Teaching People to Doubt Everything
The misinformation games governments adopted were scored on whether people rated fake headlines lower. A reanalysis separating scepticism from accuracy found they moved the first and not the second.
Key takeaways
- There are two ways to score better on these tests: get better at telling true from false, or become more sceptical of everything. Both lower the average rating of fake headlines.
- The published evaluations compared mean ratings of true and fake items before and after playing, often against a control group playing Tetris or doing nothing.
- Modirrousta-Galian and Higham reanalysed that work using receiver operating characteristic analysis and found the games did not improve discrimination.
- Seabrooke, Modirrousta-Galian and Higham then re-examined the Bad News game using Indian headlines and reported no evidence of improved discrimination.
- Our audit of Foolproof records that this distinction does not appear in the book, and that no critic of the programme is named across its source notes.
Bad News is one of the misinformation games, a browser game in which you play a disinformation merchant and work your way up from obscurity to a following using the techniques real ones use. It takes about fifteen minutes, it has been translated and deployed widely, and it is among the better-known psychological interventions of the past decade. We are not going to put figures on the reach, because we do not have a source for them.
The question this article is about is not whether it works. It is what "works" was measuring.
The two things the studies ran together
There are two ways a person can get better at the test the misinformation games are scored on, and they are not the same thing at all.
The first is discrimination: you become better at telling a true headline from a false one. The gap between how you rate the two widens. That is the stated aim and the thing worth having.
The second is response bias: you become more sceptical of everything. You rate the false headlines lower, and you rate the true ones lower too. The gap does not widen. You have not learned to tell them apart, you have learned to trust less.
Both show up as a fall in the average rating of fake headlines, which is what the published evaluations of the misinformation games mostly measured.
### Read what the reanalysis actually did
Ariana Modirrousta-Galian and Philip Higham set this out in the Journal of Experimental Psychology: General in 2023, and their contribution is methodological rather than a new experiment.
They describe the existing designs plainly. Participants rated the reliability or manipulativeness of true and fake news items before and after playing Bad News or Go Viral, usually with a control group that played Tetris or did nothing. Mean ratings were then compared between pre-test and post-test, or between the control and experimental conditions.
Then they reanalysed that body of work using receiver operating characteristic analysis, which is the standard tool for pulling discrimination apart from bias, and which asks whether an instrument measures what it is said to measure. The title states the result: gamified inoculation interventions do not improve discrimination between true and fake news.
What the games moved, in that corpus, was the bias term. People came out more sceptical and no better at sorting. Hold that loosely, because the follow-up below did not find the bias shift either.
### Read the follow-up carefully, including the part against us
A new study followed, and it complicates the bias story as well as the discrimination one.
Tina Seabrooke, with Modirrousta-Galian and Higham, re-examined the Bad News game in Psychonomic Bulletin & Review in 2025 using Indian true and fake news headlines, and reported no evidence of improved discrimination.
It also found no significant shift in response bias once both counterbalancing conditions were included. The conservative bias shift appeared only in the single condition matching the earlier experiment it was re-examining. So this study does not support the tidy claim that the games reliably make people more sceptical; it finds neither effect.
Two further things a reader should have. The sample was 150, which is small beside the figures that get quoted in this area. And it is a re-examination of somebody else's Indian-sample study with counterbalancing added, by overlapping authors, so calling it an independent replication would be wrong and we are not going to.
### Notice what this does not say
Four things, and we would rather list them than be caught having skipped one.
1. It does not say the misinformation games do nothing. Raising scepticism is a real psychological effect and may even be desirable in some settings. It says the effect is not the advertised one.
2. It does not say inoculation theory is wrong, and the misinformation games are one application of a much older idea. Separating what has survived re-examination and what has not is the work, and it is not finished here. The theory is sixty years old and much larger than two browser games, and nothing here speaks to its other applications.
3. It is one research group. A reanalysis and a follow-up by overlapping authors is not the same as independent confirmation from elsewhere, and we would be making the opposite complaint if the positive results all came from one lab, which is roughly the situation they are criticising.
4. And raising scepticism indiscriminately has a cost nobody in this exchange has measured. If an intervention makes people distrust accurate reporting as readily as fabricated reporting, that is a harm, and we have not seen it quantified in either direction.
### Read what the original group did next
Because they did respond, in the literature rather than in an argument, and it is the strongest thing against this article's case.
Johannes Leder, Lukas Schellinger, Rakoen Maertens, Sander van der Linden, Breanne Chryst and Jon Roozenbeek published in the Journal of Experimental Psychology: General in 2024, under a title that states their result: feedback exercises boost discernment of misinformation for gamified inoculation interventions.
Their own abstract names the two problems being addressed, and names them as the critics did. A potentially inadvertent impact of the games on people's evaluation of real news. And exponential decay of the effect over time where no memory-strengthening exercise is provided. Two preregistered laboratory experiments with 191 and 321 participants, plus four in-game surveys with 559, 2,558, 419 and 882.
A correction made in the literature rather than in a rebuttal is the better kind, and conceding the problem in your own abstract before testing a fix deserves to be said plainly. We have read the abstract and not the paper, so we are not going to characterise how well the fix works.
What it does to our position is specific. The claim that the games as originally deployed improved discrimination is in serious trouble. The claim that gamified inoculation cannot improve discrimination is not something anybody should be asserting, and this article is not asserting it.
Hold the book to its own standard
Sander van der Linden's Foolproof is the trade account of this research programme and we have audited it in this vault, so we can be specific about what it does well and what it leaves out.
It does several things better than books of its kind usually manage. It prints its own null results. It reports effect-size decay. It is correct and careful about the backfire effect, reporting it as much rarer than was once believed and confining the familiarity version to hedged language, which matches the current state of the literature.
What our audit records as absent is this argument. The distinction between a bias shift and a discrimination gain is the central methodological question in this area, and it is not in the book, which also names no critic of the programme across its source notes. That absence is about the book rather than about the programme, and the 2024 paper above is the reason the distinction matters: the research group has engaged with the response-bias problem in the literature while the trade account does not raise it.
Ask the question a reader can actually use
The transferable thing here has nothing to do with misinformation games.
Any intervention that claims to make you better at spotting something bad can succeed by making you suspicious of everything, and the two look identical on most measures. So the question to put to a media-literacy course, a fraud-awareness training, a scam-spotting campaign or an advertising-claims checklist is: did people get better at telling the difference, or did they just get warier?
If the evaluation only reports how badly participants now rate the bad examples, it cannot tell you.
Where we land on the evidence
Our reading is that the critics have the better of this, and that the honest summary is narrower than either side's headline.
The misinformation games demonstrably change how people rate news. The claim that they improve discrimination has not survived reanalysis or an independent-material replication, and that is the claim the deployments were justified by. We would stop describing them as improving detection.
What we would keep is the ambition and the format. A fifteen-minute game that reaches millions is a genuinely better delivery mechanism than a leaflet, and the finding here is about what to measure rather than about whether to try.
The question that is still open
Not the one we were going to ask, because the 2024 paper has already taken up the inadvertent effect on real news, which is half of it.
What remains is the net question, and it is harder. These interventions are deployed at population scale on the strength of laboratory and in-game measures. Nobody in this exchange reports what happens to a population's overall trust in accurate reporting after deployment, over months, outside a study. Both sides are arguing about instruments measured within minutes of play, and the thing being bought is a change in how millions of people read the news for years.
That is answerable in principle and expensive in practice, which is probably why it has not been done.
Common questions
Do misinformation games like Bad News actually work?
They change how people rate news, and what they change is contested. Modirrousta-Galian and Higham reanalysed the published evaluations using receiver operating characteristic analysis and found no improvement in discrimination between true and fake news. A 2025 follow-up on new material found no discrimination gain and no reliable shift in scepticism either.
What is the difference between response bias and discrimination?
Discrimination is how well you separate true from false: the gap between how you rate the two. Response bias is how readily you say yes at all. Becoming sceptical of everything lowers your ratings of fake headlines and of real ones, so it looks identical to a genuine improvement on any measure that reports only the average.
Has the original research group responded?
Yes, in the literature. Leder, Schellinger, Maertens, van der Linden, Chryst and Roozenbeek published in the Journal of Experimental Psychology: General in 2024 on feedback exercises, naming the inadvertent effect on evaluation of real news and the decay over time as the problems being addressed, and testing a fix across six studies.
How should a media literacy course be evaluated?
By whether participants got better at telling the difference, rather than by whether they rated the bad examples worse. If the evaluation reports only how harshly people now judge the fake items, it cannot distinguish accuracy from caution.
Sources
Modirrousta-Galian, A., & Higham, P. A. (2023). 'Gamified inoculation interventions do not improve discrimination between true and fake news: reanalyzing existing research with receiver operating characteristic analysis.' Journal of Experimental Psychology: General, 152(9), 2411-2437. Describes the prior pre-post designs, in which participants rated reliability or manipulativeness of true and fake items before and after playing Bad News or Go Viral, usually with a control group playing Tetris or doing nothing, and notes those studies did not separate discrimination from response bias. Verified at the published record 2026-10-05. doi:10.1037/xge0001395
Seabrooke, T., Modirrousta-Galian, A., & Higham, P. A. (2025). 'Re-examining the bad news game: no evidence of improved discrimination of Indian true and fake news headlines.' Psychonomic Bulletin & Review. Verified at the published record 2026-10-05. doi:10.3758/s13423-025-02827-x
van der Linden, S. (2023). Foolproof. Read in full in this vault, 2026-09-26, with a reliability audit. The audit records that the book prints its own nulls and effect-size decay and is correct and careful about the backfire effect, and that the response-bias distinction does not appear in it and no critic of the programme is named across 552 note pointers. The audit also records a later paper from the same research group adding a feedback exercise to the games, whose abstract names an inadvertent effect on evaluation of real news as the thing being addressed; that paper has not been read here and is reported as the audit reported it.
A note on independence: the reanalysis and the follow-up share authors, so this is one research group rather than independent confirmation from several, and the article says so.