Previous Searches vs Empty Search Box A/B Test

This comparison is worth planning when search logs show returning visitors typing the same queries over and over, or session recordings show people opening the box and pausing over what to type. If history suggestions help, those visitors select an earlier query in one tap, reach results faster and go on to view products. If you have not yet settled who owns the history, how long it is kept and how it is cleared, define those rules first, and account for what a shared device could expose.

Purchase conversion rate (CR)

Compare purchases per eligible assigned visitor with search history available, including those who do not submit a new query.

Also track

• Search start rateCompare search starts between the history-suggestion and empty-field arms.
• Time to search resultMeasure time among visitors who start a search; report non-searchers separately.
• Zero-result search rateMeasure the share of submitted queries that return no results.
• Product page visit after searchReport product-page visits among search users as a diagnostic measure.

• History suggestionsShow previous queries when search opens or leave the initial suggestion area empty.
• InputKeep the input field blank until the visitor types or explicitly selects a query.
• Other suggestionsKeep suggestions for typed queries unchanged.
• History controlsProvide the same separate access and delete controls in both arms.
• PrivacyDo not show history from private sessions or entries the user deleted; apply the same storage and deletion rules in both arms.

1. 1A higher search start rate is not decisive on its own, since old queries can send visitors to products they will not buy. Purchase conversion among all eligible visitors, including those who never submit a new query, picks the winner; zero-result search rate and product page visits after search help explain where any gap comes from.
2. 2What you learn covers returning visitors with search history and this search box only. Keep the placeholder copy, category and filter names, and the suggestion source the same for the whole test; if any of them changes midway, you can no longer separate its effect from the history suggestions.
3. 3Check that suggestions for typed queries still work in the empty-field arm; if visitors get no guidance at all, the gap comes from the setup. Visitors with history may be a narrow group, so do not stop before the planned end just because the gap looks unclear. Once it ends, the A/B test significance calculator can tell you from each arm's visitors and purchases whether the conversion gap could be chance, and separately check that the arms split in the planned ratio.