AI onboarding

AI onboarding is the first-run and early-use experience of an AI feature. Beyond the usual "here's where things are", it has to set expectations about a system that is probabilistic, can be wrong, and may change as it learns. Good AI onboarding is short, honest about limits, and spread over time rather than front-loaded into a tour.

Why it matters

The first few interactions form the user's mental model of the product. Google's PAIR guidebook warns that "Mismatched mental models can lead to unmet expectations, frustration, misuse, and product abandonment," and that marketing which promises "AI magic" sets people up for disappointment. Microsoft's HAX guidelines put two items at the start of the experience: "Make clear what the system can do" (G1) and "Make clear how well the system can do what it can do" (G2).

How to apply it

• Do state capabilities and limits up front. PAIR: "Be up-front about what your product can and can't do the first time the user interacts with it, ideally in your marketing messages."
• Do describe benefits, not technology. PAIR's stage-one advice is "Describe user benefits, not technology."
• Do give people a starting point. Apple's generative AI guidelines suggest a brief tutorial when you introduce a feature, and for open-ended inputs, "curated suggestions that make it easy to get started."
• Do teach in context. PAIR recommends "inboarding" messages and notes that "People learn better when short, explicit information appears right when they need it." A meeting summariser can explain speaker labels the first time it shows them, not in a welcome carousel.
• Do encourage low-risk experimentation, and reassure users that trying things won't lock in their future experience. PAIR's guidance: "Don't assume users want the AI to start learning from the first use."
• Do say how the product will change and how users can help. PAIR offers a template: this is the feature, it helps by X, right now it can't do Y, over time it will become more relevant, and you can help by Z.
• Don't hide the limitations in a footer. NN/g recommends including disclaimers in onboarding and placing them prominently, paired with an action like "Double-check AI outputs."

Common mistakes

• The feature list tour. PAIR advises against introducing AI features "as part of a long introductory list of product features."
• Overselling. Demo prompts that only show best-case results create expectations the product can't meet on real tasks.
• One-time onboarding. When capabilities change, users need to hear about it. HAX G18 is "Notify users about changes."
• Empty input box, no guidance. An open prompt field with no examples leaves users guessing what the system can do (see AI empty state).