AI disclosure for customer conversations
Make an AI disclosure useful to shoppers. Explain the agent’s role, data use and action controls, and distinguish a notice from customer consent.
- Published
- Last updated
An AI disclosure tells customers they’re interacting with AI and explains what it can do before they rely on its answers or authorise an action. Put a short notice at the start of the conversation. Add information when the customer uploads personal information or gives the agent permission to buy.
The useful question is what the customer needs to understand at that moment. A label saying “AI assistant” identifies the technology. It doesn’t explain whether a photo leaves the store’s systems or whether a request will place an order.
What an AI disclosure tells the customer
An AI disclosure identifies the agent’s role and explains the boundaries that affect the customer’s choice. For a shopping conversation, that means explaining where answers come from and when the agent can act. Make the route to a person easy to find.
A disclosure has a different purpose from a liability disclaimer. “AI answers may be wrong” tells the customer about a risk. It leaves unanswered whether the agent is answering from your catalogue, searching other websites or speaking as a staff member.
Avoid describing an agent as a person. A human name or avatar can sit beside a visible AI label, but the conversation must not create a false impression of human review.
A disclosure also gives no assurance that an answer is accurate. If an agent incorrectly refuses a return, its AI label doesn’t resolve that error. The Australian Competition and Consumer Commission (ACCC) says businesses must consider the overall impression of their claims, including qualifications in fine print.
Choose information for the shopping moment
Give customers the information they need before the relevant decision. An initial greeting, an upload explanation and a purchase permission each do a different job.
| Shopping moment | What the customer needs to understand | Where to put it |
|---|---|---|
| Asking about a product | That the agent is AI, the source of its answers and how to reach staff | Opening message and a persistent help control |
| Uploading a photo | The purpose of processing, who receives it and the applicable handling details | Beside the upload control, before submission |
| Placing an order | What will be bought, the total charge and whether this action authorises payment | With the order summary and purchase control |
| Enabling automatic buying | The permitted products, spending limits, timing and how to stop future orders | In the permission flow and its saved settings |
For example, shoppa helps customers find, compare and buy products in the chat, using the retailer’s catalogue and documents. It also supports customers after the sale and hands conversations to staff when they need a person. Those jobs need different explanations at different points in the conversation.
A product answer can identify the catalogue as its source. A virtual try-on needs to explain that the customer submits a still photo and receives an appearance preview. The preview doesn’t establish physical fit or recommend a size.
With automatic buying, the customer gives permission through rules set in advance. Explain that qualifying purchases can happen without a separate confirmation for each order. Put those rules where the customer can review them, alongside the way to withdraw that permission.
Treat an unexpected action as a design problem. A greeting cannot replace a purchase control that makes its effect apparent.
Write a clear notice
Write a brief opening notice in the same language your customers use. Identify the AI role and describe the information it uses. Give the customer a working way to reach staff.
The following notices are fictional examples for a store called Banksia Home. They illustrate different moments, rather than state another retailer’s data practices.
Opening a conversation
I’m Banksia Home’s AI shopping assistant. I answer from our product catalogue and store policies. Choose Talk to our team if you’d like help from a person.
Keep the AI label visible after the greeting scrolls away. A returning customer must still be able to recognise the agent’s role.
Before a photo upload
For a fictional feature that sends a photo to an image-processing supplier, the explanation could read:
Upload a room photo to create an AI preview with Banksia Home products. Our image-processing supplier receives the photo to make the preview. The image shows appearance only. Read Photo privacy before uploading, or choose Talk to our team for help without a photo.
The Photo privacy control must open the actual handling details. Explain storage and deletion there, along with any further use and overseas disclosure. Publish only the promises that match your service and supplier terms.
Before enabling automatic buying
For a fictional rule that permits a particular refill purchase, the permission summary could read:
Allow the assistant to order one 500 ml soap refill every six weeks, up to $25 including delivery. Orders within this rule will be paid automatically. Turn the rule off in Buying settings to stop future automatic orders.
The control that enables the rule must clearly authorise that action. A customer who only opens Buying settings hasn’t authorised the purchase.
Keep privacy information and buying rules available from the conversation. Repeat the relevant explanation when the purpose changes or the agent requests a new permission. Repeating the entire opening notice with every product answer makes the information harder to use.
Disclosure is not consent
Disclosure supplies information. Consent is the customer’s agreement to a particular handling of their information. Permission to spend money is a separate decision again.
The Office of the Australian Information Commissioner (OAIC) says organisations cannot infer consent simply from providing notice of a proposed collection. Its AI guidance also explains that collecting or generating sensitive information usually requires consent under Australian Privacy Principle 3.
For entities covered by the Privacy Act, Australian Privacy Principle 6 restricts secondary uses and disclosures. Consent is one permitted basis. Some related uses that a person reasonably expects can qualify under an exception. Sensitive information has a stricter directly related purpose test. The Privacy Act’s Australian Privacy Principles set out those conditions and other exceptions.
Consider a fictional store that receives a room photo to create a furniture preview. Reusing that photo for an unrelated advertising campaign is a different purpose. Adding “We use AI” to the greeting doesn’t establish permission for that reuse.
Where you rely on consent, explain the particular use and give the customer a meaningful choice. Separate optional uses from the action needed to answer the shopping question. Check whether any proposed inferences create sensitive information before asking customers to upload photos.
Check the rules that apply
Start with the Australian obligations that apply to your business and its handling of information. Keep legal requirements distinct from recommended notice design.
For entities covered by the Privacy Act, Australian Privacy Principle 5 requires reasonable steps to notify people of relevant collection matters. This applies at or before collection, or as soon as practicable afterwards if earlier notice isn’t practicable. The OAIC’s collection notice guidance explains the matters to address. They include the purpose, usual disclosures and relevant overseas recipients, as well as access and complaint information.
The Australian Consumer Law prohibits misleading or deceptive conduct in trade or commerce. Its section 18 prohibition applies to the business’s conduct, including the impression its customer communications create. Calling an answer AI-generated doesn’t make a misleading claim acceptable.
Government recommendations have a different status. The National AI Centre’s Guidance for AI adoption recommends telling people when and how they interact with AI. That guidance is separate from the statutory duties above. Likewise, a rule written for government agencies or another country needs its own applicability assessment before you treat it as a retail requirement.
Before launching a notice, walk through a product question and a photo upload. Then review the purchase flow, including any permission for future orders. At each point, confirm that the customer can identify the AI role and understand the effect of continuing. Make the notice match what the system actually does.
Give your store its own shopping agent
Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.