AI suggestion systems anticipate what a user wants to do next, surfacing relevant options, completions, or actions before they explicitly ask. They reduce typing, searching, and repetitive decisions during a task.
Use cases
If you ignore this or get it wrong, you introduce massive friction. Forcing users to remember exactly what to type or search leads to abandoned sessions.
Solving search struggles: A user types "blu" and the system suggests "blue shoes," "blouse," and "blueberry seeds." This prevents typos and gets users to results faster.
Contextual actions: While writing an email, the system suggests meeting times based on calendar availability. This removes extra tab switching and manual scheduling.
How it's used in practice
Set triggers: Determine precisely when suggestions appear (e.g., after 3 characters typed or when a user pauses on a field).
Prioritize relevance: Order suggestions based on user history, popularity, or proximity, not just alphabetical order.
Design for refusal: Ensure users can easily ignore or dismiss the suggestions (via 'Esc' key or clicking away) without interrupting their flow.
🪄
Pro-tip: Avoid "Ghosting."
Don't auto-fill a field with the top suggestion unless the user accepts it with Tab, Enter, or a click. Suggestions should reduce effort, not take control away from the user.
Challenges & limitations
Cold start problem: New users get irrelevant suggestions because the model has no signal yet
Filter bubbles: Systems trained too heavily on past behavior can repeat the same patterns and reduce discovery.
