Autocomplete candidate position and the cost of choosing
Aliases: suggestion bar · suggestion chip · inline completion
What it is
Autocomplete offers an acceptable continuation before the user has finished typing; accepting it writes the rest in one shot. Where the candidate appears sets the cost of choosing: a strip above the keyboard, ghost text at the caret, a dropdown—each demands a different aim. What autocomplete saves is taps on remaining letters; what it spends is one aim from the current key to the candidate. Position makes that ledger positive or negative.
Why it happens
Finishing the word by typing costs roughly remaining letters × time per keystroke. Hitting a candidate costs a move to its hit region plus a confirm. A strip on the keyboard’s top edge is a short thumb move, but candidates compete with space and enter for the heat map. Ghost text trails the typed prefix; confirm is often Tab or right-arrow—cheap on a hardware keyboard, a hunt for a key on touch. A dropdown leaves the key plane, so aiming resembles an ordinary tap and may cover the field. More candidates add scan time to selection cost; if the first is usually the target, the rest mainly increase glancing and miss area. Whether completion is worth a tap is “keystrokes saved” minus the round trip to the candidate and back to typing.
Studying it
Vary candidate-strip position, number of items, and confirm method (tap / Tab / swipe-to). Log take rate, mis-taps, and immediate deletions after taking. Independent variables also include prefix length and whether the target already sits first. Compute net gain as keystrokes saved minus selection-action time, not take count alone. A completion-off control measures the time to finish by typing.
Where it stops holding
Search-box and address-bar completion come from history and an index; keyboard-bar completion comes from a lexicon. The two have different placement habits; dropdown results from search do not transfer to a keyboard strip. Tab confirm is nearly zero movement on a hardware keyboard; the same ghost text on touch may never be used. When candidates cover Submit or the next field, occlusion cost stacks on selection cost.
Applying it
- Place the most likely accept on the shortest thumb path from space; do not fill the whole row with extra items that fight space.
- On touch, give confirm an explicit tap target; do not offer ghost text with no hit region.
- Measure net time: is a word taken from completion faster than typing the rest? If not, change placement before changing the ranker.