Designing AI Onboarding Flows That Set Correct User Expectations
We launched an AI drafting assistant for a client's content platform and got two contradictory complaints in the first week: some users blindly published AI drafts without checking them, and others refused to use the feature at all after one imperfect suggestion. Both problems traced back to the same root cause — the onboarding never set an accurate mental model of what the tool was actually good and bad at.
We redesigned the first-use experience around a concrete demonstration rather than an explanation. Instead of a modal saying 'AI suggestions may contain errors,' new users now see the assistant draft something in front of them, then get shown a specific real mistake it made on a sample document — a factual detail it got slightly wrong — with the correction highlighted. Showing one real failure up front did more for calibrated trust than any disclaimer copy we tried.
For the users who were over-trusting, we added friction proportional to risk: low-stakes suggestions like phrasing tweaks apply instantly, but anything touching a factual claim or a number requires an explicit confirmation click with the changed text highlighted. That single change cut the rate of unreviewed factual errors slipping into published content by roughly 70% in the following month's data.
For the users who were under-trusting after one bad experience, we added a lightweight explanation of why a given suggestion was made — which source passage it drew from, when relevant — rather than presenting suggestions as unexplained black-box output. Users didn't need the AI to be perfect; they needed to understand enough about how it reached an answer to decide when to double-check it themselves.
The broader pattern: onboarding for AI features isn't really about teaching the UI, it's about calibrating trust to match actual reliability. That means showing real limitations early, matching friction to the cost of being wrong, and giving users enough visibility into the 'why' that they can build accurate intuition instead of either blind faith or blanket skepticism.
