Does a support chatbot need to say it is not a person?

The principle stated is that people dealing with systems of that kind should be made aware they are interacting with a machine. Whether a specific deployment is caught is a legal question on your facts, but the direction is clear enough to make a single sentence the sensible default.

Who has to make generated content identifiable?

The requirement named sits with the providers of these systems rather than with everybody using them, which puts part of the work upstream. That does not settle what a business should say about its own published material, which is a separate and mostly reputational question.

Is this the same as the heavier obligations?

No, and mixing them up wastes effort in both directions. This is a disclosure duty about what people are dealing with. The heavier requirements attach to a narrow set of uses where the output can seriously affect somebody, and almost nothing an ordinary business deploys reaches them.

Glossary

Transparency obligation

A legal duty to let people know they are dealing with a machine, or that what they are looking at was generated, rather than leaving them to guess.

In plain terms

Telling people what they are dealing with. If they are talking to software rather than a person, they should know. If what they are looking at was produced by a machine, that should be identifiable. It is about being told, not about being protected from a bad answer.

Why it matters

Because it is cheap to satisfy and easy to forget, and because it is the duty most likely to apply to an ordinary business. A support widget on a website and a marketing image are both in scope of the ideas here, while nothing about either involves the heavier obligations organisations tend to worry about first.

How it works

The chatbot case is stated directly: where people are dealing with systems of that kind, they should be made aware they are interacting with a machine. That is a disclosure requirement about the interaction rather than a standard the answers have to meet.

The generated-content case sits with the supplier. Providers of these systems have to ensure that what they produce is identifiable, which puts part of the work upstream of any business using the tool rather than on the business itself.

It is entirely separate from quality. A system can be accurate, careful and useful and still breach this by saying nothing, and a poor one can comply perfectly. Conflating the two is the most common misreading, and it leads organisations to think a good tool is a compliant one.

Satisfying it usually costs a sentence. Labelling a support widget and saying so where content was generated are not projects, which is precisely why they get deferred: nobody schedules the work, because there is no work to schedule.

The awkward cases are the ones designed to feel human. An assistant given a name and a personality is more likely to leave somebody genuinely unsure, and it is the case where the disclosure is doing real work rather than stating the obvious.

It sits below the heavier tier and applies far more widely. Most organisations will never deploy anything in the categories carrying the heaviest duties, and most will deploy something covered by the ideas here, which inverts where attention usually goes.

Two questions people answer as one

The two columns get merged because both feel like doing the responsible thing, and one is much more interesting to work on. Evaluating a system, watching how it behaves and deciding who checks its output is genuine engineering with a satisfying shape to it. Adding a sentence to a widget is not, and it takes ten minutes, and it therefore never reaches anybody's plan. The result is a familiar asymmetry: organisations with real rigour on the left and nothing on the right, quite often believing the first covers the second. The useful reframe is that the right-hand column is not about protecting anybody from a bad answer. It is about not putting somebody in the position of working out, halfway through, that the thing they have been confiding in is software. The awkward case is the deliberately human-seeming assistant, complete with a name and a personality, where the disclosure does real work precisely because the design is pulling the other way.

Seen in the wild

A support widget that says up front the reply is automated.
Tidio Lyro
A generated presenter video where the audience is told what they are watching.
Synthesia
A generated voice used in a recording, identified rather than left ambiguous.
ElevenLabs

Common misconceptions

People assume

A well-built system satisfies it.

In fact

It says nothing about quality. An accurate, well-tested assistant that never tells anybody it is software has not met a disclosure duty, and a poor one that does has. The two questions are unrelated and get confused constantly.

People assume

It only matters for systems doing something important.

In fact

It applies to the interaction rather than the stakes. A support widget answering routine questions is the ordinary case, and the heavier duties people worry about first apply to a much narrower set of uses.

Questions

Does a support chatbot need to say it is not a person?: The principle stated is that people dealing with systems of that kind should be made aware they are interacting with a machine. Whether a specific deployment is caught is a legal question on your facts, but the direction is clear enough to make a single sentence the sensible default.
Who has to make generated content identifiable?: The requirement named sits with the providers of these systems rather than with everybody using them, which puts part of the work upstream. That does not settle what a business should say about its own published material, which is a separate and mostly reputational question.
Is this the same as the heavier obligations?: No, and mixing them up wastes effort in both directions. This is a disclosure duty about what people are dealing with. The heavier requirements attach to a narrow set of uses where the output can seriously affect somebody, and almost nothing an ordinary business deploys reaches them.