Do You Have to Tell People You Used AI?
It’s the quiet question hanging over every knowledge worker right now. You used AI to draft the email, tighten the report, or analyze the numbers — do you have to say so? Platforms are starting to answer it for content: Meta now auto-labels AI-generated ads, and the industry is building Systems that tag a piece of media with a record of how it was made — whether AI created or edited it — usually as invisible metadata that travels with the file. It’s how a platform can automatically detect and label an AI-generated image or video. It works for published media, but it says nothing about the everyday AI help behind an ordinary work email. tools to track it automatically. But at your desk, drafting a memo, there’s no label and no rule — just your judgment. Here’s how to think about it clearly.
AI use isn’t one thing — it’s a spectrum
The reason this feels murky is that “I used AI” covers wildly different things. Picture a spectrum:
- At one end: spellcheck. You wrote it; AI fixed a typo, tightened a sentence, suggested a better word. Nobody thinks you need to disclose spellcheck.
- At the other end: a ghostwriter. You gave a one-line prompt and AI produced the whole thing, which you’re now presenting as your own thinking.
Almost every real situation sits somewhere between those two poles, and where it sits is what determines whether disclosure matters. The question was never “did AI touch this?” It’s “how much of what makes this mine did AI actually do?”
The test: two questions
When you’re unsure, ask these:
1. Would the audience reasonably assume this is your own work? Some things carry an implicit promise of personal authorship — a condolence note, a performance review you’re giving, a piece of creative writing, an expert opinion someone is paying you for. Other things don’t — a formatted meeting summary, a routine status update, boilerplate. The more someone assumes you personally made it, the more AI’s involvement is something they’d want to know.
2. Do the stakes make “how it was made” matter? If being wrong is expensive, or if your personal judgment is the actual product, the method matters. A doctor, lawyer, or analyst whose expertise you’re trusting is different from someone reformatting a spreadsheet.
If the honest answer to both is “no” — the audience doesn’t assume sole authorship and the stakes are low — you’re in spellcheck territory. Use AI freely and say nothing; it’s a tool. If either is “yes,” disclosure starts to matter.
Where you clearly don’t need to disclose
Treat these like any other software — no announcement required:
- Fixing grammar, tightening, or rewording your own draft.
- Brainstorming or thinking out loud before you write the real thing.
- Summarizing a document for yourself.
- Formatting, reformatting, and boilerplate.
- The routine internal email that nobody imagines you agonized over.
Disclosing here isn’t more honest — it’s just noise, and it can even undersell work that’s genuinely yours.
Where it genuinely matters
- When personal authorship is the point. Creative work, a heartfelt message, anything whose value is that it came from you. Passing off AI-generated writing as your own voice is where people feel deceived.
- When someone is trusting your expertise. If people are relying on your professional judgment, AI-generated analysis presented as your own careful work crosses a line — especially if you didn’t verify it.
- In academic, legal, or regulated settings, where rules or honor codes may explicitly require it.
- When someone asks. “Did you write this?” has an honest answer. Dodging it is the disclosure failure people remember.
The principle underneath
This isn’t about AI being shameful — it’s about not misleading people regarding what they’re getting. Which means there are two ways to get it wrong, not one. Hiding AI where authorship or trust is on the line erodes credibility the moment it’s discovered. But reflexively stamping “made with AI” on everything is its own mistake — it buries a real signal under noise and quietly discounts work you actually did. Aim for the honest middle.
How to disclose without making it weird
When it does matter, keep it matter-of-fact — no confession, no fanfare:
- “I used AI to pull together a first draft, then edited it.”
- “AI-assisted analysis — I’ve checked the figures.”
- “Drafted with AI, reviewed and approved by me.”
Notice what those all share: they signal that a human is accountable. Which is the part people actually care about.
One thing disclosure does not do
It doesn’t transfer responsibility. Whether or not you mention AI, you own what goes out under your name — including a confident When AI states something false with total confidence, because it’s built to produce a plausible-sounding answer rather than to look facts up. Saying “AI wrote it” doesn’t make its mistakes any less yours once you’ve sent it — disclosure is about honesty, not about offloading the blame for errors. you didn’t catch. Disclosure is honesty about authorship; verification is your responsibility about accuracy. Don’t confuse the two, and don’t let the first substitute for the second.
The takeaway
Stop asking “did I use AI?” and start asking “would my audience feel misled if they knew how this was made?” If the honest answer is no — low stakes, no assumption of sole authorship — it’s a tool, use it and move on. If yes, a plain one-line acknowledgment that a human is accountable covers you. Honesty about what people are getting, without the theater. That’s the whole rule.
Related: Write a One-Page AI Policy Your Team Will Actually Follow for setting disclosure norms across a team, and Giving AI Context: The Skill That Actually Makes It Useful on why the human stays accountable for the output.