into a chat window.
• Verify every output against the source. A generative tool can state a damages figure, a date, or a diagnosis that is not in your file. The attorney checks every factual claim in the letter against the case file before it is sent. An unverified AI claim in a legal document is the lawyer's responsibility, not the tool's. The ABA's first formal ethics guidance on generative AI places the duties of competence, confidentiality, and verification on the lawyer using the tool.
• The attorney sets the number. Valuation is judgment and advocacy, not arithmetic a model should originate. Claude organizes and argues your figure. It does not decide what the case is worth.

What AI Does Not Replace

An AI demand letter is a drafting accelerator, not a strategy. It will not read your adjuster, judge whether to anchor high or stay credible, decide when to hold a demand for one more treatment record, or carry the negotiation that follows. It does not establish liability, it does not value pain and suffering, and it does not exercise the professional judgment that is the entire reason a client hired a lawyer. The firms that get the most out of this treat the AI as a fast, tireless associate who drafts well and never tires of revisions, supervised by an attorney who still owns every decision that matters.

Build the Skill or Rent the Tool

This is the real decision behind the search. Purpose built platforms such as the AI legal tools the ABA tracks, EvenUp among them, will draft demands and benchmark them against claim databases. For very high volume shops that want comparable valuation data, that can be worth the per case price. For most personal injury practices it is not, because two costs add up: you pay for every file, and your most sensitive workflow lives inside someone else's pipeline.

Training your own team inverts both. Once your attorneys and paralegals can run the workflow, the marginal cost of the next demand is a few minutes of prompting, not another invoice. The work product stays under your controls. You adapt the prompts to your jurisdiction and your style, and you are never waiting on a vendor's queue the week a demand is due. That is the premise of The Leveraged Attorney: build the capability into your people so the leverage is yours to keep.