Seven prompts, tested against real demand letter workflows — here’s what each one does, where to adjust it, and the one rule that applies to every citation it produces.
Demand letters are high-stakes but formulaic enough that AI models can carry a significant share of the drafting load. The problem is that generic prompts produce generic letters — flat tone, missing damages math, zero anticipation of the other side’s response. These seven prompts are built around the specific jobs a demand letter needs to do: tell a compelling factual story, calculate and justify damages, calibrate tone, anchor legal claims, frame a settlement range, anticipate the counterargument, and pass a final quality check. They run in any major chat model — Claude 3.5 Sonnet, GPT-4o, or Gemini 1.5 Pro all produce usable output. Each prompt is modular, so you can run them in sequence as a full drafting workflow or pull individual ones as needed.
1. Facts-into-Narrative
Use this first. Drop in your raw intake notes — dates, events, names, amounts — and get back a coherent chronological narrative built for a demand letter’s opening section. The model organizes the facts into a readable sequence and flags any apparent gaps where a date or amount is missing. Expect one to three paragraphs. Trim anything that softens the client’s position without adding credibility.
You are a legal drafting assistant. I will give you a set of unorganized facts about a dispute. Your job is to turn them into a clear, chronological factual narrative suitable for the opening section of a demand letter.
Rules:
- Write in third person (referring to my client as "our client" or by name if I provide one).
- Be specific: include all dates, dollar amounts, and names I provide.
- Do not add facts I have not given you.
- If you notice a gap — a missing date, an unclear sequence, an unspecified amount — flag it with [MISSING: describe the gap] rather than inventing a detail.
- Tone: factual and professional. Save persuasive framing for the legal claims section.
- Output: the narrative paragraphs only, no commentary.
Here are the facts:
[PASTE INTAKE NOTES HERE]Notes: Replace “our client” with the actual party name before sending. If you’re in a jurisdiction where the demand letter will be attached to a complaint, you may want to ask the model to mirror the factual allegations structure you use in pleadings — just add that instruction at the end of the rules list.
2. Damages Calculation Breakdown
This prompt takes a list of claimed damages — actual losses, consequential damages, fees, interest — and builds a numbered breakdown with subtotals and a grand total. It’s useful for making sure you haven’t left a category on the floor and for producing the kind of itemized damages section opposing counsel actually has to respond to line by line.
You are a legal drafting assistant helping structure a damages section for a demand letter.
I will provide a list of damages categories and amounts. Your job is to:
1. Organize them into a numbered list with a one-sentence explanation of each category.
2. Show a subtotal for each category where there are multiple items.
3. Provide a grand total.
4. Flag any category where the calculation method is unclear or where I have not provided a number — use [NEEDS AMOUNT: category name].
5. Do not invent dollar figures. Use only the numbers I provide.
6. If I specify a prejudgment interest rate and accrual start date, calculate and include accrued interest as a line item.
Output format: numbered list followed by a totals section. No narrative paragraphs — just the structured damages breakdown.
Damages information:
[PASTE DAMAGES CATEGORIES AND AMOUNTS HERE]
Prejudgment interest rate (if applicable): [RATE]
Interest accrual start date (if applicable): [DATE]Notes: Prejudgment interest rules vary by jurisdiction and claim type. The model will do the arithmetic, but you need to confirm the applicable rate and accrual rules for your state before the number goes in a letter. Treat the model’s interest figure as a calculator output, not legal authority.
3. Tone Modulation
The same factual record and legal claims can support a letter that reads as measured and professional, one that reads as pointedly aggressive, or one that reads as cordial and settlement-oriented. Each serves a different strategic moment. This prompt takes a draft letter and rewrites it in whichever register you specify without changing the substance.
You are a legal drafting editor. I will give you a demand letter draft and a target tone. Rewrite the letter to match that tone while preserving all factual claims, legal claims, damages figures, and deadlines exactly as stated.
Target tone options — choose one and specify it:
- FIRM-PROFESSIONAL: Direct and confident, no hedging, no hostility. Reads like a lawyer who expects to be taken seriously and doesn't need to raise their voice.
- AGGRESSIVE: Sharp and unambiguous. Makes clear litigation is imminent and the client is prepared for it. No threats beyond what the legal claims support, but no softening language.
- CORDIAL: Collegial and solution-oriented. Acknowledges the other party's potential perspective. Leaves the door open for a quick resolution. Still firm on the legal position — just warmer in delivery.
Do not add legal claims or facts I have not included. Do not remove any claim, damages figure, or deadline. If a sentence in the original cannot be made to fit the target tone without distorting its meaning, flag it with [TONE CONFLICT: original sentence] and leave it unchanged.
Target tone: [CHOOSE ONE]
Draft letter:
[PASTE DRAFT HERE]Notes: Run this after prompts 1 and 2 have given you a working draft. The AGGRESSIVE version occasionally produces phrasing that overshoots — read it carefully before sending. The CORDIAL version is useful when the recipient is a long-standing client of your client’s, or when early resolution is the actual goal.
4. Case-Citation Suggestions
This prompt asks the model to suggest relevant legal authorities — case law, statutes, and restatement sections — that could support the claims in the letter. It is the most useful and the most dangerous prompt in this collection. Use it to surface research directions quickly. Do not use any citation in a letter without pulling the actual case and confirming it says what the model says it says. AI models hallucinate citations with confidence. This is not a warning to ignore.
You are a legal research assistant. I will describe the legal claims in my demand letter. For each claim, suggest up to three legal authorities — cases, statutes, or restatement sections — that commonly support that claim in [JURISDICTION].
Format your response as:
Claim: [claim name]
Suggested authorities:
1. [Citation] — one sentence describing what it stands for.
2. [Citation] — one sentence describing what it stands for.
3. [Citation] — one sentence describing what it stands for.
IMPORTANT: You must include this warning line before every citation block, verbatim:
"⚠ AI-generated citations require independent verification. Do not use any citation in a document without pulling and reading the source."
Do not fabricate citations. If you are uncertain whether a case exists or whether it supports the point, say so explicitly rather than providing a citation you cannot confirm.
Jurisdiction: [STATE/FEDERAL]
Claims in my demand letter:
[LIST CLAIMS HERE, e.g., "breach of contract — failure to pay invoices," "conversion of client property," "unjust enrichment"]Notes: Jurisdictional adaptation matters most here. A contract damages case that’s solid authority in Texas may be distinguishable or simply unknown in Oregon. Always specify the jurisdiction. Then verify every single citation in Westlaw, Lexis, or Fastcase before it touches a document. If the model says “I am uncertain whether this case exists” — believe it and go research from scratch.
5. Settlement-Range Generator
This prompt doesn’t predict outcomes — it asks the model to reason through a realistic settlement range based on the damages figures and claim strengths you provide, using factors you specify. The output is a starting-point framework for your own analysis, not a valuation. It’s useful for drafting the implied “or we go to court” calculus in the letter, and for your own pre-letter client conversation about realistic expectations.
You are a legal strategy assistant helping a lawyer think through a settlement range for a demand letter matter.
I will provide: claimed damages, the strength of each claim (I will rate as strong / moderate / uncertain), and any relevant factors (insurance involvement, defendant's apparent financial position, litigation cost estimate, jurisdiction).
Based on these inputs, reason through:
1. A likely floor (minimum realistic settlement, accounting for litigation risk and cost).
2. A likely ceiling (full damages if claims succeed, adjusted for collectability).
3. A reasonable demand figure to put in the letter, with brief reasoning.
Present your reasoning step by step. Flag any input where I have not given you enough information to reason reliably — use [NEEDS INPUT: describe what's missing].
Note: This is a reasoning framework, not a legal opinion or a prediction of outcome. The lawyer must apply their own judgment and jurisdiction-specific knowledge.
Total claimed damages: [AMOUNT]
Claim strength ratings: [LIST EACH CLAIM AND RATE IT]
Litigation cost estimate (through trial): [AMOUNT OR RANGE]
Other relevant factors: [DESCRIBE]Notes: This prompt works best when you give it accurate claim-strength ratings rather than optimistic ones. If you rate everything “strong,” the model will give you an inflated ceiling that won’t survive a client conversation. The framework is more useful than the specific numbers — read the reasoning, not just the output figures.
6. Response Anticipation
Before the letter goes out, this prompt simulates the opposing party’s likely response — the defenses they’ll raise, the facts they’ll dispute, and the counterclaims they might assert. It’s a one-pass red-team review. The output helps you decide whether to address a likely defense in the letter itself or leave it for the reply, and whether any facts need to be shored up before you send.
You are playing the role of opposing counsel reviewing this demand letter on behalf of the recipient. Your job is to identify:
1. The three to five strongest defenses or factual disputes the recipient is likely to raise in response.
2. Any counterclaims or affirmative claims the recipient might assert.
3. Any weaknesses in the sender's factual narrative that opposing counsel would target.
4. Any legal claims in the letter that appear vulnerable to a motion to dismiss or a simple factual rebuttal.
Format: numbered list per category above. Be specific — reference the actual claims and facts in the letter. Do not be diplomatic; the goal is to surface the hardest challenges before the letter is sent.
After the analysis, provide a one-paragraph summary of the letter's overall defensibility from an opposing counsel perspective.
Demand letter:
[PASTE FINAL DRAFT HERE]Notes: This prompt produces the most variable output of the seven — quality depends heavily on how much factual and legal substance is in the draft you feed it. A thin draft produces a thin red team. If the model identifies a vulnerability you hadn’t considered, that’s the prompt working. Don’t dismiss it because the output is uncomfortable.
7. Revision Pass Against a Checklist
The final prompt runs the letter against a demand-letter checklist before it goes out. It catches missing elements — no response deadline, no statement of consequences, unclear damages total — and flags anything that reads as inconsistent or contradictory across sections. Think of it as a proofreader with a checklist, not a substantive editor.
You are a legal document reviewer. Review the following demand letter and check it against this checklist. For each item, note: PRESENT (and where), MISSING, or FLAG (if present but unclear or potentially inconsistent).
Checklist:
1. Identification of sender and recipient with correct party names.
2. Statement of the legal basis for each claim.
3. Itemized damages with a stated total.
4. Specific response deadline with a calendar date (not just "within 30 days").
5. Statement of the consequences of non-response (e.g., commencement of litigation).
6. Preservation demand or litigation hold notice, if appropriate.
7. Instructions for response (to whom, by what method).
8. Consistent use of party names throughout.
9. No factual statements that contradict each other across sections.
10. Closing signature block with attorney name and contact information.
After the checklist, provide a short list of any edits needed before the letter is ready to send. If the letter passes all ten items cleanly, say so.
Demand letter:
[PASTE DRAFT HERE]Notes: Add or remove checklist items to match your firm’s standard. If your jurisdiction requires specific statutory language in demand letters — consumer protection claims in some states, for example — add that as item 11. The model will check for whatever you put on the list.
Notes on Using These Prompts
All seven prompts run in Claude 3.5 Sonnet, GPT-4o, or Gemini 1.5 Pro without modification. Claude tends to produce cleaner structured output; GPT-4o tends to be slightly more verbose in its reasoning. Neither difference is large enough to matter much — pick the model you already have access to.
Jurisdictional adaptation is your responsibility throughout. The prompts flag where jurisdiction matters, but the model’s defaults will lean toward majority-rule positions or common-law standards that may not match your state. Specify your jurisdiction in every prompt that involves legal claims, and treat anything jurisdiction-specific in the output as a starting point for your own research.
The citation warning in prompt 4 applies to every prompt in this collection. Any time a model produces a case name or statutory citation — even in passing, even in a prompt that isn’t specifically about research — verify it before it goes in a document. This is not a sometimes rule.
Run the prompts in order — 1 through 7 — when drafting from scratch. When revising an existing draft, prompts 3, 6, and 7 are the ones worth pulling individually. The full sequence takes roughly 25-40 minutes of active time for a standard single-claim demand letter, compared to 90 minutes or more drafting cold. That estimate holds on matters with clean facts. When the facts are messy, the time savings shrink — the model can’t organize what you haven’t organized yourself.
Related reading
- 10 ChatGPT Prompts Every Solo Lawyer Should Save (Tested on Real Matters)
- The 5-Prompt Sequence for First-Pass Contract Review with Claude or GPT
- 10 Claude Prompts for Faster Discovery Document Review
- Document Automation with Claude and Microsoft Word: A Walkthrough for Small Firms
- Briefpoint Tested: AI Discovery Drafting for Solo and Small-Firm Litigators