People Still Follow Sources They Know Are Unreliable, Study Finds

Told to Ignore a Source, People Still Lean on It

A source labeled 50% reliable is no better than a coin flip, and a new study shows that people lean on it anyway. Participants were told in plain terms to ignore these sources, yet their choices drifted toward whatever color the useless sources pointed to. They showed some awareness of being swayed, and it did not stop them.

Researchers from Britain, Japan, Germany, and France ran five online experiments, reported in PLOS Computational Biology, to see how people use reliability scores.

Labels like these are common online, from star ratings on products to fact-check flags on social posts. The experiments used colored squares instead of real news, so the results do not show how people handle actual misinformation. They offer a controlled look at what happens when a number says to stop trusting a source.

Unreliable Sources Swayed Choices in All Five Online Experiments

Of the 158 people recruited online from the United Kingdom, 109 were women, and average ages in each experiment ran from about 25 to 30. After 18 who rushed through the task were removed, 140 remained, in groups of 23 to 32 per experiment.

On each round, participants watched six colored squares, red or blue, each predicting the correct color for that round. A percentage above every square showed how often that source told the truth. Sources above 50% were right more often than wrong. Sources at exactly 50%, the unreliable ones, were pure coin flips. Sources below 50% were usually wrong, so a sharp participant would bet on the opposite color. After the sixth square, participants picked a color and learned whether they were correct. Each person completed roughly 300 rounds.

Source mixes changed across experiments. Two used coin-flip and mostly right sources, two added a usually wrong source to the mix, and one used only coin-flip and usually wrong sources.

Participants beat random guessing, picking the correct color about 58% to 64% of the time. Even so, coin-flip sources left a mark in every experiment. When more of the 50% squares showed blue, participants grew more likely to pick blue, even though those squares held no information. A statistical model of the choices estimated that people treated these sources as right between about 54% and 62% of the time. In effect, no information became a little information. The authors floated several explanations, including that seeing a color repeatedly may nudge preference toward it, and that people may have used the useless squares to break ties when the real evidence was thin. The coin-flip pull was weaker in the two experiments that mixed mostly right and usually wrong sources, perhaps, the authors say, because reading colors both ways made the colors themselves matter less.

Sources That Are Usually Wrong Proved Hardest to Use

Sources labeled below 50% caused a different problem. A source that is wrong most of the time is valuable once its advice gets reversed, and a 35% source carries as much information as a 65% one. Participants did not treat them equally. Their choices followed the wrong sources less than they should have, and accuracy fell in the experiment where every source was either a coin flip or usually wrong, compared with the matching experiment using mostly right sources. Smaller dips in two other experiments could have been chance.

Order mattered, too. The last square counted for more than the first, with the first square’s influence running 11% to 40% lower than the last square’s, depending on the experiment. In the researchers’ model, that fading looked stronger in the experiment where most sources were misleading. The authors propose that flipping a source’s advice is mentally costly, which may strain short-term memory for earlier squares.

Participants Sensed the Pull of Unreliable Sources but Kept Following Them

In the last two experiments, 64 participants rated, after each choice, how much one randomly chosen group of squares, such as the 65% ones, had influenced them, on a scale running from going against the squares, through no influence, to going along with them.

Ratings moved in step with the evidence. The more coin-flip squares that agreed with a person’s pick, the more that person said those squares had nudged the decision. Across individuals, people who leaned harder on a given source tended to report feeling its influence more strongly.

Yet the authors suggest that “awareness of bias is not necessarily sufficient to overcome it.” The experiments never tested whether people could resist when asked to, so an inability to resist is an interpretation, not a proven limit. One possible reason, the authors say, is that the bias was not tied to a clear drop in accuracy in most experiments, so resisting it may not have seemed worth the effort. A real-world version might be a flagged post whose claim still lingers in a reader’s mind.

A coin flip labeled as a coin flip still tugged at people’s choices. Designers of credibility scores and fact-check flags should test how people respond to labels, not assume a clear label does the work.

---

Disclaimer: This article reports on the findings of a single peer-reviewed study and is intended for general informational purposes only. It is not professional, medical, financial, or legal advice, and results from controlled experiments may not apply to every situation.

---