Age, purpose, privacy, and support belong in the same design conversation.
Start with the context of use
A homework assistant, a creative tool, and a companion raise different questions. Young people also have different needs at different ages. A single toggle cannot capture every context.
Privacy and support can be in tension
Caregivers may need information to help. Young people may need privacy to ask sensitive questions. Product decisions should make the intended boundaries understandable and examine unintended consequences.
Make the system’s role clear
An assistant should not create a false impression of being a person, a confidential professional, or a guaranteed source of correct advice. Interfaces can communicate limitations without making ordinary use feel alarming.
Safety includes opportunity and participation
UNICEF’s third-edition guidance connects protection with children’s privacy, development, inclusion and participation. This broader view matters: a learning tool should support beneficial use, while considering who may be excluded or exposed by its safeguards.
Source: UNICEF Innocenti · Guidance on AI and children · version 3Avoid one rule for every age and situation
Our editorial view is that a young child using a classroom tutor and an older teenager seeking private information should not automatically receive identical controls. Consider developmental needs, the adults involved, the task and whether sharing information could itself create risk. Design choices should be explained in language the child can understand.
Evaluate relationships as well as isolated answers
A companion product can shape expectations over repeated conversations. Review how it describes its role, responds when a user wants to leave, and encourages appropriate real-world support. A content filter on individual replies cannot by itself establish that the overall relationship is beneficial. This is a proposed evaluation lens, not a claim that a particular product meets it.
A situation to think through
A homework tool notices repeated late-night use. Automatically sharing full transcripts may reveal unrelated private information. A thoughtful design would explain available support, minimize any information shared and make the rules visible before the situation arises.
Questions to take with you
- Can a child understand the system’s role and privacy choices?
- Does the design support learning without unnecessary collection?
- Have children and relevant caregivers helped evaluate realistic use?
For more reading
The sources behind this page, with a reason to open each one. Practical examples and recommendations are our editorial interpretation.
- Guidance on AI and children · version 3
Connects safety, privacy, inclusion, development and children’s participation in AI design.
- Guidance for large multimodal models in health
Introduces risks and governance needs for generative AI in health, with a link to the full WHO guidance.
- Generative AI Profile · NIST AI 600-1
A framework for identifying, measuring and managing generative AI risks across the system lifecycle.
Sources reviewed 13 September 2026. Product documentation can change. How we use evidence