Prediction of self-efficacy in recognizing deepfakes based on personality traits 

Background: While deepfake technology is still relatively new, concerns are increasing as they are getting harder to spot. The first question we need to ask is how good humans are at recognizing deepfakes - the realistic-looking videos or images that show people doing or saying things that they never actually did or said generated by an artificial intelligence-based technology. Research has shown that an individual’s self-efficacy correlates with their ability to detect deepfakes. Previous studies suggest that one of the most fundamental predictors of self-efficacy are personality traits. In this study, we ask the question: how can people’s personality traits influence their efficacy in recognizing deepfakes? Methods: Predictive correlational design with a multiple linear regression data analysis technique was used in this study. The participants of this study were 200 Indonesian young adults. Results: The results showed that only traits of Honesty-humility and Agreeableness were able to predict the efficacy, in the negative and positive directions, respectively. Meanwhile, traits of Emotionality, Extraversion, Conscientiousness, and Openness cannot predict it. Conclusion: Self-efficacy in spotting deepfakes can be predicted by certain personality traits.


Introduction
One of the biggest threats and disruptions to privacy and democracy in this digital age is deepfake technology.A 'deepfake' or synthetic media, is a video editing technology that manipulates and mimic a person's facial expressions, mannerisms, voice, and inflections based on a large amount of data of other people to create a hyper-realistic video depicting them doing or saying things that never happened (Westerlund, 2019).
The current consensus is that the average human's ability in recognizing deepfakes is similar to the machines (Vitak, 2022).However, the result seems to vary depending on their own confidence and belief in their cognitive abilities.Some studies suggest that some individual differences determine if a person is good at recognizing deepfakes or not (Shahid et al., 2022).In this study, we will look at the relationship between personality traits and people's self-reported efficacy in recognizing deepfakes.
The HEXACO personality model describes six facets of personality structures: Honesty-humility, Emotionality, Extraversion, Agreeableness, Conscientiousness, and Openness to experience (Lee & Ashton, 2009;unpublished report;Zettler et al., 2020).This instrument model is selected because of three reasons: (1) the measurement covers a wider and more complex range of personality facets that go beyond the five-factor model (Ashton & Lee, 2007); (2) the Honesty-Humility factor measures traits like sincerity, boastfulness, pretentiousness, and fair-mindedness that are associated with dishonest or inauthentic behaviors (Ashton & Lee, 2008) relating to self-reported efficacy; and (3) the model provides flexibility in measuring contextually unique situations (Oostrom et al., 2019;Pletzer et al., 2020).Advancements in information technology, including AI, socially intelligent robots, and other autonomous systems, will have a profound impact on human life, necessitating research in typical personality to understand and address individual differences in adapting to these new challenges (Matthews et al., 2021), not to mention deepfakes.Multiple studies in various contexts have shown that personality traits influence an individual's self-efficacy (Lodewyk, 2018).
The Honesty-humility dimension reflects an individual's fair-mindedness, modesty, and cooperation.A person with high Honesty-humility might not think they are good at recognizing deepfakes, regardless of their true ability while an individual with low Honesty-humility might be biased in their self-reported ability in recognizing a deepfake.
Emotionality reflects an individual's degree of anxiousness, fearfulness, and sentimentality -the experience of anxiety in response to life's stressors.To overcome this anxiety, the sense of being able to recognize deepfakes is important to reduce that anxiety.One way to become less anxious is to appreciate deepfakes as a "cultural technology" (Cover, 2022) that contains artistic and creative values.People with high Emotionality may be more motivated to use deepfakes as an "antidote" from the pressures of everyday life, so they have higher self-reported efficacy to detect them, not to be avoided but as potential things to be used according to their interests (technology appropriation; see Prayoga & Abraham, 2017) Extraversion reflects an individual's degree of sociability.Individuals high in Extraversion might have higher selfefficacy due to their higher social esteem, boldness, and familiarity.Van der Zee et al. (2002) found that extroverts are friendly and less formal in their interactions with others.This is closely connected with emotion recognition (part of emotional intelligence) which affects the success of negotiations.By using the paradigm of the social construction of technology (Kwok & Koh, 2021), humans are parties who "negotiate" with technology to better recognize the technology, including deepfakes, and can adapt it to not become victims of technology-or misappropriate technology for evil interests-but rather agents who utilize technology to improve humanity and prevent harm posed by technology (such as deepfakes).An individual's degree of cooperation, tolerance, flexibility, and patience is reflected in the Agreeableness dimension.More agreeable people are at a larger risk for security, and social engineers (like deepfake designers) specifically target Agreeableness attributes like benevolence and compliance.
Conscientiousness reflects precisions, cautiousness, and a degree of self-control.Individuals with higher Conscientiousness thread might have higher self-efficacy in recognizing deepfakes.This is in line with the hypothesis of Köbis et al. (2021) that increasing Conscientiousness will make people motivated to invest cognitive resources to detect deepfakes, thereby enhancing their capacity to recognize truth and decreasing their desire to spread false information.

REVISED Amendments from Version 2
The authors add the directions of the hypotheses, conclusion, and implications of this present study.
Any further responses from the reviewers can be found at the end of the article Openness reflects the willingness to experience new things and is associated with lower risk aversion.Research by Uebelacker and Quiel (2014) shows that open people don't create suitable coping mechanisms because they misjudge their vulnerability to being a target of social engineering (like deepfake designers).This confirmatory study tested the hypotheses that the dimensions of HEXACO personality traits, i.e. (1) Honestyhumility (H), ( 2) Emotionality (E), (3) Extraversion (X), (4) Agreeableness (A), ( 5) Conscientiousness (C), and ( 6) Openness (O) can predict self-reported efficacy in recognizing deepfakes.H, A, and O traits would predict the efficacy in negative directions, while E, X, and C would predict it in positive directions.
The entire HEXACO model offers a thorough picture of a person's personality traits.The dynamics of the six qualities may provide a multidimensional perspective on one's cognitive, emotional, and behavioral responses to deceptive information when taking into account its ability to predict self-reported efficacy in spotting deep fakes.For instance, people with higher levels of Conscientiousness may investigate media with great care, whereas people with higher levels of Agreeableness may be less sceptic and more trusting.Based on the predictive powers of the six qualities, HEXACO may be able to shed light on a person's overall sense of vulnerability or resilience to digital deception.Written informed consent was obtained from all participants of this study, which included consent for the research procedure to be carried out and for the publication of this article containing anonymized, analyzed, and interpreted data.

Ethical considerations
Participants filled out an electronic survey questionnaire consisting of demographic data and two scales, namely HEXACO Personality Traits (as the predictors) and Self-efficacy in recognizing deepfake (as the criterion variable).The design of this study was predictive correlation.There is only one data collection stage.There is no exposure in this study because the research was not an experimental study.
The eligibility criteria of the samples were young adults aged 18-25 years (Generation Z), which, according to a YouGov survey, is an age group who are concerned about a deepfake video of themselves going viral online (Help Net Security, 2022; unpublished report).In addition, Generation Z account for more than a quarter, precisely 26.47% of the total Indonesia's population (Badan Pusat Statistik, 2020a, 2020b).This group were the less likely to risk falling victim to misinformation like deepfakes compared to the older generation (Caramancion, 2021).The 18 to 24 age group was the most confident one in detecting deepfakes (iProov, 2020).Thus, understanding the self-efficacy of this age group in relation to their individual differences provides a huge potential for deepfakes detection strategies.
The participants of this study were 200 young adults (139 women, 61 men; M=22.06 years old; SD=1.98 year) who came from a non-Western country, Indonesia, and were recruited using a convenience sampling technique.The number of sample came from a calculation using the Sample Size Calculator (Calculator.net,2022), with the following parameters: Confidence level of 95%, population size of 71,509,082 and population proportion of 26.47% -which was the total population of generation Z in Indonesia, as well as a margin of error of 6.2% -which is still in the range of 3-7%, the acceptable one (National Institutes of Health, 2005; unpublished report).
The research was conducted for 6 months from planning, participant recruitment, to data analysis.The research location is in Indonesia in an online setting for 3 months, namely 1 May to 31 July 2022.The research was a cross-sectional study, so no follow-up procedure was applied.
To measure self-efficacy in recognizing deepfakes, the authors constructed a self-efficacy measuring tool based on Bandura's theory (1977) which was adapted with the recommended checklists to pay attention when detecting deepfakes from the cyber-security company Norton taken from its unpublished report (Johansen, 2020).The introductory question was: "How sure are you that you can recognize or detect the presence of non-original or unnatural or unnatural elements (e.g. because it has been EDITED/MANIPULATED) from every image, photo, sound, and video you encounter?"Examples of items were: (1) I feel able to see abnormal eye movements; (2) I feel that I recognize awkward faces, e.g. if someone's face is pointing in one direction and the nose is pointing the other way; (3) I feel able to see any inappropriate skin tone in a video; (4) I am confident of being able to recognize when a person's face does not seem to convey the emotion that should be in line with what the person is supposed to say.There were six answer choices, ranging from "Feeling Very Incompetent" (scored 1) to "Feeling Very Capable" (scored 6).
To measure personality traits, this study used the short version of HEXACO-PI-R (60 items) (Lee & Ashton, 2009) with a scoring key.The response option ranged from "Strongly Disagree" (scored 1) to "Strongly Agree" (scored 6).The author translated the measuring tool into Indonesian.
All psychological scales in the questionnaire were tested for validity and reliability with the criteria of item validity (corrected item-total correlation) of at least 0.250 and internal consistency (Cronbach's α) of at least 0.600.A number of HEXACO trait items were eliminated because they did not meet these criteria.The test results are listed in Table 1.
Multiple regression, when examining the hypothesized relationship between HEXACO traits and self-reported efficacy in identifying deep fakes, offers several advantages over simple correlational analysis.It enables the simultaneous examination of all traits, determining each trait's unique contribution while controlling for others.This method also filters out shared variance, detects multicollinearity, and allows for predictive model development.Additionally, multiple regression can discern the relative importance of each predictor, and overall, provides a more nuanced and precise assessment of how HEXACO traits jointly influence the self-efficacy to detect deep fakes.

Results
Demographically, some participants were residents of DKI Jakarta province (N=90) which is the capital of Indonesia.In addition, other participants were residents of the Java Island (non-DKI Jakarta; N=86); Sumatera Island (N=21); and the rest (N=3) came from East Kalimantan, North Maluku, and West Nusa Tenggara provinces.
The psychometric properties and descriptive statistics of the variables are shown in Table 1.The results of this study indicate that the residuals of the multiple regression model are normally distributed (Figure 1) and all HEXACO personality dimensions are negatively correlated with self-efficacy in recognizing deepfakes; except for Agreeableness, which positively correlated (see Table 2).However, the results of the regression analysis with F(6,199)=13,295, p=0.000,R 2 =0.292, showed that only Honesty-humility and Agreeableness were able to predict the efficacy (see Table 3).
Table 3 shows the unadjusted (B) and adjusted (β) estimates for each predictor of which the potential confounders are the personality traits dimensions other than the focused predictor.

Discussion
Recognizing all deep fakes elements requires a certain level analytical capability and general intelligence (Ahmed, 2021).We need to look not just at people's cognitive abilities, but also at their belief in carrying out these abilities to recognize the information of deep fakes contextually.In other words, their self-efficacy.
This study found that, to a certain degree, individuals' personality traits do affect their self-efficacy in terms of detecting deepfakes.Because self-efficacy expression depends on context-to-context, it is not surprising that some traits can predict it better than the others.
Personality trait of Honesty-humility had negative predictive correlation with self-efficacy in recognizing deepfakes, β=-0.255,t(193)=-2.491,p<0.05 (Table 3)."Persons with very high scores on the Honesty-Humility scale avoid manipulating others for personal gain, feel little temptation to break rules, are uninterested in lavish wealth and luxuries, and feel no special entitlement to elevated social status" (Lee & Ashton, 2009, para 1).A person's Honesty-humility trait do not want to engineer others but, ironically, this trait makes them vulnerable to being manipulated by others (Ternovski et al., 2021), including deepfakes.It can drive higher errors for the trait in recognizing deepfakes, exposing weaknesses that could be exploited.
Thompson et al. (2016, p. 54) once stated, "Honesty-Humility may not only be less likely to exploit others, they may also be strongly opposed to being the target of exploitation."However, this study found that when faced with deepfakes, Generation Z individuals high in Honesty-Humility feel overwhelmed and struggle to identify these manipulations.
That is a notable discovery of this present study, and could be explained by the findings of Weger et al. (2022) that Honesty-Humility has a negative correlation with general (r=-0.168,p=0.002) and specific (r=-.0270,p<0.001) technology acceptance.This is reinforced by the findings of Sindermann et al. (2020) that Honesty-Humility has a negative correlation with all aspects of technology acceptance, namely perceived usefulness (r=-0.25,p<0.001), perceived ease of use (r=-0.16,p<0.001), intention to use (r=-0.17,p<0.001), and predicted usage (r=-0.18,p<0.001).In fact, someone with high technology affinity is able to perceive deepfakes less negatively (Kleine, 2022).This is presumably because they feel they have knowledge and "master" deepfakes.
Therefore, to not fall for deepfakes, Generation Z with a high Honesty-Humility trait need to reduce their conservative attitude towards technology in order to detect potential harm and even utilize deepfakes effectively.Future studies can test this with an experimental design that involves measuring these two traits and people's ability to detect malicious vs. nonmalicious deepfakes videos.
Not as hypothesized, Emotionality cannot predict self-efficacy in recognizing deepfakes, β=-0.029,t(193)=0.332,p>0.05 (Table 3).Austin and Vahle (2016) found that Emotionality-a trait that is positively correlated with empathy and social engagement-can predict the dimensions of Enhance (providing support and reassurance as interpersonal emotion management strategies) and Divert (the practice of using humor and pleasure pursuits to lift the spirits of others) of the Managing the Emotion of Others Scale (MEOS).This means that the Emotionality dimension is also positively correlated with the emotional intelligence needed to recognize deepfakes.Yang et al. (2022) emphasized the pivotal role of emotional intelligence in improving artificial intelligence technology so that it becomes a useful deepfake in the context of clinical encounters.By knowing that deepfakes themselves are increasingly being prepared with elements of emotional intelligence, then recognizing deepfakes also requires a better one; and this intelligence can actually be found in people with higher Emotionality.However, individuals high in Emotionality might be less confident in their own ability to accurately recognize deepfakes, as they might consider more factors and doubt themselves more (Thompson, 1998).With this uncertain direction, it is not surprising that no predictive power of Emotionality was found on self-efficacy.
Extraversion is a personality trait that cannot predict self-efficacy in recognizing deepfakes, β=0.003, t(193)=0.044,p>0.05 (Table 3).Hosler et al. (2021) put forward that detecting deepfakes is actually recognizing unnatural displays of emotion in voices and faces.Emotion apparently plays a central role in recognizing deepfakes because emotion is a higher-level semantic construct-which is difficult to counterfeit up to now-that could offer hints for detection.In an unpublished report, Kill states that emotion recognition is an ability that is honed in someone with a high extraversion trait (2021).However, Extraversion is also found to be positively correlated with excitement-seeking and a lower preference for consistency (Uebelacker & Quiel, 2014) -whereas "pairwise self-consistency learning" (Zhao et al., 2021, p. 15023) is needed to recognize deepfakes.Therefore, the positive and negative predictive powers of Extraversion on deepfake detection seem to neutralize each other.
Agreeableness trait can predict self-efficacy in recognizing deepfakes; however, not as hypothesized, the direction was found positivenot negative, β=0.309, t(193)=4.090,p<0.05 (Table 3).People with high Agreeableness are eager to cooperate and reach a compromise with others (Lee & Ashton, 2009).One of the good "others" in the context of deepfake recognition or detection is the "wisdom of the crowds" (Groh et al., 2022), which Surowiecki ( 2004) defines as "the collective intelligence that arises when our imperfect judgments are aggregated".Agreeing with (or high Agreeableness to) the collective intelligence should reduce the chance of falsely recognizing deepfakes, including its algorithm attempts that present visual obstructions such as misalignment, partial occlusion, and inversion.
Agreeableness trait has a positive correlation with perception of forensic science (Sarki & Mat Saat, 2020).Deepfakes detection can be seen as part of forensic science.People with high agreeableness are known for their cooperativeness; agreeableness is often referred to as safeguards against antisocial behavior (Frias Armenta & Corral-Frías, 2021), including -in the context of this present study -deepfakes creation and distribution.They esteem innovative forensic methods in their environment and have a positive attitude toward it for the common good (Sarki & Mat Saat, 2020).
People who are more agreeable tend to make more accurate decisions about whether to believe information, which reduces their vulnerability to victimization (Cho et al., 2016).This is confirmed by the empirical findings of van Winsen (2020) that agreeable individuals exhibit more secure online behavior and have a lesser likelihood of becoming a victim of cybercrime.
This study found that Conscientiousness was not able to predict self-efficacy in recognizing deepfakes, β=-0.079,t(193)=-0.798,p>0.05 (Table 3).Although deepfake recognition requires conscientious characteristics such as prudence and a sense of responsibility, Lawson and Kakkar's (as cited in Sütterlin et al., 2022) research recently found that Conscientiousness is partially correlated with belief in conspiracy and conservatism -making it less efficacious in recognizing deepfakes.
This study found that Openness was not able to predict self-efficacy in recognizing deepfakes β=-0.028,t(193)=-0.312,p>0.05 (Table 3).In an unpublished report, Jin (2020) found that values of Openness to change do not correlate with the perceived ethical implications of deepfakes (e.g., "These videos can uncontrollably deceive and influence many people", p. 24).In addition, contrast with the certain direction of the influence of Agreeableness and Honesty-humility on the selfefficacy; the direction of the Openness prediction is ambiguous.On the one hand, Openness is related to the low ability to recognize deepfakes.It is because Openness was found to be positively correlated with cognitive ability (Curtis et al., 2015;Rammstedt et al., 2016), but cognitive abilities encourage more protective online behavior, indicated by more interest in discussing how people who use deepfakes manipulate their audiences -rather than developing ability to apply scepticism on the authenticity of videos (Ahmed, 2021).On the other hand, there is a logic in favor of Openness as a buffer to prevent vulnerabilities from being manipulated by social engineering.For example, Eftimie et al. (2022) associated Openness with cognitive exploration tendencies which, based on their study, will stimulate responsible behavior including security best practices -which in the context of this study is deepfake recognition.
Based on the study findings, to avoid falling for deep fakes, there are two "optimal" personality traits that are worth exercising, i.e.Honesty-Humility and Agreeableness.First, the Honesty-Humility trait needs to be positioned strategically so that people with this trait can not easily be trapped or "absorbed" by the counterfeits from deepfakes technology, ie by reducing conventionalism (Leone et al., 2012) towards technology, that is allegedly inherent in this trait.Second, agreeableness trait should be directed at various deepfake detection methods and technologies that are beneficial to community members.
A number of studies have shown that both general and technological self-efficacy are able to predict the actual ability associated with the use of the technology (Alnoor et al., 2020;Raghuram et al., 2003;Tetri & Juujärvi, 2022).This is because the efficacies determine organizing actions, behavioral intention and strategies, and preparedness for change, as well as reducing emotional sensitivity which is a source of performance anxiety.
Of course, there is no denying the possibility of inflated or overestimated belief, or the Dunning-Kruger effect (Koc et al., 2022), which in the context of this study means that people who have high self-efficacy in detecting deepfakes actually have low actual abilities.In their research on bullshit detection, Cavojová et al. (2022) explained that the overestimation is caused by metacognitive (un) awareness, i.e. "These highly overconfident people suffer from a double cursenot only they do not know, but they also do not know that they do not know ... [that] is the result of self-enhancement motivation" (p. 1, 2).
To conclude, an individual's self-efficacy in spotting deepfakes is significantly predicted by personality traits, especially Honesty-Humility and Agreeableness.Higher Agreeableness encourages greater recognition, however higher Honesty-Humility may make one more vulnerable to such manipulations because of its negative link with technology acceptance.Furthermore, other traits, such as extraversion, conscientiousness, and openness, did not significantly predict selfefficacy in spotting deepfakes.Contradictory factors neutralized Extraversion's impacts, Conscientiousness' anticipated advantages were eclipsed by beliefs linked with the trait, and Openness, despite its association with cognitive capacity, had no direct effect on the effectiveness of deepfake recognition.These results imply that understanding these interactions between personality traits, technology acceptance, and self-efficacy is essential for developing successful defenses against deepfakes.
The limitation of this research is the use of non-probability sampling with limited generalizability.Nevertheless, this study has implication for the development of psychoinformatics -a branch of psychology that explains attitudes, competencies, and behavior in using information technology.Further research is suggested to implement random sampling and experimental methods to ensure a causal-not only predictive-relationship between personality traits and deepfakes detection self-efficacy.dependent was the self-efficacy in recognizing deepfakes, where the authors constructed a selfefficacy measuring tool based on Bandura's theory.And this was more adopted using the Notion company criteria.
Overall, authors have significantly improved the article, in terms of the clarity in explaining the methods and evaluations.
I would recommend not to explain the methods first with "There is only one data collection stage.
There is no exposure in this study because the research was not an experimental study."Although as a reviewer I know you may be addressing a comment, but if this is published, the readers have no clue on it.Explain the method first and then to be clear its a survey questionnaire .
I would also recommend to add a conclusion part to the manuscript summarizing the key takes aways -why some traits were removed and what implications does it reflect to the predicted efficacy.Also add the overall implication of such predictors to the efficacy items.
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Computational social science, deepfakes social implications I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

Sandra Grinschgl
Institute of Psychology, University of Graz, Graz, Austria This is a revised version of a manuscript that I previously reviewed.The authors addressed my previously raised comments and improved their manuscript.
Here are a few comments that the authors might still want to consider: The authors now included a brief paragraph on their hypotheses, however, to me those seem rather unspecific.In what direction would each personality trait predict self-reported efficacy in identifying deep fakes?How could the traits together predict variance in selfreported efficacy in identifying deep fakes?A related question: Why did the authors not have hypotheses for the correlational analyses? 1.
Some of the interpretations in the discussion seem a bit too extreme and could be downscaled.For instance: "Generation Z with trait Honesty-Humility feels helpless, so it is less functional in detecting deepfakes."I don't think the authors can draw this conclusion as 2.
they did not measure the actual ability to detect deepfakes.
"Therefore, the effects of Extraversion traits appear to cancel out of each other resulting in no predictive correlation with the self-efficacy" -> this should be formulated softer as it's only a suggestion.
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Human-technology interaction, personality psychology, cognitive psychology

Dilrukshi Gamage
Department of Innovation Science, Tokyo Institute of Technology, Tokyo, Japan Overall, the objective of this brief report is to find out the personality traits that affect the efficacy of spotting deepfakes.

Is the work clearly and accurately presented and does it cite the current literature?
○ Since some of the literature cited are unpublished reports, I am not sure of the credibility.Since those were absorbed to the main study, this is pretty questionable.

Is the study design appropriate and does the work have academic merit?
○ As mentioned in the introduction, the authors have taken the personality traits HEXACO from an unpublished report in 2009.I am curious what the authors consider other personality traits and also why did not cite any peer reviewed article with verified and validated factors.On the other hand, introduction does not provide a smooth understanding to why such model was selected and how others have conducted such explanations.

Are sufficient details of methods and analysis provided to allow replication by others?
○ Methods were illustrated in a very awkward pattern, for example at once I was not sure doing such a sample size for what-the method of data collection, it took a while to understand that this is a questionnaire, and authors' explanation of other statements made this bit complicated than a straight forward mention.
The survey questionnaire was unclear -as I understand it's a 6 point Likert scale, but I am not https://doi.org/10.5256/f1000research.141554.r160499 © 2023 Grinschgl S. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Sandra Grinschgl
Institute of Psychology, University of Graz, Graz, Austria Summary: This article deals with a highly relevant topic -the identification of deepfakes.The authors investigated potential predictors of the self-reported ability to detect deepfakes, namely individuals' personality traits based on the HEXACO model.While the traits honesty-humility and agreeableness were indeed shown to be predictors for self-reported ability to identify deepfakes, emotionality, extraversion, conscientiousness and openness were not.A higher honesty-humility was related to a lower self-reported ability to detect deepfakes whereas a higher agreeableness was related to a higher self-reported ability to detect deepfakes.
I think this is an interesting article on a very acute topic, however, I think it would benefit from some revisions.I outline my concerns below.
Major comments: Regarding the clear and accurate presentation of the authors' work: In my opinion, the introduction is lacking an elaborate justification for testing the HEXACO traits as potential predictors of the (self-reported) ability to detect deepfakes.Why might especially those traits act as predictors?The authors could, for instance, argue that typical personality traits might also be relevant when it comes to the application of other artificial intelligence technologies (e.g., robots).For a reference on this behalf see Matthews et al. (2021) 1 .

○
The introduction does not end with clear hypotheses, thus, to me it is unclear whether this research was exploratory or confirmatory.
○ Table 1: Why was the number of items reduced for the HEXACO traits?What does "after validation" refer do?This choice of method might potentially influence the interpretation of results.
○ I think the discussion would benefit from a paragraph that focuses on the bigger picture in which the results can be placed.For instance, it could be discussed what conclusions we can draw from the study findings when it comes to the increasing distribution of deepfakes.
What might be the "optimal" personality to not fall for deepfakes?Also, I think the authors should discuss how self-reported abilities in detecting deepfakes might be related to actual abilities in detecting deepfakes.This seems especially relevant as individuals' self-estimation is not always accurate (e.g., above-average effect, Dunning-Kruger effect).

○
Minor comments: Page 3: "The current consensus is that the average human's ability in recognizing deepfakes is similar to the machines (Vitak, 2022)."I'm not sure what this exactly means.Does it mean that artificial intelligence technologies analyzing videos/pictures are as good as humans in identifying deepfakes from real videos?○ Only within the methods section it became fully clear to me that the authors did not actually measure participants' ability to detect deepfakes but rather participants' self-reported ability ○ to do so.I would recommend the authors to refer to "self-reported efficacy in recognizing deepfakes" already from the beginning of their article (and in the abstract) to avoid any confusion.Figure 1: I guess this graph shows the residuals for the multiple regression model.Please clearly state this in the figure description and corresponding manuscript text.

○
Please report an effect size with the t-Test statistics on gender differences.
○ I think especially the discussion could need some language-editing.Every paragraph starts with "This study found" which does not induce an optimal reading flow.
The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias • You can publish traditional articles, null/negative results, case reports, data notes and more • The peer review process is transparent and collaborative • Your article is indexed in PubMed after passing peer review • Dedicated customer support at every stage • For pre-submission enquiries, contact research@f1000.com Note.M = mean, SD = standard deviation, SE = standard error.

Reviewer Report 17
July 2023 https://doi.org/10.5256/f1000research.152590.r185695© 2023 Grinschgl S. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This present study was initially approved by the Bina Nusantara University Research Committee, vide Letter of Approval No. 042/VR.RTT/VI/2021, strengthened with Letter No. 127/VR.RTT/VI/2022.The ethical decree is stated in Article 1 Paragraph 2 of the Letter.

confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above. Version 1
https://doi.org/10.5256/f1000research.141554.r162152© 2023 Gamage D. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.