Autocomplete best practices

Effective autocomplete triggers from the second character, matches mid-word rather than prefix-only, tolerates typos, mixes query completions with categories and a few real results, highlights the matched text, and responds in under 100ms. Its real job is preventing zero-result searches before they are submitted.

Rules that change outcomes

• Trigger at two characters. Waiting for three loses short, high-intent queries.
• Match anywhere in the term, not only at the start — this is the most common user complaint.
• Apply typo tolerance in suggestions, where correction is cheapest and least intrusive.
• Mix types: query completions, categories, and three to five concrete results.
• Bold the matched substring so the list can be scanned rather than read.
• Cap at eight to ten items. Longer lists are ignored, not scrolled.
• Full keyboard support, and an escape that closes without clearing the field.

Mobile is a different problem

The keyboard occupies roughly half the viewport, so only two or three suggestions are genuinely visible. Rank ruthlessly, keep the list from reflowing as characters arrive, and never let the field lose focus while the visitor is still typing.

Measure it separately

Track suggestion click-through as its own metric. A high rate means visitors are finishing sessions in the dropdown — the fastest successful search there is — and it will not show up in results-page reporting at all.