Dotted · behaviour good, perception poor
Phantom friction
Behaviour is fine. Users report it as bad. Tasks complete at reasonable cost and people still call the product confusing, slow or untrustworthy. The most commonly misdiagnosed state in the discipline.
What is happening
Five mechanisms produce it, and none of them lives in the interaction itself.
Missing feedback. The task succeeds but the product never says so. The user completes the job and leaves unsure whether it worked, and uncertainty is remembered as difficulty.
Unclear progress. A three-step flow that never shows where you are feels longer than a five-step flow that does. Perceived effort tracks visibility, not step count.
An expectation set upstream. Marketing promised instant. The product delivers in ninety seconds. Ninety seconds is fast, and every one of them is experienced as a broken promise.
Absent trust signals. The payment goes through reliably, and the page asking for the card number looks like nobody owns it. Users complete it and describe it as sketchy.
Naming that misses the mental model. The feature works. It is called something the user would never call it, so finding it once felt like an accident they expect to repeat.
How to detect it
- Solid completion rates and times alongside poor satisfaction or effort scores for the same task, gathered independently
- Complaints that describe feelings — “confusing”, “slow”, “sketchy” — without pointing at a step that fails
- Support tickets asking “did it work?” after actions that succeeded
- Users who complete a flow and then avoid it, delegating it or batching it as if it were expensive
Treatment
Change what the product says, not what it does. Add confirmation, name things in the user’s vocabulary, show progress, surface evidence of security.
This is copy, sequencing and feedback design. It is a fraction of the cost of an interaction rebuild, and unlike a rebuild it addresses the thing that is actually broken: the story the user is left with.
The standard wrong answer
The reflex here is to rebuild the interaction, because the complaints sound like usability complaints and the team knows how to fix usability. It is expensive and it does not help. The behaviour was already fine, so the rebuild has no behavioural headroom to improve into — but it can regress it, and it resets users’ learned habits, which they will report as the product getting worse.
Worse, a rebuild consumes the budget the real fix needed, and when perception fails to move, the conclusion drawn is usually “users don’t know what they want” — which poisons the next round of research too.
What to expect afterwards
Unlike silent friction, fixing phantom friction is loud. Perception scores move within one or two cycles, and users often praise “the redesign” when nothing about the interaction changed. Take the win and log honestly what it was: the flow was always fine, the story was not.