Deciding When and How to Tell Customers You Use AI
A customer asks, "Was that reply written by a person?" and nobody on the team knows what the business is supposed to say. That's the problem this page fixes: deciding in advance when to tell customers about AI, and how.
First, the legal question
Whether you must tell customers depends on the country, the industry and the kind of use. This site can't tell you what applies to you. Consumer protection and privacy regulators publish the official position. Examples are the ACCC and the OAIC in Australia, the FTC in the US and the ICO in the UK. A lawyer can tell you how those rules apply to your business.
What follows is what official guidance suggests as good practice. It isn't a statement of what the law demands.
When customers should be told
The Australian industry department's guidance gives the clearest example. Its example is a chatbot. The customer "should know this is not a human specialist"1.
Its reasoning is that clearly communicating "when and how AI is being developed or deployed by the organisation" builds accountability and trust.1
Sort your uses into three groups.
- The customer talks to AI directly. A chat window on your website, an automated phone assistant. Tell them it's AI, at the start.
- AI shapes what the customer receives. A quote, a recommendation or a decision that AI helped produce. Consider saying so, particularly if the result affects them.
- AI helps staff behind the scenes. Drafting, sorting, tidying wording, with a person checking the output. You can decide whether to mention it in a general statement.
Group two is the one to think about hardest. A policy that says "we tell customers when AI materially shaped the outcome" is easier to follow than one that tries to list every case.
An example from government guidance
Staff guidance for Australian government workers describes a case where a staff member summarises a report using AI. Depending on "team or agency expectations regarding disclosure of AI-influenced content", the person "may choose to include a statement advising readers how AI was used to develop the summary".2
Two things to take from it. The decision depends on what the organisation expects. And the statement is short and says how AI was used, not only that it was.
Writing it into the policy
The industry department's guidance asks organisations to create "policy and process for how to transparently communicate" about AI.1 For a small team that means three decisions, written in a few sentences.
- Which uses get disclosed. Use the three groups above.
- What the wording is. A fixed line, so staff don't improvise. See the examples below.
- Who can change it. Usually the policy owner.
Wording examples
For a chat assistant on a website:
You're chatting with an automated assistant. A member of our team can take over if you'd like.
For a document or reply where AI helped and a person checked:
We used an AI tool to help draft this. A member of our team has read and checked it.
For the privacy side, tell customers what happens to anything they type. The cyber security agency advises limiting what a chatbot collects: "Collect only what is essential for the chatbot".3 It also says business contracts with chatbot vendors "should include clear clauses on data protection".3
Don't overclaim
Don't write "our AI is always accurate" or "no one reads your messages" unless you've confirmed it with the vendor's terms. Say only what's true. A person should read output before it goes out, as set out in the check-before-you-use rule.
More on where this fits in the whole document is in the policy outline, and ready sentences are on the sample sentences.
Notes and sources
- Guidance for AI adoption, Department of Industry, Science and Resources (Australian Government)
- Staff guidance on public generative AI, Digital Transformation Agency (Australian Government)
- Artificial intelligence for small business, Australian Cyber Security Centre (ASD)