As AI features appear in more products, users are asking a reasonable question: why is it suggesting this? Design trend reports this year point the same way. People want to see where a suggestion came from, how confident the system is, and how to change it. Products that answer those questions earn trust. Products that hide them see people ignore or switch off their AI features. Here are the patterns we use when designing AI into products.
Explain the suggestion
A short reason next to a suggestion helps people decide whether to trust it. “Based on your last three invoices” or “Because you usually pay this on the 1st” gives enough context to judge the suggestion in a second.
Explanations also make errors easier to fix. When the reason is visible and wrong, people know exactly what to correct. When it is hidden, they only know the suggestion feels off.
Show confidence honestly
Not every suggestion is equally reliable. Where it helps, show how confident the system is, in plain language rather than percentages. “Likely match” and “Please check” communicate more clearly than “73%”.
Be careful not to overstate certainty. A system that sounds sure and is wrong damages trust far more than one that admits doubt.
If changing a suggestion is harder than doing the task, people stop using the feature.
Make it easy to override
Every AI action should be easy to accept, edit or undo, ideally in a single step. If changing a suggestion is harder than doing the task manually, people will stop using the feature.
Undo deserves special attention. Being able to reverse an automated action quickly makes people far more willing to let the system act in the first place.
Beside the work, not instead of it
Conversational assistants are powerful, but they work best alongside structured views, not as a replacement for them. A clear list, table or form remains the reliable way to review information and finish a task.
Use the assistant to help people get there faster, such as filtering, drafting or summarising, and keep the structured view as the source of truth people can check.
Be clear about data
People increasingly want to know what information an AI feature uses and where it goes. Say it plainly, at the moment it matters, not only in a privacy policy. Offer clear choices where possible.
Clarity about data is not only a legal requirement in many places; it is a trust signal that makes people more comfortable using the feature.
An example flow
Imagine an expense app that suggests a category for each receipt. A good design shows the suggested category with a short reason, “Restaurant, based on the merchant name”, lets people change it with one tap, and remembers the correction for next time.
If the system is unsure, it says so and asks for confirmation instead of guessing silently. The person stays in control, and the system gets better with every correction.
Designing for mistakes
AI systems will make mistakes. Design for them from the start: clear ways to report a wrong suggestion, easy undo, and graceful fallbacks when the system cannot help.
How a product handles its errors often determines whether people trust it more or less over time.
Measure the right things
Track not only how often suggestions are accepted, but how often they are edited, undone or ignored. High acceptance with frequent undo is a warning sign. Low acceptance may mean the explanations or timing need work, not the model.
Patterns to apply
- Show a short, plain reason for each suggestion.
- Express confidence honestly and in words.
- Make accept, edit and undo one step each.
- Keep the structured view as the source of truth.
