Telling users what an AI assistant actually does

Users need different context when a tool drafts a response, ranks records or executes an action. The interface should not imply that all three carry the same level of authority or review.

Describe the function in plain language

Explain what the assistant can do, what information it uses at a meaningful level and where a person remains responsible. Avoid claims of human-like understanding or guaranteed correctness.

NIST’s AI RMF Playbook includes suggested actions for managing trustworthy AI. Its voluntary nature means a disclosure should describe the actual service rather than claim compliance with a label alone.

An illustrative support application might state that it drafts an answer from an approved help library and offers a human contact when the answer is insufficient.

Put consequential information near the decision

If an action changes an account, show the relevant details before confirmation. A long policy page elsewhere does not make an ambiguous action understandable.

Provide a usable route to correct information, request clarification or reach authorised support. Do not promise human review unless the operating team can deliver it.

Use the relevant language and accessible presentation for the audience.

Test comprehension

Ask users what they believe happened and who they think is responsible. If they assume a generated suggestion was independently verified, improve the explanation or workflow.

Include the disclosure in usability testing rather than asking only whether legal or technical reviewers approve its wording. Users may interpret a technically accurate statement differently from its authors. Their understanding should inform the design, especially where they must decide whether to rely on a suggestion or request human assistance.

GAT’s AI-first buyer-journey coverage examines trust in discovery. Disclosure supports that trust when it helps people make an informed choice about the actual automated service in front of them.