Feedback Loops: Why a Button Needs to Feel Pressed

Press a key on a mechanical keyboard and three things happen at once. It moves, it makes a noise, and it stops moving at a defined point. None of those are necessary to enter a character, and all of them are why people care what keyboard they use.

A touchscreen has none of that. Glass does not move, does not click, and offers no resistance. So every sense of having pressed something on a phone is manufactured, entirely, out of what appears on screen in the moments after contact. There is no physical event to fall back on.

This is easy to underrate because it sounds like polish. It is not polish. It is the mechanism by which a player learns that they are the cause of what is happening — and that belief is the thing every argument for this format depends on.

Consider what a player actually knows. They touched the screen. Something happened. The connection between those two facts is not given to them — it is inferred, and the inference relies almost entirely on timing.

If the screen changes immediately, the change is attributed to them. If it changes after a pause, the attribution weakens, and at some point the change is read as something the ad did rather than something they caused. The event is identical; only the delay differs.

That is the whole of it, and it explains why feedback cannot be treated as a finishing touch. A playable ad with no immediate response is not a slightly less satisfying playable ad. It is a video the player happens to be touching, because nothing has established that their touching matters.

It also explains a failure that puzzles teams when they see it in testing: a player taps, the ad responds a few hundred milliseconds later, and the player taps again. They are not impatient. They have concluded the first tap did not register, because from their side there is no difference between a slow response and no response.

Different delays produce different experiences, and the boundaries are sharper than they sound. These are the thresholds worth designing against.

The practical consequence is a split that solves most feedback problems: separate the acknowledgement from the outcome.

The acknowledgement has to be instant — a press state, a scale change, a highlight, anything that says contact was received. The outcome can take as long as it needs. A piece that slides into place over half a second is fine, provided something confirmed the grab at the moment of contact. What is not fine is nothing happening until the animation begins, because the player spends that gap not knowing whether they did anything.

Most feedback effort goes into rewarding correct actions, which is understandable and backwards.

A player who does the right thing already has evidence that they did — the game progressed. A player who does the wrong thing has nothing, and silence in that moment is ambiguous in the worst possible way. Did I miss the target? Is this the wrong move? Is the ad broken? Each of those leads somewhere different, and a player who picks the third leaves.

So an invalid action needs a response more urgently than a valid one. Not a punishment and not an error message — just evidence of contact. The thing wobbles and returns, the target flashes and refuses, the piece lifts and settles back. Any of these says two things at once: I heard you, and that is not it.

This is also the cheapest reliability improvement available in the format. A large share of ads that players describe as broken are not broken at all. They simply do not acknowledge the inputs that fail, and a stranger given no response assumes the software rather than the aim.

Calling this a feedback loop rather than feedback is not just terminology. The important property is that each response changes what the player tries next.

A player arrives knowing nothing about your mechanic. They touch something. What happens next is the entire curriculum: it tells them whether that was the right kind of action, whether the thing is grabbable, whether precision matters, whether speed matters. Then they act again with that knowledge, and the next response refines it.

In a format that cannot afford instructions, this loop is the tutorial. Not a hand pointing at a target — the accumulated evidence of what the screen does when touched. Which means the quality of feedback determines how fast the mechanic is learned, and a mechanic learned in two seconds rather than five has gained three seconds of the ad back.

The corollary is that inconsistent feedback teaches inconsistent lessons. If a valid tap sometimes produces a strong response and sometimes a weak one, the player cannot build a model of the rules, and the confusion presents as the ad being fiddly rather than as the feedback being unreliable.

Two limits, because this is one of the easier principles to overapply.

More response is not linearly better. Once every input is clearly acknowledged, additional effects stop adding clarity and start competing with the thing the player is trying to watch. An ad where every tap produces particles, a sound, a screen shake and a number is not more responsive than one with a clean press state — it is louder, and the extra motion makes the actual state of the board harder to read. The test is whether a new effect tells the player something they did not already know.

The specific thresholds here also vary with what is being touched. A response feels immediate at a longer delay when the player is dragging something continuously than when they are tapping, because continuous contact provides its own ongoing evidence. The numbers above are useful as a shape rather than as a specification, and the way to tune them is on a real device rather than from a table.