Why disclosure comes up
Sooner or later, someone asks: “Do we have to tell people we used AI for this?” The honest answer is that it depends. Some situations carry legal duties, some are governed by contracts or platform rules, and many are simply matters of judgment about what people would reasonably want to know.
This article offers a way to think it through. It is general information, and rules differ by country, sector and context, so check the requirements that apply to you.
The trust question
A useful test is expectation. Would the person be surprised, or feel misled, to learn that AI was involved? If they would, that is a strong sign you should say so up front. People generally accept tools that make a service faster or more consistent. What they resent is discovering that they were talking to a machine while believing it was a person, or that a significant decision about them was shaped by a system nobody mentioned.
The cost of disclosure is usually low. The cost of being found out after concealing it can be a loss of trust that is hard to repair.
Situations where disclosure matters most
Interacting with a chatbot or voice system. People should generally be able to tell they are not dealing with a human. Several places have rules or proposals along these lines, and ethical practice points the same way, so make this clear at the start of the conversation rather than burying it in terms.
Decisions that affect people. If automated tools help screen job applicants, set prices, assess eligibility or prioritize patients, the people affected may be entitled to know, and in some places to request human review or an explanation. These are the kinds of sensitive uses where extra care is warranted. The concerns around hiring fairness and recruitment screening show why applicants deserve a clear explanation.
Synthetic or altered media. Generated voices, faces or realistic images should usually be labeled, especially if they could be mistaken for real people or events. The wider risks of fabricated content are one reason this matters.
Published content presented as expertise. If an article, report or advice leans heavily on generated text, readers may reasonably want to know, particularly when accuracy matters. Some publishers and platforms have their own policies.
Professional work for clients. Contracts and professional standards may address tool use, confidentiality and review. Where a client expects personal expertise, silent use of AI can be a problem even if the result is good.
Situations where it matters less
Not every use needs a label. Spell check, grammar suggestions, translation of a simple message that a person reviews, or an AI-assisted brainstorm that a human then reworked are usually viewed differently from fully generated content sent without review. Many organizations treat routine drafting aids like any other software tool and do not mention them in every email. The line is not always sharp, so write down where your organization draws it.
Colleagues and internal use
Transparency applies inside organizations too. If a colleague receives a report, summary or analysis, they may want to know whether it was generated, because that changes how carefully they should check it. A simple norm, such as noting when AI produced a first draft, helps everyone calibrate their trust. It also discourages people from passing along unchecked output as if it were their own research.
Staff should also know what is being done with AI in the organization. Using AI to monitor or evaluate employees, for example, raises serious fairness and legal questions, and the lines around employee monitoring deserve open discussion with the people affected.
How to disclose well
Good disclosure is brief, plain and placed where people will see it.
- Say what and why. “This reply was drafted with an automated assistant and reviewed by a team member” tells readers what they need to know.
- Place it early. Put it at the start of a chat, near the top of a page or alongside the relevant output, not in a footer nobody reads.
- Avoid jargon and vague phrases. “Powered by advanced technology” is not disclosure.
- Offer a way out. Where appropriate, give people a route to reach a human or ask for a review.
- Be accurate about human involvement. Do not claim “reviewed by an expert” unless that really happened.
- Keep it consistent. Use the same wording across channels so it becomes familiar.
Mistakes to avoid
- Making a bot sound human, with a fake name and photo, in a way that deceives.
- Overstating what the tool does, or implying it is more accurate than it is.
- Disclosing only in fine print.
- Treating disclosure as a substitute for quality checks. Saying “AI-generated” does not excuse errors.
- Assuming one rule fits everywhere. Different markets and regulators may require different things.
Putting it into policy
A workplace AI use policy is a natural home for these decisions. It can list the situations where disclosure is required, offer approved wording and say who to ask in unclear cases. Revisit it periodically, since expectations and regulations are changing and customer attitudes shift.
In general, a default of openness is the safer path. It respects the people you serve, makes errors easier to correct when someone points them out, and puts you on firmer ground if rules tighten later.