0 cumulative citations
View corpus contextSource citations alone don’t reduce false beliefs: they boost detection of deliberate disinformation only for digitally literate users, while offering no help for unintentional misinformation or for people who rely on intuitive processing.
The Roles of Source Citations in Identifying Misinformation and Disinformation: Moderating Effects of Digital Literacy and Information Types
openalex
rct
medium evidence
7/10 relevance
Summary only
summary available; pdf_status=paywall
DOI
Source
PDF
Providing source citations does not improve overall detection of false information, but citations help detect deliberate disinformation among respondents with higher digital literacy while offering no benefit for misinformation or for intuitive processors.
Citation observations
Cumulative provider counts captured on specific dates; providers are never combined.
This study investigates whether source citations enhance detection of false information and whether their effects depend on digital literacy and information processing style. In an experimental survey of 1,013 South Koreans, citations show no overall effect. However, they significantly improve detection of disinformation—but not misinformation—among individuals with higher digital literacy. Intuitive processing lowers detection ability, and citations do not counteract this tendency. The findings challenge assumptions that transparency universally improves evaluation, showing that citation effectiveness depends on user capabilities and the type of falsehood, and highlighting the need to strengthen digital literacy alongside transparency initiatives.
Summary
Main Finding
Source citations do not improve false-information detection overall. However, they significantly help detect disinformation (deliberate falsehoods) for people with higher digital literacy. Citations do not improve detection of misinformation (unintentional falsehoods), nor do they offset the negative effect of intuitive (vs. reflective) information processing.
Key Points
- Sample: experimental survey of 1,013 South Korean respondents.
- Treatment: presence versus absence of source citations (experimental manipulation).
- Heterogeneous effects:
- No average treatment effect of citations on ability to spot false information.
- Positive effect of citations on detecting disinformation, but only among respondents with higher digital literacy.
- No citation benefit for detecting misinformation.
- Cognitive style matters: respondents who rely on intuitive processing are worse at spotting false information, and adding citations does not mitigate that deficit.
- Conceptual implication: transparency (here, providing citations) is not universally effective — its utility depends on user capabilities and the type of falsehood.
Data & Methods
- Design: randomized experimental survey (n = 1,013) that varied whether items included source citations and presented both misinformation and disinformation stimuli.
- Key measurements:
- Outcome: ability to detect false information (separately for misinformation and disinformation).
- Moderators: digital literacy (measured via standard digital-literacy items) and information-processing style (intuitive vs. reflective).
- Analysis: estimation of average treatment effects and interaction terms to assess heterogeneity by digital literacy and cognitive style; statistical significance reported for the interaction between citations and digital literacy on disinformation detection.
- Limitations (noted or implicit): single-country sample (South Korea) may limit external generalizability; survey/experimental context may differ from real-world information environments; details on effect sizes and exact measures not provided here.
Implications for AI Economics
- Transparency alone is an insufficient policy lever: simply attaching provenance/citation metadata to algorithmic outputs (e.g., model citations, source links) will not uniformly increase user scrutiny or reduce false-belief uptake.
- Targeted complementarities: investments in provenance mechanisms should be coupled with digital-literacy interventions; the return to citation/provenance features is higher among more digitally literate users.
- Heterogeneous demand for verification tools: economic models of information markets should account for user heterogeneity (skills and cognitive style) when predicting uptake and welfare effects of transparency technologies.
- Design trade-offs: platform design and regulation should consider the marginal benefit of provenance features conditional on user capabilities; for low-literacy or highly intuitive audiences, alternative interventions (e.g., automated fact-check signals, friction, interface nudges) may be more cost-effective.
- Policy prioritization: regulators promoting disclosure/transparency (e.g., source tags, model cards) should also fund and target digital-literacy programs to maximize social returns and reduce harms from disinformation.
- Evaluation metrics: empirical assessment of transparency features should report heterogeneous effects (by literacy and cognitive style) and separately examine misinformation versus disinformation, as their responsiveness to interventions differs.
Assessment
Paper Typerct
Evidence Strengthmedium — A preregistered RCT (n=1,013) provides strong internal validity for the causal effect of citations in the survey setting and for heterogeneity analyses, but the average null effect, reliance on a single-country sample, survey/experimental context (not real-world platforms), and limited detail here on pre-registration, power for subgroups, and measurement validity constrain overall strength.
Methods Rigormedium — Randomized design and interaction analyses are appropriate and give credible internal validity; however, summary lacks detail on randomization checks, balance, pre-registration, exact outcome measures and effect sizes, multiple-hypothesis adjustments, and ecological validity (external context), so methodological rigor appears solid but not demonstrably best-practice.
SampleExperimental online survey of 1,013 adult respondents in South Korea; respondents exposed to stimuli labeled as misinformation (unintentional falsehoods) or disinformation (deliberate falsehoods) with randomized presence vs. absence of source citations; moderators include measured digital literacy and self-reported intuitive vs. reflective information-processing style.
Themesgovernance adoption
IdentificationRandomized experimental manipulation of whether each survey stimulus included source citations (presence vs. absence); randomized assignment ensures causal identification of the citation treatment on detection outcomes.
GeneralizabilitySingle-country sample (South Korea) — cultural and media-system differences may limit transferability to other countries., Survey/experimental stimuli may not replicate real-world platform dynamics, incentives, social context, or information overload., Findings depend on the specific citation format, quality, and credibility of presented sources (not fully described here)., Results may not generalize to different populations (age, education, online habits) or to automated AI-generated content outside the experimental framing.
Claims (7)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| Providing source citations does not improve respondents' overall ability to detect false information. Decision Quality | null_result | Overall ability to detect false information |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| Source citations improve detection of disinformation among respondents with higher digital literacy. Decision Quality | positive | Ability to detect disinformation, defined as deliberate falsehoods |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| Source citations do not improve detection of misinformation. Decision Quality | null_result | Ability to detect misinformation, defined as unintentional falsehoods |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| Respondents who rely more on intuitive rather than reflective information processing are worse at detecting false information. Decision Quality | negative | Ability to detect false information |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| Adding source citations does not mitigate the lower false-information detection ability associated with intuitive information processing. Decision Quality | null_result | Ability to detect false information conditional on intuitive versus reflective processing style |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| The effect of citations on false-information detection depends on both users' digital literacy and the type of falsehood being evaluated. Decision Quality | mixed | False-information detection, separated into misinformation and disinformation detection |
Reading fidelity
high
Study strength
medium
|
n=1013
|
| Transparency interventions such as citations should be paired with digital-literacy interventions rather than treated as a universally effective standalone policy lever. Governance And Regulation | positive | Effectiveness of transparency interventions in improving false-information detection |
Reading fidelity
high
Study strength
low
|
n=1013
|