Summary:
A cognitive psychology study explored how people use explicit reliability information when deciding which sources to trust. The experiments, funded by the Economic and Social Research Council (ESRC), show that while participants tended to favor sources labeled as more credible, they could not fully ignore inputs explicitly marked as no better than chance. Information tagged with a 50% chance of being correct—a value that mathematically carries no diagnostic weight—nonetheless influenced decisions. Metacognitive ratings revealed that participants were partly aware that unreliable inputs affected their choices, but that awareness did not prevent those inputs from biasing outcomes.
Key Facts:
- Persistent bias from zero-value data: Inputs explicitly labeled as 50% likely to be correct—i.e., equivalent to random chance and therefore uninformative—still swayed participants toward the content of those inputs.
- Imperfect reliability weighting: Participants generally gave greater weight to sources marked as reliable, yet their integration of those reliability metrics deviated from the optimal probabilistic calculations researchers would expect.
- Metacognitive awareness: When asked to evaluate the influence of each source, participants reported that unreliable sources had affected their judgments, indicating a gap between awareness and the ability to exclude unwanted information.
Source: Neuroscience News / ESRC Research Project
In modern digital environments, people make hundreds of judgments daily based on a mix of social feeds, algorithmic news, user reviews, and direct messages—each carrying widely varying reliability. To help users, platforms increasingly add quantitative signals such as credibility scores, verified badges, and star ratings to indicate source trustworthiness.
This raises a central cognitive question: when a source’s reliability is presented as an explicit probability, do people use that number appropriately to form objective judgments?
A set of behavioral and metacognitive experiments addressing how explicit reliability information is combined shows a clear limitation: people struggle to exclude dubious inputs, even when those inputs are explicitly labeled as uninformative. In practice, data marked as equivalent to a coin flip still biased participants’ decisions.
The cognitive friction of multi-source integration
In the main experimental task, participants had to combine predictions from several independent sources to choose between two alternatives. Each source came with an explicit numerical reliability score that indicated the stated probability its claim was correct.
Behavioral outcomes made two patterns clear:
- Proportional weighting: Participants were able to distinguish sources and generally weighted higher-scoring sources more heavily, showing an intuitive sensitivity to reliability labels.
- Calibration failures: Despite that sensitivity, people did not treat the stated probabilities as exact mathematical inputs. Their aggregation of evidence departed systematically from optimal Bayesian updating.
- The 50% problem: Most strikingly, information tagged as 50% accurate—mathematically neutral and diagnostically meaningless—was not treated as neutral. Instead, it shifted choices in the direction of its content.
Rather than converting entirely to neutral background noise, such uninformative inputs seem to be processed at a semantic level, leaving a residual effect on the decision process that explicit probability markers fail to eliminate.
Recognizing the bias: metacognitive gaps
To determine whether participants noticed this contamination, researchers conducted follow-up tests of metacognitive awareness. After making decisions, participants rated how much each source had influenced their final choice.
Results showed a nuanced split: people were consciously aware, to some extent, that low-credibility inputs had affected their judgments. Yet this awareness did not translate into cognitive control sufficient to isolate or neutralize those inputs during decision making.
In short, participants both recognized and struggled to prevent the unwanted influence of unreliable evidence—an important dissociation between monitoring and control in everyday information processing.
System design and the limits of fact-checking
These findings have practical implications for platform designers, content moderators, and public communicators. Labeling claims with “low credibility” or presenting numerical reliability indicators may not be enough to stop unreliable content from shaping beliefs and choices, because once content is read it can leave an imprint that numerical tags alone do not remove.
Designers should therefore consider how credibility signals are presented and combined with other interface strategies, media literacy training, and moderation practices that reduce the initial impact of uninformative content. Understanding how people misinterpret, partially internalize, and fail to fully disregard explicit reliability metrics is crucial for developing better tools to limit misinformation.
Funding: This work was supported by ESRC grant ES/V00378X/1 awarded to Lucie Charles and Patrick Haggard.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by staff.
About this psychology Research:
- Media Contact: YAMASHINA Naoko
- Source: University of Tsukuba
- Image Credit: Image credited to Neuroscience News
- Original Research is Open Access: PLOS Computational Biology (Oct 1, 2026). “How does explicit reliability guide choices? Effect of trustworthiness on choice and metacognition.” Authors: Keiji Ota, Anthony Ciston, Patrick Haggard, Thibault Gajdos-Preuss, and Lucie Charles.
- DOI: 10.1371/journal.pcbi.1014818
Abstract
How does explicit reliability guide choices? Effect of trustworthiness on choice and metacognition
Integrating information from multiple sources with varying reliability is a central challenge in contemporary information environments. Despite the widespread use of explicit probabilistic indicators of source accuracy, we know little about how such markers are used in decision-making.
This research introduced a task in which participants viewed predictions from sources with explicit, stated reliability values to choose between two options. After each decision, participants reported how much each source had influenced their choice. Computational modeling estimated the effective reliability participants assigned to each source.
Findings indicate that people do not accept stated reliability values at face value. Participants systematically distorted the provided probabilities when combining evidence, and even sources explicitly labeled as unreliable biased choices as if they held moderate credibility. Furthermore, the presence of sources known to predict incorrect answers degraded performance by introducing leakiness into the evidence accumulation process.
Despite these distortions, participants retained some metacognitive insight: they could identify and report the influence of unreliable sources on their decisions. Overall, the results show a divergence between how people consciously perceive the role of reliability markers and how those markers are actually integrated into choice—a gap with direct implications for platform design, fact-checking, and media literacy initiatives.
