AI can produce a usable first draft of a demand letter, engagement letter, or client email in under a minute. That speed is the point. But speed means nothing if the document goes out with a wrong number, a misattributed fact, or a placeholder you forgot to fill in.
The review step is not optional. It is also not long. Here is what to check.
1. Verify every number
Dollar amounts, dates, percentages, statute of limitations deadlines. These are the facts most likely to be wrong in an AI-generated document.
The model works from what you gave it in the prompt. If you typed a number incorrectly, or the model inferred a figure from ambiguous input, the document will reflect that error.
Example: You typed $12,500 in medical specials in the prompt, but the actual total from the billing records is $12,050. The demand letter goes out with the inflated number. The adjuster catches it before you do and it costs you credibility on the whole claim.
Check every number against your source material before the document leaves your desk.
2. Read the placeholder fields
Any prompt worth using has placeholder fields like [CLIENT NAME], [DATE], or [FIRM NAME]. AI output occasionally includes unfilled placeholders, especially if the input was incomplete.
Example: You filled in the client name and matter type but left the fee amount blank because you had not decided yet. The draft goes out with [FEE RATE] still in the fee arrangement paragraph. The client notices before you do.
Scan the full document for brackets before sending. Do not assume they are all filled in because most of them are.
3. Confirm the facts trace back to your file
AI models can hallucinate. In a legal context this means a document that reads correctly but attributes a fact to the wrong source, or includes a detail that was never in your notes.
Example: Your intake notes say the accident happened on March 12. You paraphrased the facts loosely in the prompt and the model picked up "mid-March" and rendered it as March 15 in the demand letter. You send it. The police report says March 12. Now opposing counsel has a discrepancy to work with.
For demand letters, verify that each factual claim about the incident, injuries, and damages matches what is actually in the file.
4. Check the scope of representation
Engagement letters need to accurately describe what you are and are not handling. AI will draft reasonable-sounding scope language based on what you provided, but it will fill gaps with generic language.
Example: The client hired you to handle the negotiation and settlement only, not litigation. You described the matter as a "personal injury claim" without specifying the scope limit. The draft says you are representing the client "through resolution of all claims arising from the incident." That language is broader than what you agreed to and creates an expectation problem.
Read the scope section carefully. If the letter says you are handling "all aspects of the matter" and you only agreed to part of it, that language matters.
5. Look at the tone
AI drafts tend toward formal and neutral. That is usually fine, but not always right.
Example: Your client is a 67-year-old retired teacher who was rear-ended at a stoplight. The liability is clear, the injuries are documented, and the adjuster you are dealing with is reasonable. The AI draft opens with: "We write to present the claim of our client." That is technically correct and completely flat. A more direct opener lands better: "Your insured caused a cervical fracture. Demand: $85,000."
Adjust the tone before sending. This takes two minutes and it is part of the work.
6. Flag anything marked for attorney review
If you are using any of the tools on this site, the output includes an explicit attorney review flag for ambiguous terms or missing information. Do not skip these.
Example: The engagement letter tool returns this in the fee section: "Attorney will bill at a rate of [ATTORNEY REVIEW REQUIRED] per hour." That marker means you did not specify a rate in the intake form and the model left it blank rather than inventing a number. It is doing you a favor. An engagement letter that goes out without a fee rate is not an engagement letter.
The missing_info field in the intake and deposition tools works the same way. Those flagged items are the things most likely to cause a problem later.
How long this takes
For a demand letter or engagement letter, this review should take five to ten minutes, not an hour. The point is not to rewrite the draft. It is to catch the specific categories of errors that AI gets wrong: numbers, placeholders, hallucinated facts, and missing scope.
If you find yourself rewriting most of the document, the input prompt was probably too vague. Go back and add more detail to the prompt rather than fixing the output by hand.
The right mental model
Think of the AI draft the same way you would think of a first draft from a junior associate. Read it critically, verify the facts, fix the tone, and send it. The value is in not starting from a blank page, not in skipping the review entirely.
The tools on this site are built with that assumption. The output is a starting point, not a finished work product.