What 50 Playable Ads Have in Common (and What Only the Good Ones Do)

If you sit and watch fifty playable ads in a row — which we have done, mostly on a phone, mostly wincing — the first thing that strikes you is how similar they are. Not in subject or art style, but in shape. The same parts arrive in the same order, often with the same timing.

The obvious reading is that everyone is copying everyone. There is some of that. But the better explanation is that the format is unusually constrained: a few seconds of attention, a few megabytes of file, no instructions, one hand, a stranger who did not ask for this. Constraints that tight do not leave many workable answers, and a field converging on the same structure is usually a field that has found one.

What is more useful is the second thing you notice. Within that shared skeleton, a small number of decisions separate ads that hold together from ads that do not — and they are not the decisions most production effort goes into.

Stripped of art and theme, the common structure runs roughly like this. Individual ads compress or reorder parts of it, but the sequence is remarkably stable.

None of that is a criticism. It is a sensible structure and we would build against it too. The question worth asking of any individual ad is not whether it uses this shape, but what it does inside it.

Sorting the fifty into ones that worked and ones that did not needs a criterion, and ours was deliberately simple: would a person who does not already play this kind of game come away with an accurate idea of what the product is, and any inclination to try it.

By that standard the separating decisions were surprisingly few.

Almost nothing in that table is about production budget. Several of the ads we thought worked best were visually unremarkable — flat colours, simple shapes, no particular polish. Several of the ones that failed were clearly expensive.

The separating factor was closer to editing than to art direction: knowing what to leave out, and knowing when to stop.

One fault appeared more than any other, and it is not a craft problem. It is a mismatch between the ad and the product it advertises.

The pattern is familiar to anyone who installs games from ads: a puzzle mechanic that is not in the game, a rescue scenario that turns out to be a minor side mode, a building sequence that exists only as a menu. The ad is often well made on its own terms. It simply is not about the product.

It is worth understanding why this keeps happening, because it is not stupidity. Certain mechanics — extracting a pin, filling a container, sorting by colour — are unusually legible in three seconds, and a mechanic that reads instantly outperforms one that needs explaining. So the ad drifts toward what demonstrates well rather than what is true, one reasonable-seeming decision at a time.

The result is an ad that acquires people who wanted something else. Whatever it does to install numbers, it does the opposite to everything downstream, and the people making the ad are usually not the people who see that.

Near enough every ad in the set used a pointing hand, and it is the element we ended up most unsure about.

The case for it is strong and practical. A player has about a second to work out what to do, cannot be given instructions, and may never have seen this mechanic. A hand solves that immediately and in every language.

The case against is that it removes the discovery. A player who is shown exactly what to do has performed a gesture rather than worked something out, and the small satisfaction of understanding a system is one of the few things the format has to offer.

The versions that seemed to work best split the difference: the hand appears after a beat of hesitation rather than immediately, indicates a region rather than a precise target, and stops appearing once the player has acted correctly once. That keeps the teaching function and returns most of the discovery.

This is one of the few questions in the format that a controlled test would settle in an afternoon and that argument has never settled at all. Four variants of the same ad, one hand behaviour each, and the answer would be sitting in front of you by the evening. It is on our list for Field Notes.

The point of doing this is not our conclusions. It is that the exercise is genuinely useful and almost nobody in this industry does it deliberately — most people see hundreds of these ads and examine none of them.

A note on method, since this piece opens a section built on it.

Fifty is not a sample in any statistical sense, and ours was not randomly drawn — it came from the ads we were actually served, which reflects our phones, our locations and whatever the networks decided we looked like. A different fifty would produce a different list, and anyone treating this as a survey of the format is reading more into it than it holds.

Campaign performance stays inside the teams that ran it, so a teardown is a reading of craft rather than a report on results. That distinction matters more than it sounds: a poorly built ad can outperform a well built one for reasons that have nothing to do with the creative — a bigger budget, a better-matched audience, a product people already wanted. Anyone claiming to rank ads by effectiveness from the outside is ranking them by taste and calling it data.

What a teardown can actually do is narrower and still worth doing: it makes the decisions visible. Once you can see that an ad chose to stop at a particular moment, you can ask why, and then make that choice deliberately in your own work instead of inheriting it by default.