Game Tutorial Design: What a Three-Game Study Found

Ask what the instructions are supposed to help a player do

Game tutorial design involves more than deciding whether a message should appear. A first-time player might need to discover a control, understand a goal or recognize when to use an unfamiliar tool. Those are different instructional problems. A message that names a button may leave the reason for using it unexplained.

Start with an observable action: can the player perform the relevant move and recognize its result? This gives a tutorial a concrete purpose. It also makes the comparison clearer than assuming that more explanation must always produce a better introduction, regardless of the game or the audience.

The experiment varied tutorial features within three games

Andersen and colleagues' 2012 experiment involved over 45,000 players across Refraction, Hello Worlds and Foldit. New players were randomly assigned to eight conditions varying tutorial presence, context, restrictions on actions and optional help. The researchers recorded playing time, completed levels and returns.

That design examines what people did with particular instructions in particular games. It does not make every tutorial one interchangeable treatment. Introducing help at the relevant moment, presenting a manual first and requiring a specific action are choices that need to be identified before interpreting a result.

The clearest gains appeared in Foldit

For Foldit, the paper reports 29% longer play and 75% more completed levels with context-sensitive instructions versus no tutorials, without a significant return-rate improvement. Tutorial presence did not significantly improve time or progress in the other two games; some return-rate comparisons were less favorable.

These are separate outcomes. More completed levels does not mean 75% better understanding, and longer play does not establish a corresponding increase in enjoyment. Keep the comparison and its denominator attached to each percentage. Otherwise, a result about one recorded behavior can quietly become a much broader claim about learning or satisfaction.

Instruction timing is a question you can describe precisely

In Foldit, instructions presented in context also outperformed the up-front manual on time and progress.

Consider an invented robot puzzle. A player must rotate a mirror before a beam can reach a receiver. One version explains rotation before the first puzzle; another introduces that same explanation when the mirror becomes relevant. To discuss the difference clearly, specify the message, its trigger and what the player can still do while it appears.

This imagined scene illustrates a comparison, not a tested improvement. It separates the timing of an instruction from adding new information or making the puzzle easier. Without that distinction, a successful redesign could be credited to the wrong change.

Activity records do not reveal every part of learning

The study measured behavior without directly asking players what they were thinking or feeling.

Imagine two players who each finish seven levels. One can explain the rule; another has followed instructions successfully but cannot yet choose the rule independently. The same completion total could fit either account. That hypothetical difference shows why a progress measure and a separate understanding check answer different questions.

A useful interpretation states the measurement before the explanation. If a comparison records more activity, report that result. Questions about remembered knowledge, independent problem solving or enjoyment need evidence suited to those questions rather than being assumed from the activity total.

Keep the game and audience differences visible

The authors lacked demographic information and noted possible self-selection differences between games. Only three games and picture-and-text tutorials were studied.

Randomizing instructions within a game does not randomize people into different games. A contrast between their results therefore cannot isolate complexity from every other difference in audience, interface or entry conditions. Treat complexity as a relevant explanation to investigate, rather than a universal causal verdict.

This is a dated 2012 comparison, not a new feature announcement or a prescription to remove tutorials. The full paper and its university project page describe the same experiment, not independent replications. They support a more specific question: which instructional choice helps which observable task, for the players who encounter it?