Ghasemi et al.: People know what a default does — and aim it differently at allies and rivals — Fukai Reads

TL;DR

Every settings screen has options that are already selected. Difficulty on "Normal". Assists on. Board size 9×9. To the person building the game these look like placement decisions. To the player they read as a recommendation.

Today's paper measures whether people can deliberately use that preselected option as a tool. A well-known 2017 study concluded they cannot. Across three experiments here, participants set the default in their own favour on more than 80% of trials — the high-value card for teammates, the low-value card for opponents. The same person flips the placement when only the recipient changes.

In Experiment 3, people who saw nothing but someone else's default choice could work out which card was actually better 79.5% of the time. A default is not furniture. It is a message that gets sent and read.

The paper, and who wrote it

Eight authors. The first, Omid Ghasemi, and the last, Ben R. Newell, are at the Institute for Climate Risk & Response and the School of Psychology at UNSW Sydney. The other six are at universities and research institutes in Iran.

It appears in Judgment and Decision Making, volume 21, article e23, published on 4 September 2026. This is not an arXiv preprint — it is a peer-reviewed journal article, open access under CC-BY, so anyone can read the whole thing. DOI 10.1017/jdm.2026.10046. The authors state the study received no external funding. Data, analysis code and materials are posted on OSF.

I usually track game AI papers. I picked this one for two reasons. First, its subject is exactly the thing a designer can change tomorrow: the preselected value on a settings screen, the "recommended" tag, the preset. Second, it has the unusual shape of overturning a well-known earlier finding, which makes it a good place to think about how carefully psychology results should be carried into game design.

Background — the claim that people cannot use defaults

The default effect is the tendency to end up with whichever option was preselected. The standard illustration is organ donor registration, where countries that make participation the default show very different sign-up rates from those that do not. The effect itself has been studied for decades.

The harder question is the next one. Can someone who knows the effect exists actually use it on other people? In 2017, Julian J. Zlatev, David P. Daniels, Hajin Kim and Margaret A. Neale published a paper in PNAS concluding that they cannot. Across 11 samples totalling 2,844 participants, only 50.8% set the option they wanted chosen as the default — indistinguishable from a coin flip. They called it "default neglect".

Rebuttals followed. In 2018 Jung and colleagues published a reanalysis in the same journal showing people can learn when shown whether the default was accepted. In 2021 Craig R. M. McKenzie, Lim M. Leong and Shlomi Sher reported in Psychonomic Bulletin & Review that simplifying how the task is presented raises optimal default-setting to 87.7% (Experiment 1A) and 77.8% (Experiment 1B). The dispute had shifted from "do people lack the ability" to "was the task too confusing".

This paper enters there. It replaces the task with a legible card game, and then asks not merely whether people pick the right default but whether they change the placement depending on who is receiving it. That is the main departure from earlier work.

Approach — two cards, and a teammate, a rival, a stranger

Participants play the "informant": for another player, they set one of two cards as the preselected option. Whatever they set becomes that player's starting state.

There are two cards. The sure card always pays $50. The chance card pays $100 with probability p, where p is one of 5% / 10% / 50% / 90% / 95%. The expected value — the average payout over many repetitions — is therefore $5 / $10 / $50 / $90 / $95. At low p the sure card is better; at high p the chance card is.

Recipients come in types: a teammate whose gains become your gains, an opponent whose losses become your gains, an unknown player of unclear allegiance, and from Experiment 2 onward a neutral player whose outcome does not touch your earnings. The recipient never sees the probability or the amounts. Only the informant does.

One design detail matters a great deal. The informant is never told whether the recipient accepted the default. There is no outcome feedback at all; only the placement is measured.

Experiment 1: 160 members of the general public in Iran, run in Farsi, mean age 28.5; 15 trials (5 card pairs × 3 recipient types). Experiment 2: 161 UNSW undergraduates, mean age 19.9, 20 trials, where one card is already set as the default and changing it costs $10 out of a $200 budget. Experiment 3: 149 undergraduates, mean age 19.0, who did the 20-trial informant task plus an 8-trial observer block in which they saw only someone else's default choice and had to judge which card was more valuable.

All payoffs and costs were hypothetical. No real money moved. Keep that in mind — it comes back later.

Findings — over 80%, and a flip by recipient

In Experiment 1, participants set the default that served their own interest on 83.6% of trials [95% CI 80.1–86.6], clearly away from the 50% chance level (χ²(1)=497.36, p<.001). The gap from Zlatev et al.'s 50.8% is the paper's central claim.

The breakdown is where it gets interesting. For a teammate, the probability of setting the chance card rises from .14 when its expected value is $5 to .88 when it is $95. For an opponent, the same people run it backwards, from .80 down to .15. Good card to the ally, bad card to the rival. Change nothing but the label on the recipient and the placement inverts.

For the unknown recipient the slope was slight and tilted toward the opponent pattern, .56 down to .39 (b=−.008, p=.003). A weak effect.

Experiment 2 asked whether people would pay $10 to move the default. Optimal placement: 84.9% [81.8–87.5]. The pull of the preset itself is real — the group starting on the chance card kept choosing it more often (b=1.75, SE=.16, p<.001). But the pattern of tailoring by recipient was the same regardless of which card started as the default (the three-way interaction was not significant; χ²(3)=1.54, p=.673).

One result caught my eye. In the unknown condition of Experiment 2, the effect of expected value vanished (b=.002, p=.533). The faint slope from Experiment 1 did not replicate.

Experiment 3. As informants, participants were optimal on 86.8% of trials [84–89.2]. As observers, they were right 79.5% of the time [75.2–83.3]. The breakdown shows how they read it: when the chance card had been defaulted for a teammate, 83% judged the chance card the more valuable one; when it had been defaulted for an opponent, that fell to 28%. And when the sure card was defaulted for an opponent, 73% concluded the chance card must be better. Whatever is aimed at an enemy gets read as the bad option.

Participants also rated why they think defaults work, on a 0–100 scale: "it reads as an implicit recommendation" 65.2, "people don't want to lose what they have" 67.8, "thinking it through is effortful" 51.1. Their average estimate of how many recipients simply accept the default was 64.4%.

The authors close on an asymmetry: the helping direction came through more strongly than the harming direction. It reads as a stronger pull to help an ally than to hurt a rival.

Six things a game maker can take from this

1. The preselected difficulty is read as a recommendation. If observers could reverse-engineer the better option 79.5% of the time, players are actively asking why this one is selected. So don't pick the default because it is safe. Pick it because you want to recommend it. If you can't write one line next to it explaining why you recommend it, you haven't chosen it yet.

2. Assume shop and subscription presets will be treated as suspect. The 73% / 28% asymmetry shows how readily a default aimed at a rival gets reread as the worse option. The moment a player feels a preset is tilted toward the seller, it stops persuading and starts warning. If you preselect a recommended bundle, make it visible who benefits.

3. Don't protect a default with switching costs. In Experiment 2 people paid $10 to move it. Making a preset hard to change does not keep people on it, and it reads as hostility. Aim for a preset nobody needs to move, not one they can't.

4. In asymmetric or competitive games where one side sets the options, don't expect neutrality. Map picks, handicaps, proposed house rules — the flip by recipient label is measured, not hypothetical. Show who set the option, or alternate who gets to set it. If I were building an asymmetric competitive puzzle, I would decide up front that the name of whoever set the configuration is always displayed.

5. A level's starting arrangement is a default too. In a puzzle like Baba Is You, where the rules themselves are movable, the arrangement at the moment the level opens is the author's chosen starting point for interpretation. Players read it as intent before they read it as difficulty. Treating the initial layout as message rather than merely as difficulty tightens the design.

6. When you can't read the recipient's interests, don't guess. The one condition where this study's results disagreed with themselves was "unknown" — a faint effect in Experiment 1, none in Experiment 2. And "unknown" is precisely the position a designer occupies. You do not know what the player in front of you wants. So set defaults from measurement — drop-off, settings-change rate, first-clear rate — not from intuition. Even a design as transparent as Into the Breach, which shows you what is coming, still has an author deciding what to show first.

Limits — theirs, and mine

Start with what the authors concede. Participants were explicitly told they would be perceived as supportive teammates. Who is friend and who is foe is given from the outset. The authors write that "in real-world settings, such clarity is rare", and note that real choice architects work where incentives are not spelled out.

The second concession bites harder. Informants never observed whether the default was accepted. So, as the authors state, "optimal" here means optimal given participants' beliefs about how well defaults work. Actual influence was not measured. All payoffs were hypothetical. They limit the conclusion to showing that people are capable of strategic default-setting under clear, controlled conditions, and call it an open question whether they do so spontaneously without prompting.

Three things I would add. First, I find no preregistration statement — no prior public record of the hypotheses and analysis plan. Data and materials are on OSF, so the work is checkable, but when carrying a psychology result into design I want to hold it loosely until a replication appears.

Second, the samples are not matched across experiments: Iranian adults in Farsi (mean age 28.5) in Experiment 1, Australian undergraduates in English (19.9 and 19.0) in Experiments 2 and 3. The authors treat the consistency of the pattern as a strength, which is fair. But "consistent across different groups" and "replicated in the same group" are different claims.

Third, and this is what stuck with me: the unknown condition. A faint effect in Experiment 1 (p=.003), nothing in Experiment 2 (p=.533). That disagreement does not threaten the paper's main claim. But the position a game designer actually occupies is exactly this one — setting a default for a recipient who is neither ally nor rival, and whose interests you cannot see. The only unstable measurement was the one covering the ground we stand on. That seems worth carrying away separately from the conclusion.

How I read it

This section is my own reading. I want to place this study in a line of thinking that says a default is not furniture but an utterance. The 2017 "default neglect" result painted people as incompetent senders. This paper reads as showing that, at least when conditions are legible, they can choose what to send. But the part that matters to me is Experiment 3 — the receiver's competence. In the vocabulary of design criticism, setting a default is not UI placement; it is the first sentence the author speaks to the player. If intent is reverse-engineered at 79.5% accuracy, then clumsiness in the placement can be read at roughly the same accuracy. A "Normal" dropped carelessly onto a settings screen is not silent. It has already said something.

Closing — three papers that draw the map

Read alone, this paper looks like a simple correction: default neglect was wrong. Line up four papers and something better appears. Zlatev et al. (2017) said people cannot use defaults; Jung et al. (2018) answered that they learn given feedback; McKenzie et al. (2021) showed that a simpler task lifts performance to around 80%; and now Ghasemi et al. show people tailor the default to the recipient.

You can watch the question move from "is there an effect" to "under what task does the ability show up". That movement is itself the lesson for bringing psychology into game design. Don't take a result as "people are like this". Take it as "under these conditions this was observed", then check how close your design is to those conditions. Here, our design sits on the "unknown recipient" side of the board.

The note I wrote for myself: next time I build a settings screen, ask whether I can write one line beside the default saying why I recommend it. If I can't, it hasn't been chosen yet.

References

Papers and materials referenced in this article:

・People are aware of the impact of defaults and use them strategically (Ghasemi, Ghane-Ezabadi, Hojjat, Afshin Mahjoub, Arastouy Irani, Gholampour, Solhirad & Newell, 2026, Judgment and Decision Making, 21, e23) — the subject of this article; peer-reviewed, open access (CC-BY)

・OSF: data, analysis code and materials

・Default neglect in attempts at social influence (Zlatev, Daniels, Kim & Neale, 2017, PNAS, 114(52), 13643–13648) — the paper that proposed "default neglect"

・Default sensitivity in attempts at social influence (McKenzie, Leong & Sher, 2021, Psychonomic Bulletin & Review, 28(2), 695–702) — the rebuttal showing simpler tasks raise performance

・People can recognize, learn, and apply default effects in social influence (Jung et al., 2018, PNAS) — a reanalysis of Zlatev et al.'s data