On-screen keyboards lack physical nubs, so eyes-free accuracy falls well below hardware boards

What it is

Phone and tablet soft keyboards can draw visual marks on F and J but cannot offer a touchable nub, and there is no trough between keys to catch a fingertip. Motor programs trained on hardware drift on glass; letter error rates typically sit well above a physical board with the same layout. The comparison is spatial hit rate, not “did a click sound play.” Audio can confirm actuation; it cannot confirm the column.

Why it happens

Physical keys supply three spatial cues: nubs, dished caps, and gaps. Glass erases all three. Fingers estimate coordinates from the bezel and from occasional vibration. Vibration is a broadband whole-device or region signal and cannot separate adjacent columns. Skilled typists then produce gestures that look like touch typing while taking a peripheral glance every few words. Language models and gesture typing can pull landing points back onto dictionary words after the fact—that is recognition patching, not restored homing. Passwords, code, and foreign names without dictionary cover show the accuracy drop in full.

Studying it

Have the same touch typists transcribe on a hardware board and on the device’s soft keyboard: ordinary prose, random strings with no dictionary, and natural sentences with correction off. Independent variables include permission to look, haptics on or off, and screen size. Dependent measures are uncorrected error rate, glances per minute, and the gap between random strings and prose. “Almost as fast” on prose is often the language model wiping errors; random strings measure spatial accuracy. Do not treat hunt-and-peck touch users as the ceiling.

Where it stops holding

Landscape tablets whose keys approach hardware width shrink the gap, but nubs are still absent. A case with raised caps or a fold-out Bluetooth board turns the comparison back into hardware. Gesture typing changes the movement itself and must not be pooled with discrete-key touch typing. On-screen keyboards driven by a gamepad stick fail differently. For thumb hunters and peckers, eyes-free accuracy was never their baseline.

Applying it

• Do not promise that a touch product types “without looking, like a computer.” Set the expectation that users will watch the keyboard or lean on candidates.
• For passwords, one-time codes, and code editing, leave dictionary smoothing off so spatial error stays visible, and offer paste or scan as alternatives.
• If spatial accuracy on glass must improve, prefer discriminable zonal haptics or a physical accessory over merely drawing larger F/J glyphs.
• Verify with autocorrect off: have touch typists enter nonsense strings on hardware and on the soft keyboard and compare uncorrected errors. If the soft side is not clearly worse, the test is measuring the language model, not the keyboard.

Related

• Same group: C6.19.1 Homing bumps let fingers return to the home row without vision · C6.19.2 Touch-typing skill is built on stable key pitch and travel · C6.19.4 Without tactile homing references, users must look to confirm, cutting both speed and attention to the text
• Adjacent: C6.02 Key travel and tactile confirmation · C6.06 Soft keyboards and screen occupancy
• Search: soft keyboard · eyes-free typing · touch typing accuracy