AI writing tools can produce a clean first draft of a letter in seconds, but the sender stays responsible for every fact, date, and promise in it. Canada's privacy regulators advise keeping personal information out of prompts where possible, and a 2023 U.S. court sanction showed what happens when machine-written text is sent without checking. The safe pattern is to write down the facts first, let the tool shape the wording, and verify the result line by line.
Key takeaways
- AI is useful for tone, structure, and trimming a letter, and unreliable for facts, dates, and legal citations it was not given.
- Canadian privacy regulators advise using anonymized or de-identified information in prompts where possible.
- A 2023 sanctions order in Mata v. Avianca penalized lawyers USD 5,000 for filings that contained invented case citations.
- Letters with legal or financial consequences need a human check against the original documents before sending.
What AI does well in a letter
A language model predicts likely text from the instructions it receives. That makes it good at jobs where wording matters more than new information: turning rough notes into polite paragraphs, shifting a message from blunt to formal, cutting a long complaint to one page, or offering three ways to open a cover letter. It can also check a draft for repeated phrases and unclear sentences.
The same mechanism explains its weak spot. A model produces plausible text, not verified text, so a letter that asks it to "mention the relevant law" may receive a confident citation that does not exist. Anything the sender did not supply, such as an account number, a policy date, or a statute, has to be treated as unconfirmed until it is checked against a real document.
What to keep out of the prompt
The Office of the Privacy Commissioner of Canada and its provincial and territorial counterparts issued principles for generative AI on December 7, 2023. They are aimed at organizations, but several points apply to anyone drafting a letter. Where possible, prompts should use anonymized or de-identified information instead of personal information, and sensitive or confidential details should be entered only where authorized. The principles also say developers should tell users when prompt content is used to train or refine a model, and that public-facing tools should warn users not to enter personal information unless necessary.
In practice, a draft can use placeholders such as [Name], [Account number], and [Date], and the real details can be added in a separate document afterward. The same habit works for U.S. readers, since many consumer tools state in their terms whether conversations may be reviewed or reused, and those terms can change.
What happens when no one checks
The best-documented example comes from a court rather than a personal letter, but the mechanism is identical. In the court's June 22, 2023 sanctions ruling in Mata v. Avianca, a New York federal court found that lawyers had filed a brief resting on cases ChatGPT made up, and that they kept defending it after the judge raised doubts. The penalty was USD 5,000, and the lawyers also had to notify their client and the judges falsely named. The court added that using a reliable AI tool is not improper in itself, provided lawyers check what they file.
A tenant, job seeker, or customer is not under court rules, yet the lesson carries over: the signature on the letter makes the content the signer's own.
A quick guide by letter type
The table is general orientation, not legal advice. Deadlines and notice rules for housing, employment, and consumer disputes differ by province and state, so the official source for the relevant jurisdiction decides.
| Letter type | Good use of AI | Check before sending |
|---|---|---|
| Cover letter | Structure, trimming, matching tone to the posting | Every claim about skills and dates matches the resume |
| Complaint to a company | Clear, calm wording; one-page length | Order numbers, amounts, and dates copied from receipts |
| Notice to a landlord or employer | Formal tone and layout | Required notice period, delivery method, and wording per local law |
| Condolence or thank-you note | Help starting a draft | Personal details and names are the sender's own, and the voice sounds like the sender |
| Letter that cites a law or regulation | Plain-language explanation of a passage already pasted in | Citation read directly on the official legislation site |
A five-step routine
- List the facts in plain notes: who, what, when, amounts, and the outcome requested.
- Replace names, addresses, and numbers with placeholders before pasting anything into a tool.
- Ask for a draft in a stated tone and length, and give the recipient's role.
- Compare every factual statement in the draft with the notes, deleting anything that was not supplied.
- Read the final text aloud, restore the real details, and keep a copy of what was sent.
The last step matters because recipients such as employers, landlords, and agencies often read for personal voice and specific detail, which a generic draft lacks.
FAQ
Is it acceptable to send a letter written with AI?
Generally yes, for ordinary correspondence, and nothing in the sources reviewed here prohibits it. Some settings carry their own rules, such as courts, schools, and employers, so the receiving institution's policy applies.
Should personal details be typed into an AI tool?
The Canadian regulators' principles favor anonymized information in prompts and warn about entering personal information unless necessary. Placeholders reduce the risk without losing much quality.
Can AI cite the law in a demand letter?
It can write something that looks like a citation, but the Mata v. Avianca case shows such citations can be invented. Each one needs to be read on an official legislation or court site.
Does AI make a letter sound the same as everyone else's?
Often. Drafts tend toward generic phrasing, so adding one specific, true detail and rewriting the opening in the sender's own words usually fixes it.
Related reading
- AI tools in academic writing
- Ethical questions around AI-generated work
- Where AI and creativity meet
- AI in creating and refining art
More in AI Basics.
Last verified: October 2026
Sources
- Principles for responsible, trustworthy and privacy-protective generative AI technologies, Office of the Privacy Commissioner of Canada with provincial and territorial counterparts, December 7, 2023
- Opinion and Order on Sanctions, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), U.S. District Court, Southern District of New York, June 22, 2023