Autocomplete and Suggestions: Steering Without Trapping

Autocomplete works when it predicts what this shopper is likely to mean and offers a short, truthful route to it: a product, a category or a phrase other shoppers use. It fails when it pushes promoted items that do not match, shows suggestions so slowly that typing finishes first, or traps keyboard and touch users in a list they cannot leave easily. Keep suggestions fast, few and relevant. Label anything that is a category or a search phrase rather than a product. And when there is no good suggestion, show none and let the full search handle the query.

A suggestion list mixes different kinds of promises. A product suggestion promises that exact item. A category suggestion promises a shelf to browse. A phrase suggestion promises a new search. Visitors read them faster when each type is labelled or grouped consistently, with a product image only where a product is meant. Blending them without signals creates small misunderstandings that add up: the visitor taps expecting a product page and lands on a broad category, feeling steered. Steering is acceptable only when it is transparent. The visitor should always be able to run their original words as a search.

Early letters are ambiguous. A short typed string could begin a brand, a product type or a use phrase. Ranking should use what your store knows: which completions real shoppers choose, which lead to results, and which products are available. Do not let a promotion override relevance at this moment. The search box is a place of high trust, because the visitor is telling you their intent directly. Using that moment to push an unrelated campaign item spends trust for little gain. Promoted products belong in clearly marked places elsewhere, as covered in recommendations that actually relate.

Suggestions that arrive after the visitor has finished typing are decoration. They need to appear while the thought is forming, update without flicker, and not shift the page layout under a tapping finger. Test on a mid-range phone over a normal mobile connection, not only on fast office wifi. If your platform or app cannot deliver suggestions reliably at that speed, a clean search box with a good results page is better than a slow, jumpy suggestion list. Performance standards for responsive interaction are covered in the speed cluster; here the practical test is simply whether the list helps or interrupts.

Autocomplete must work for everyone who uses the search box. Keyboard users need visible focus, arrow-key movement that behaves predictably and an escape route that closes suggestions without clearing their words. Touch users need targets large enough to tap accurately. Screen reader users need suggestions announced in a way that does not overwhelm. The accessibility standard behind these details is listed in the sources. The operator test is simpler: complete a search using only a keyboard, then only touch, then try to abandon the suggestions and type something else. Any route that fights back needs fixing before visual polish.

The best phrase suggestions come from your own logs: the completions shoppers actually chose and that led to products. Review them for language. If visitors say pram and your suggestions insist on pushchair terminology they never use, the list will feel foreign even when it is technically correct. Keep both terms working in search, and let the suggestion wording follow evidence of use. Remove suggestions that lead to zero results, seasonal phrases long after the season, and products that are no longer available. A stale suggestion list is a small broken promise repeated on every visit.

Judge autocomplete by what happens after it appears: how often a suggestion is chosen, whether the chosen suggestion leads to a product view, and whether visitors who see suggestions search again immediately, which can signal a miss. Compare a short period before and after a change, keeping the query mix in mind. Do not judge by clicks alone, because a misleading suggestion can earn a click and lose the visit. The goal is fewer keystrokes to a relevant result. Keep a small hand test from the pillar guide alongside the numbers, so a tidy metric never hides a clumsy experience.