Should You Tell Customers a Chatbot Is AI? A Practical Guide

Yes. A customer-facing AI assistant should be clearly identified as AI from the first message. Tell visitors what it can help with, what it cannot confirm and how to reach a person. Transparency is good practice even where no law requires it, and in Australia the privacy regulator already expects public-facing AI tools to be identified.

A scenario#

A visitor opens the chat on a cleaning company's website. The header says "Sarah", with a friendly photo. Sarah answers questions about end-of-lease cleans, then says "I've booked you in for Thursday."

Nobody has booked anything. "Sarah" is an AI assistant that collected the visitor's details for a callback. When the customer rings on Thursday morning asking where the cleaner is, the business has a complaint, a refund conversation and a customer who now distrusts everything on the website.

The problem was not the AI. It was that the interface let the customer believe they were talking to a staff member who could confirm a booking.

Put the disclosure where the conversation starts#

Disclosure belongs in the chat itself, not only in a privacy policy. Use direct language in the first message. For a fictional cleaning business:

"Hi, I'm the AI assistant for Coastal Cleaning. I answer from our published information and can take your details for the team. I can't confirm bookings; a person will do that."

That opening sets three expectations:

1. What it can do: answer from approved business information and collect details.
2. What it cannot confirm: bookings, availability, prices outside published rules, anything requiring judgement.
3. How to reach a person: a button, a phone number or a callback.

Names and avatars#

A persona name is not automatically deceptive, but a human name plus a human photo, with no mention of AI, invites the misunderstanding in the scenario above. Safer options:

• Use the business name: "Coastal Cleaning assistant".
• If you use a persona name, put "AI assistant" next to it in the first message.
• Avoid stock photos of people as the chatbot's avatar; use your logo.

Repeat it when the stakes rise#

Surface the AI identity again when a conversation becomes consequential:

• Quotes: "This is an estimate based on our published price guide; the team confirms the final quote."
• Bookings: "I've shared our booking link. Your booking is confirmed only when you receive the confirmation from that system."
• Complaints: offer a person straight away.
• Personal information: explain why a detail is needed before asking for it.

Never let the assistant say a human has reviewed something when no one has.

What the rules say#

In Australia, the Office of the Australian Information Commissioner says businesses should update privacy policies and notifications with clear information about their use of AI, including ensuring that public-facing AI tools such as chatbots are clearly identified as such to users (OAIC guidance; OAIC top five takeaways). The same guidance says collection of personal information through public-facing AI chatbots must comply with APP 3 and APP 5 where the Privacy Act applies.

In the United States, requirements vary by state and context. A growing number of state laws require disclosure when companies deploy chatbots in certain one-to-one consumer interactions, and deception standards can apply when a bot's artificial identity is material (DLA Piper).

In India, the Digital Personal Data Protection Rules, 2025 require clear consent notices that explain the specific purpose for which personal data is collected (PIB). Telling people they are dealing with an AI assistant fits naturally alongside that notice.

This is general product guidance, not legal advice. Requirements depend on your location, sector and use case.

Disclosure is also good conversion design#

Some owners worry that saying "AI" will put customers off. In practice, disclosure gives the assistant permission to be honest: "I don't have an approved answer for that; I can take your details so the team can confirm." It also makes asking for a person feel like a normal option rather than a failure.

A customer who knows what they are talking to forgives "I'm not sure" far more readily than a confident wrong answer from someone they thought was staff.

How to set this up in Pivra#

• Welcome message: write your AI disclosure into the first message visitors see (appearance settings).
• Chatbot name: use your business name, or add "AI assistant" to a persona name.
• Instructions: in your system prompt, tell the assistant never to say a booking is confirmed or that a person has reviewed something (improving responses).
• Handover (Starter and above): let visitors ask for a person; when a team member claims the conversation, the header shows "A team member is here", so the change from AI to human is visible (human handover).
• Lead capture (Starter and above): explain why each field is needed before asking (lead capture).

For more on collecting details carefully, see privacy-first AI chatbot lead capture, and for handover design see the human handover playbook.

The five-person comprehension test#

Ask five people who have never seen your website to have a short chat with the assistant. Then ask them:

1. Who or what were you talking to?
2. What information could it access?
3. Was anything booked or confirmed?
4. What happens to the details you gave?
5. How would you reach a person?

If any answer does not match reality, change the wording and test again.

Sources#

• OAIC: Guidance on privacy and the use of commercially available AI products
• OAIC: Top five takeaways
• DLA Piper: AI disclosure laws on commercial chatbot interactions
• PIB: Digital Personal Data Protection Rules, 2025

Next step: write an AI disclosure a customer can understand in five seconds, and put it in your welcome message today.