Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 59 of the 63 references it cites were found in a registry or an index.

Fooled Twice – People Cannot Detect Deepfakes But Think They Can

Open access

Abstract

Hyper-realistic manipulation of audio-visual content, i.e., deepfakes, presents a new challenge for establishing veracity of online content. Research on the human impact of deepfakes, addressing both behaviors in response to and cognitive processing of deepfakes, remains sparse. In a pre-registered behavioral experiment ( N = 210), we show that (a) people cannot reliably detect deepfakes, and (b) neither raising awareness nor introducing financial incentives improves their detection accuracy. Zeroing in on the underlying cognitive processes, we find that (c) people are biased towards mistaking deepfakes as authentic videos (rather than vice versa) and (d) overestimate their own detection abilities. Together, these results suggest that people adopt a ``seeing-is-believing'' heuristic for deepfake detection while being overconfident in their (low) detection abilities. The combination renders people particularly susceptible to be influenced by inauthentic deepfake content.

Overview

Integrity signals

Registry

Crossref · 10.2139/ssrn.3832978

Cites

59 of 59 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Figures

Graphical abstract
Graphical abstract CC BY-NC-ND · PubMed Central
Figure 1 Illustration of video treatments Design of the experiment. Videos were separated into two sets, each consisting of eight deepfakes and eight authentic videos. Deepfake and authentic versions of the same videos were presented at the same slot in different sets.
Figure 1 Illustration of video treatments Design of the experiment. Videos were separated into two sets, each consisting of eight deepfakes and eight authentic videos. Deepfake and authentic versions of the same videos were presented at the same slot in different sets.CC BY-NC-ND · PubMed Central
Figure 2 Accuracy by video number Average accuracy levels for videos in descending order. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.
Figure 2 Accuracy by video number Average accuracy levels for videos in descending order. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.CC BY-NC-ND · PubMed Central
Figure 3 Accuracy dis-aggregated by video's number and video's type Average accuracy levels for each video, dis-aggregated for authentic (left) and fake (right) videos. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.
Figure 3 Accuracy dis-aggregated by video's number and video's type Average accuracy levels for each video, dis-aggregated for authentic (left) and fake (right) videos. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.CC BY-NC-ND · PubMed Central
Figure 4 Guessing behavior—Correct and “authentic” guesses (A) Violin plots of the distribution of the number of correct guesses a participant made (y axis) by the treatment (x axis). (B) Violin plots of the distribution of frequency of “authentic” guesses (y axis) by the treatment (x axis). Dark gray lines represent medians; the dark red dots represent means. Boxes indicate the interquartile range; each dot shows a raw data point. Plots created with the Ggstatsplot package (Patil, 2018).
Figure 4 Guessing behavior—Correct and “authentic” guesses (A) Violin plots of the distribution of the number of correct guesses a participant made (y axis) by the treatment (x axis). (B) Violin plots of the distribution of frequency of “authentic” guesses (y axis) by the treatment (x axis). Dark gray lines represent medians; the dark red dots represent means. Boxes indicate the interquartile range; each dot shows a raw data point. Plots created with the Ggstatsplot package (Patil, 2018).CC BY-NC-ND · PubMed Central
Figure 5 Fraction of correct guesses by fraction of “authentic” guesses Distribution of participants according to the percentage of authentic guesses (x axis) and the percentage of correct guesses (y axis) they made. The size of the bubble is proportional to the number of participants (exact N reported inside the bubble). The gray diamond delimits the maximum and minimum fractions of correct guesses conditional on the number of times the participant guessed that the video was authentic. Histograms on each side represent marginal distributions.
Figure 5 Fraction of correct guesses by fraction of “authentic” guesses Distribution of participants according to the percentage of authentic guesses (x axis) and the percentage of correct guesses (y axis) they made. The size of the bubble is proportional to the number of participants (exact N reported inside the bubble). The gray diamond delimits the maximum and minimum fractions of correct guesses conditional on the number of times the participant guessed that the video was authentic. Histograms on each side represent marginal distributions.CC BY-NC-ND · PubMed Central
Figure 6 Accuracy and confidence by video number Average accuracy levels, i.e., fraction of correct guesses, and average confidence, i.e., average belief about the probability to have guessed correctly the video, by video. Videos are ordered in descending order by level of accuracy. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.
Figure 6 Accuracy and confidence by video number Average accuracy levels, i.e., fraction of correct guesses, and average confidence, i.e., average belief about the probability to have guessed correctly the video, by video. Videos are ordered in descending order by level of accuracy. The error bars denote 95% confidence intervals. The red line indicates the 50% accuracy rate of random guessing.CC BY-NC-ND · PubMed Central
Figure 7 Overconfidence by number of correct guesses The figure displays the distribution of participants according to the number of correct guesses they made (x axis) and the overconfidence, i.e., believed minus actual number of correct guesses (y axis). The size of the bubble is proportional to the number of participants. The number of participants is reported inside the bubble. The downward sloping gray lines delimit the maximum and minimum levels of overconfidence that can be achieved conditional on the number of correct guesses. The dashed black line shows the linear regression.
Figure 7 Overconfidence by number of correct guesses The figure displays the distribution of participants according to the number of correct guesses they made (x axis) and the overconfidence, i.e., believed minus actual number of correct guesses (y axis). The size of the bubble is proportional to the number of participants. The number of participants is reported inside the bubble. The downward sloping gray lines delimit the maximum and minimum levels of overconfidence that can be achieved conditional on the number of correct guesses. The dashed black line shows the linear regression.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

We couldn’t retrieve this paper’s full text, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 63 references

Cited by

See all 194

Metadata

Authors
Köbis, N., Doležalová, B., & Soraperra, I.
Affiliations
  • Max Planck Institute for Human Development (DE)
  • University of Amsterdam (NL)
Corresponding author
Köbis, Nils
Venue
iScience
Publisher
Elsevier BV
Published
2021
Volume
24
Issue
11
Pages
103364-103364
ISSN
2589-0042 (ISSN-L)
OpenAlex
W3208526032
Cited by
194
References
63
Citations by year
2026: 64 · 2025: 48 · 2024: 48 · 2023: 23 · 2022: 11
FWCI
28.45
Type
journal article
Language
en
Open access
Licence
cc-by-nc-nd
Locations
  • iScience (publishedVersion, cc-by-nc-nd)
  • SSRN Electronic Journal (acceptedVersion)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • UvA-DARE (University of Amsterdam) (publishedVersion, cc-by-nc-nd)
  • Data Archiving and Networked Services (DANS) (submittedVersion, other-oa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Universiteit van Amsterdam (202006110906)
  • Max-Planck-Institut für Bildungsforschung
  • Toulouse School of Economics
Topics
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
  • Social Sciences › Psychology › Social Psychology › Deception detection and forensic psychology
Keywords
deepfakes, deepfake detection, misinformation, overconfidence, belief bias, awareness intervention, financial incentives

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Fooled Twice – People Cannot Detect Deepfakes But Think They Can. https://app.cochat.ai/papers/10.2139/ssrn.3832978
The paper
Köbis, N., Doležalová, B., & Soraperra, I. (2021). Fooled Twice – People Cannot Detect Deepfakes But Think They Can. iScience. https://doi.org/10.2139/ssrn.3832978

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.