Transparency and disclosure: telling people it's AI

As AI shows up in more of what people read, see, and interact with, a simple question keeps recurring: when should you tell people that AI was involved? The instinct to say "always" is appealing but too blunt, and the instinct to say "only when forced" is corrosive to trust. This piece is a plain-language guide to disclosure norms — why they matter, what makes them tricky, and how to reason toward what is genuinely honest rather than merely defensible.

Why disclosure matters at all

Disclosure matters because people make decisions based on what they believe about who or what they are dealing with. We weigh a doctor's advice differently from a stranger's, a paid review differently from an unpaid one, a human's letter differently from a form. When AI is involved and that involvement is hidden, people may act on assumptions that are no longer true — trusting it as human judgment, or relying on it in ways they would not if they knew.

The harm from non-disclosure is rarely the AI itself. It is the gap between what people believe is happening and what is actually happening. Disclosure exists to close that gap so people can make informed choices. That framing — informed choice, not box-ticking — is the foundation everything else builds on.

The principle underneath the rules

Specific disclosure rules vary by place and over time, so chasing the exact rule is the wrong approach. The durable principle is this: disclose when a reasonable person would want to know, and would act differently if they knew.

That single test handles most cases. It explains why you should disclose that a "person" someone is confiding in is actually a system, and why you need not announce that spell-check touched an email. The first changes how a reasonable person would behave; the second does not. When you are unsure whether disclosure is required, asking whether the information would matter to a reasonable person is more reliable than hunting for a rule that happens to cover your situation.

Where disclosure clearly matters

Some situations call for disclosure under almost any reasonable standard:

• When people might think they are dealing with a human and it matters. If someone would reasonably assume a person is on the other end — in conversation, advice, or care — and that assumption would change their behavior, they should be told.
• When AI shapes consequential decisions. If a system influences outcomes that affect someone's money, health, rights, or opportunities, the people affected have a strong claim to know.
• When content could be mistaken for authentic human work and that mistake matters. Synthetic media presented as real, or AI output passed off as firsthand human experience, misleads in ways people care about.
• When trust is the whole point. In relationships built on authenticity — personal, advisory, journalistic — hidden AI involvement betrays the specific thing the relationship is for.

The common thread is consequence: disclosure matters most where the information would change how someone reasonably acts.

Where the line is genuinely blurry

Honesty requires admitting that not every case is clear. Where does ordinary tool use end and disclosure-worthy AI involvement begin? Few people announce that they used spell-check, autocomplete, or a search engine, and demanding disclosure of every such assist would be absurd and useless. Yet somewhere along the spectrum from "tool that helped me write" to "system that wrote this for me," disclosure starts to matter.

There is no crisp universal line, and pretending there is one does more harm than good. The reasonable-person test is again the best guide: the more the AI did the substantive work, and the more its involvement would change how a reasonable person receives the result, the stronger the case for saying so. Trivial assistance that changes nothing about how the work should be understood sits comfortably below the threshold.

How to disclose well

Disclosure is not only about whether but about how. Done badly, it is technically present and practically useless.

1. Be clear, not buried. A disclosure hidden in fine print that nobody reads satisfies the letter and defeats the purpose. Put it where the people it is for will actually see it.
2. Be specific enough to be useful. "AI may have been used" tells people little. Saying what role AI played helps them calibrate.
3. Match prominence to stakes. The higher the consequences, the more visible and unambiguous the disclosure should be.
4. Avoid disclosure theater. A vague label slapped on everything trains people to ignore it. Reserve emphasis for where it carries real information.
5. Default to honesty when unsure. If you find yourself constructing arguments for why you do not have to disclose, that hesitation is itself a signal that you probably should.

Good disclosure serves the reader's understanding, not the discloser's liability.

Why getting this right is in your interest

It is tempting to see disclosure as pure cost — an admission that makes your work seem less impressive or more replaceable. The longer view is the opposite. Trust, once lost to a discovered deception, is extremely hard to rebuild, and audiences increasingly assume AI might be involved anyway. Being straightforward about it positions you as honest in an environment where honesty is scarce and valued. Norms here are also still forming, which means the people who disclose thoughtfully now help set the standard rather than scramble to meet someone else's later.

The takeaway

Disclosure about AI is not a compliance chore; it is about preserving the informed choices people make based on what they believe is happening. The durable test is simple: disclose when a reasonable person would want to know and would act differently if they did. That principle outlasts any specific rule, handles the clear cases, and gives honest footing in the genuinely blurry ones between ordinary tool use and substantive AI authorship. Disclose clearly, specifically, and in proportion to stakes — and when you catch yourself arguing your way out of it, treat that as your answer. In a world that increasingly assumes AI is involved, plain honesty is both the ethical choice and the durable one.