CoChat
Sign Up
Cited by 84Misinformation and Its ImpactsSpam and Phishing DetectionLinguistics and Discourse Analysis

The fingerprints of misinformation: how deceptive content differs from reliable sources in terms of cognitive effort and appeal to emotions

Open access
Carrasco-Farré, Carlos·2022·Humanities and Social Sciences Communications
DOI 10.1057/s41599-022-01174-9
Open this paper in CoChat — free accountOpen PDF

Abstract

Not all misinformation is created equal. It can adopt many different forms like conspiracy theories, fake news, junk science, or rumors among others. However, most of the existing research does not account for these differences. This paper explores the characteristics of misinformation content compared to factual news—the “fingerprints of misinformation”—using 92,112 news articles classified into several categories: clickbait, conspiracy theories, fake news, hate speech, junk science, and rumors. These misinformation categories are compared with factual news measuring the cognitive effort needed to process the content (grammar and lexical complexity) and its emotional evocation (sentiment analysis and appeal to morality). The results show that misinformation, on average, is easier to process in terms of cognitive effort (3% easier to read and 15% less lexically diverse) and more emotional (10 times more relying on negative sentiment and 37% more appealing to morality). This paper is a call for more fine-grained research since these results indicate that we should not treat all misinformation equally since there are significant differences among misinformation categories that are not considered in previous studies.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper compares 92,112 news articles across six misinformation categories and factual news, finding that misinformation tends to require less cognitive effort and rely more on emotion and morality, but categories differ substantially. It argues for fine-grained, explainable content-feature approaches rather than treating all misinformation alike.

Key findings

  • Across misinformation on average, the abstract reports it is 3% easier to read, 15% less lexically diverse, 10 times more reliant on negative sentiment, and 37% more appealing to morality than factual news.
  • Category averages differ: fake news has the lowest readability (FK = 11.25) and high morality (4.66); junk science has the highest readability (FK = 13.94); factual news is near-neutral in sentiment (−0.19) and lowest in morality (3.12); hate speech is most negative (−6.01); rumors have the lowest perplexity (140.56), while factual news has the highest (174.23).
  • Clustering shows two main groups—rumors, hate speech, conspiracy theories, and fake news versus factual news, clickbait, and junk science—with factual news separating first at height 16.86 and conspiracy theories and fake news most similar (height 1.84). Euclidean distances from factual news are 12.00 for junk science, 21.71 for clickbait, 34.33 for conspiracy theories, 34.84 for hate speech, 35.69 for fake news, and 37.68 for rumors.
  • Multinomial logit results show higher readability lowers the odds of clickbait, conspiracy theories, fake news, and rumors but raises the odds of hate speech and junk science; higher perplexity lowers the odds of misinformation; negative sentiment is associated with clickbait, conspiracy theories, fake news, and hate speech, while positive sentiment is associated with junk science and rumors; higher morality is associated with all misinformation categories. Simulations show raising readability from 6.78 to 22 decreases fake news probability by 35% and increases junk science by 40.19%; raising perplexity from 91.37 to 215.01 increases factual news by 8.72% and junk science by 33.75%, while decreasing rumors by 18.27%.

Methods

  • Data: the Fake News Corpus (9.4 million news items from 194 webpages) with OpenSources categories; the paper states it extracted 30,000 random articles from each category, generating 210,000 misinformation articles, and separately used Factiva for 3,177 factual articles from The New York Times, The Wall Street Journal, and The Guardian. After filtering articles under 200 or over 2,000 words and removing outliers (0.025 and 0.975 quantiles), the final dataset is 92,112 articles: clickbait 12,955; conspiracy theories 15,493; fake news 16,158; hate speech 15,353; junk science 16,252; factual news 1,743; rumors 14,158. Median length is 461 words, range 201–1,961, totaling 58,087,516 words across 197 websites.
  • Measures: Flesch-Kincaid readability for grammatical cognitive effort; perplexity for lexical cognitive effort; AFINN sentiment lexicon (2,477 coded words, −5 to +5) for sentiment; and a 411-word morality dictionary measured as moral words per 500 words, interacted with negativity.
  • Analysis: Euclidean distance, hierarchical clustering using unweighted pair group method with arithmetic mean, k-means clustering, and multinomial logistic regression with factual news as the reference category; predicted probabilities and first differences were based on 100 simulations.

Limitations

  • Paper-stated: the sample comes from a selection of websites, so results may reflect domain-specific features; more complex measures and models may be more useful than the parsimonious explainable model; results are category-level averages and sources within the same category may behave differently; findings depend on the specific dictionaries used.
  • Evident from the text: the factual news sample (1,743 articles) is much smaller than each misinformation category (12,955–16,252 articles), and labels are assigned at the source/website level via OpenSources rather than article-by-article, which may limit generalizability and category purity.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1057/s41599-022-01174-9

Cites

111 of 111 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 48 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Higher readability is associated with higher log odds of hate speech and junk science.

    “a one-unit increase in the Flesch-Kincaid score is associated with an increase in the log odds of hate speech (β readability; hate ¼ 0:01, p < 0.1) and junk science categories (β readability; junk science ¼ 0:04, p < 0.001).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.
  • Findingown data

    Junk science and factual content are the most similar categories.

    “junk science and factual content are the most similar categories”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The database's category labels (clickbait, conspiracy theories, fake news, hate content, junk science, rumors, factual information) accurately reflect the type of content.
  • Findingown data

    Negative sentiment is associated with higher log odds of clickbait, conspiracy, fake news, hate speech.

    “a one-unit increase in negative sentiment is associated with an increase in the log odds of a content being clickbait β sentiment; clickbait ¼ À0:01, p < 0.001), conspiracy theory (β sentiment; conspiracy ¼ À0:02, p < 0.001), fake news (β sentiment;fake news ¼ À0:02, p < 0.001) or hate speech (β sentiment;hate ¼ À0:03, p < 0.001)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.
  • Findingown data

    Junk science and rumors are significantly more positive than reliable news.

    “Both junk science and rumors tend to be significantly more positive than reliable news.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.
  • Findingown data

    Hierarchical clustering at height 24.90 yields two clusters: (rumors, hate speech, conspiracy theories, fake news) and (factual content, clickbait, junk science).

    “there are two big clusters (height = 24.90):rumors, hate speech, conspiracy theories and fake news on one side and factual content, clickbait, and junk science on the other.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The database's category labels (clickbait, conspiracy theories, fake news, hate content, junk science, rumors, factual information) accurately reflect the type of content.
  • May combine two claims. 23
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 27

  • May combine two claims.

    Findingown data

    Fake news and conspiracy theories are 18% more reliant on negative emotions and 45% more appealing to morality than factual news.

    “and 18% more reliant on negative emotions and 45% more appealing to morality than factual news.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 92,112 articles in the final dataset are representative of misinformation and factual news content generally.

  • May combine two claims.

    Findingown data

    Misinformation, on average, is easier to process in terms of cognitive effort than factual news: 3% easier to read and 15% less lexically diverse.

    “The results show that misinformation, on average, is easier to process in terms of cognitive effort (3% easier to read and 15% less lexically diverse)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 92,112 articles in the final dataset are representative of misinformation and factual news content generally.

  • May combine two claims.

    Findingown data

    Fake news and conspiracy theories are, on average, 8% simpler in terms of readability and 18% simpler in terms of perplexity than factual news.

    “Among others, these results show that fake news and conspiracy theories are, on average, 8% simpler in terms of readability and 18% simpler in terms of perplexity”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 92,112 articles in the final dataset are representative of misinformation and factual news content generally.

  • May combine two claims.

    Findingown data

    Misinformation, on average, is more emotional than factual news: 10 times more relying on negative sentiment and 37% more appealing to morality.

    “and more emotional (10 times more relying on negative sentiment and 37% more appealing to morality).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 92,112 articles in the final dataset are representative of misinformation and factual news content generally.

  • May combine two claims.

    Findingown data

    Hate speech has the most negative sentiment score (-6.01), followed by fake news (-3.51), conspiracy theories (-3.41), and clickbait (-3.09).

    “hate has the highest negative value (sentiment = -6.01,[CI = -6.21,-5.82]), followed by fake news (sentiment = -3.51,[CI = -3.70,-3.32]), conspiracy theories (sentiment = -3.41,[CI = -3.59,-3.22]) and clickbait (sentiment = -3.09,[CI = -3.31,-2.87]).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Lower morality appeal increases the probability of being classified as factual news, rumors, hate speech, or junk science; higher morality appeal increases the probability of being classified as fake news, clickbait, or conspiracy theories.

    “Morality plays a role in identifying categories by augmenting their probabilities of being classified as factual news, rumors, hate speech or junk science as the appeal to morality decreases; and on the contrary, it leads to higher probabilities of being classified as fake news clickbait or conspiracy theories as the appeal to morality in a given text increases.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Fake news appeal to moral values 37% more than factual content; hate speech appeals 50% more to morality than factual sources.

    “fake news appeal to moral values 37% more than factual content (although hate speech appeals 50% more to morality than factual sources)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • The quote only partly states this claim.

    Findingown data

    Increasing readability from 6.78 to 22 increases the probability of being classified as junk science by 40.19% (CI 0.390-0.414).

    “an increase of 40.19% of being classified as junk science (CI = [0.390,0.414])”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.

  • May combine two claims.

    Findingown data

    Positive sentiment increases the probability of being classified as junk science or rumors; negative sentiment increases the probability of being classified as hate speech, fake news, conspiracy theories, or clickbait.

    “These results indicate that highly positive sentiments increase the probability of a content being classified as junk science or rumors, while an increase in the negative sentiment leads to an increased probability of being classified as hate speech, fake news, conspiracy theories or clickbait.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Euclidean distances from factual news to misinformation categories: rumors 37.68, fake news 35.69, hate 34.84, conspiracy 34.33, clickbait 21.71, junk science 12.00.

    “In Fig. 2A, B one can observe that the Euclidean distance between factual news and rumors is 37.68, with fake news is 35.69, with hate content is 34.84, with conspiracy theories is 34.33, with clickbait is 21.71 and with junk science is 12.00.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The database's category labels (clickbait, conspiracy theories, fake news, hate content, junk science, rumors, factual information) accurately reflect the type of content.

  • May combine two claims.

    Findingown data

    Factual information is 15% more lexically diverse than misinformation; rumors are 19% less diverse than factual news.

    “factual information is 15% more lexically diverse (with rumors being 19% less diverse than factual news)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingown data

    Higher readability is associated with lower log odds of being clickbait, conspiracy, fake news, or rumor.

    “a one-unit increase in the readability score is associated with a decrease in the log odds of categorizing content in the clickbait (β readability; clickbait ¼ À0:06, p < 0.001) conspiracy theories (β readability; conspiracy ¼ À0:05, p < 0.001), fake news (β readability; fake news ¼ À0:21, p < 0.001), and rumor (β readability; rumor ¼ À0:04, p < 0.001) categories.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.

  • May combine two claims.

    Findingown data

    Fake news are 18 times more negative than factual news; hate speech is 30 times more negative.

    “fake news are, on average, 18 times more negative than factual news (having hate speech the highest difference with 30 times more negativity than factual news)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Rumors and junk science have positive sentiment scores (0.86 and 2.52, respectively).

    “rumors (sentiment = 0.86, [CI = 0.65, 1.08]) and junk science (sentiment = 2.52, [CI = 2.31, 2.72])”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Fake news has the highest morality score (4.66), followed by conspiracy theories (4.44), hate speech (4.41), and clickbait (4.38).

    “clickbait news (morality = 4.38, [CI = 4.34, 4.43]), hate speech (morality = 4.41, [CI = 4.36, 4.46]), conspiracy theories (morality = 4.44, [CI = 4.39, 4.48]) and, finally, fake news (morality = 4.66, [CI = 4.61, 4.70]).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Findingown data

    Higher perplexity increases the probability of being classified as junk science, clickbait, or factual news; lower perplexity increases the probability of being classified as conspiracy theory, fake news, hate speech, or rumors.

    “These results indicate that increasing the perplexity of content increases the probabilities of a content being classified as junk science, clickbait or factual news while decreasing perplexity augments the probability of a piece of text being classified as conspiracy theory, fake news, hate speech or rumors.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Findingnone given

    The model is easily understandable in daily language.

    “Finally, the model that is easily understandable in daily language (the explainability challenge)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Findingreasoning

    The method takes less than one second for a 2000-word article, so real-time deployment is feasible.

    “The feasibility of real-time deployment should not be a big problem since this method uses manageable calculations in a reasonable amount of time (less than one second for a 2000 words article).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Findingreasoning

    The findings increase the evidence about structural differences among misinformation types.

    “Secondly, they increase the evidence about structural differences among misinformation types (the breadth challenge)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Findingown data

    Not all misinformation is created equal.

    “the findings presented in this paper are a compelling and impactful demonstration that not all misinformation is created equal.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Deceptive content can adopt different forms, including rumors, clickbait, junk science, fake news, and conspiracy theories.

    “Misinformation can appear as rumors, clickbait or junk science (trying to maximize visitors to a webpage or selling”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Scheufele & Krause, 2019 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Content that requires less cognitive effort to be processed is more engaging and viral.

    “content that requires less effort to be processed is more engaging and viral (Alhabash et al., 2019)”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Alhabash et al., 2019 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The Flesch-Kincaid index estimates the educational level needed to understand a given text.

    “This measure estimates the educational level that is needed to understand a given text.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Kincaid et al., 1975

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The AFINN lexicon is one of the most used lexicons for sentiment analysis.

    “I employ the AFINN lexicon developed by Finn Årup Nielsen (Nielsen, 2011) one of the most used lexicons for sentiment analysis (Akhtar et al., 2020;Chakraborty et al., 2020;Hee et al., 2018;Ragini et al., 2018)”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Nielsen, 2011

      Not checked in the free run.

    • cites Akhtar et al., 2020 source

      Not checked in the free run.

    • cites Chakraborty et al., 2020 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Content that appeals to what individuals think is moral (or not moral) is likely to be more viral, especially in conjunction with negative emotions that challenge their social identity.

    “content that appeals to what individuals think is moral (or not moral), is likely to be more viral (Brady et al., 2020), especially in conjunction with negative emotions that challenge their social identity (Brady et al., 2017;Horberget al., 2011)”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Brady et al., 2020 source

      Not checked in the free run.

    • cites Brady et al., 2017 source

      Not checked in the free run.

    • cites Horberget al., 2011

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Perplexity measures uncertainty in predicting word appearance.

    “I calculate the uncertainty in predicting each word appearance in every text through perplexity (Griffiths & Steyvers, 2004)”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Griffiths & Steyvers, 2004 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    More advanced and diverse language allows encoding more complex ideas, which generates higher cognitive load.

    “In general, more advanced and diverse language allows to encode more complex ideas (Ellis & Yuan, 2004), which generates a higher cognitive load (Swabey et al., 2016).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Ellis & Yuan, 2004 source

      Not checked in the free run.

    • cites Swabey et al., 2016 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Morality was measured using a previously validated dictionary.

    “To measure morality, I will use a previously validated dictionary (Graham et al., 2009).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Graham et al., 2009 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Reliance on emotion is associated with misinformation sharing or believing in conspiracy theories.

    “reliance on emotion is associated with misinformation sharing (Weeks, 2015) or believing in conspiracy theories (Garrett & Weeks, 2017;Martel et al., 2019)”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Weeks, 2015 source

      Not checked in the free run.

    • cites Garrett & Weeks, 2017 source

      Not checked in the free run.

    • cites Martel et al., 2019

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    This type of model has been rarely available.

    “This is important because this type of model have been rarely available (Miller et al., 2017).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Miller et al., 2017

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Misinformation requires less cognitive effort and is more reliant on emotions than factual information.

    “misinformation requires less cognitive effort for being processed and is more reliant on emotions than factual information: it is easier to read, and falsified content creators use a less diverse vocabulary”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Readability and perplexity are valid measures of cognitive effort.
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • May combine two claims.

    Interpretationown data

    Misinformation tends to rely on emotional language, with varying polarity across categories.

    “Therefore, misinformation tends to rely on emotional language.However, the polarity of these emotions varies across misinformation categories.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The study used the k-means algorithm to add robustness to the results.

    “To add robustness to the results, I also used the k-means algorithm”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The study follows a similar approach as previous studies to classify sources.

    “Here, it is important to note that I follow a similar approach as previous studies to classify sources (Broniatowski et al., 2022;Cinelli et al., 2020;Cox et al., 2020;Lazer et al., 2018;Singh et al., 2020).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Broniatowski et al., 2022 source

      Not checked in the free run.

    • cites Cinelli et al., 2020

      Not checked in the free run.

    • cites Cox et al., 2020 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The labeling of each website was done through crowdsourcing following six steps.

    “The labeling of each website was done through crowdsourcing following the instructions as follows (Zimdars, 2017):”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Zimdars, 2017

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationown data

    Existing research tends to overemphasize the role of sentiment while neglecting the prominent role of appeal to morality.

    “This finding is important because existing research tends to overemphasize the role of sentiment while neglecting the prominent role of appeal to morality.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The proposed method can be useful to identify misinformation before it becomes viral.

    “Thirdly, I propose a method that can be useful to identify misinformation before it becomes viral (the speed challenge)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    The increased evidence about structural differences allows focusing on or differentiating specific categories and creating tailored countermeasures for each one.

    “which allows to focus on or differentiate specific categories and create tailored countermeasures for each one.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Improving media literacy to process more complex ideas will likely reduce the cognitive load associated with factual information and therefore the prevalence of misinformation.

    “improving media literacy to process more complex ideas will likely reduce the cognitive load associated with factual information (Ellis & Yuan, 2004;Swabey et al., 2016) and, therefore, the prevalence of misinformation.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Ellis & Yuan, 2004 source

      Not checked in the free run.

    • cites Swabey et al., 2016 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The categorization of an URL can be done at the exact moment it is posted.

    “since the categorization of an URL can be done at the exact moment it is posted”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductreasoning

    Sources belonging to the same misinformation category may behave differently.

    “it is also worth mentioning that sources belonging to the same category may behave differently.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Only 17% of US adults have the skills to safely navigate the information ecosystem.

    “only 17% of US adults have the skills to safely navigate the information ecosystem (Horrigan, 2016).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Horrigan, 2016

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Study design and conductnone given

    There may be more complex measures and models that may be more useful to discern between misinformation content and factual sources.

    “I also acknowledge that-potentially-there may be more complex measures and models that may be more useful to discern between misinformation content and factual sources.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The results are contingent to the specificities of each dictionary employed.

    “Finally, although we employed widely used dictionaries in the literature, the results are also contingent to the specificities of each dictionary I employed.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Misinformation often escalates into violence.

    “misinformation often escalates into violence, like in the case of ethnic brutality in India (McLaughlin, 2018) or violence against doctors in Latin America (Taylor, 2020).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites McLaughlin, 2018

      Not checked in the free run.

    • cites Taylor, 2020

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The results could be based on domain-specific features because the news articles come from a selection of websites.

    “I am aware of the potential limitation that the results could be based on domain-specific features.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    The model's understandability facilitates its adoption by technical and non-technical audiences.

    “facilitating its adoption by technical and, specially, non-technical audiences.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning

    Some fake news sources may use a more neutral tone, while others may show a higher appeal to negative emotions.

    “For example, some fake news sources may use a more neutral tone, while others may show a higher appeal to negative emotions.”
    Quote in the paperNot a substantial claim
48 of 48

  • The paper relies on this without support.

    The 92,112 articles in the final dataset are representative of misinformation and factual news content generally.

    “This resulted in the final dataset consisting of 92,112 articles with the following distribution by type: clickbait (12,955 articles), conspiracy theories (15,493 articles), fake news (16,158 articles), hate speech (15,353 articles), junk science (16,252 articles), factual news (1743 articles) and rumors (14,158 articles).”

  • The paper relies on this without support.

    The seven content categories are mutually exclusive for the multinomial logistic regression.

    “I will employ a multinominal logistic regression, a technique suitable for mutually exclusive variables with multiple discrete outcomes.”

  • The paper relies on this without support.

    The paper's dataset includes more sources and more articles than any other.

    “In comparison to other datasets, the one used in this paper includes more sources and more articles than any other:”

  • The paper relies on this without support.

    Readability and perplexity are valid measures of cognitive effort.

    “For each article, I calculated measures of cognitive effort (readability and perplexity) and evocation to emotions (sentiment analysis and appeal to morality).”

  • The paper relies on this without support.

    Sentiment analysis and appeal to morality are valid measures of evocation to emotions.

    “For each article, I calculated measures of cognitive effort (readability and perplexity) and evocation to emotions (sentiment analysis and appeal to morality).”

  • The paper relies on this without support.

    The database's category labels (clickbait, conspiracy theories, fake news, hate content, junk science, rumors, factual information) accurately reflect the type of content.

    “I analyze 92,112 news articles classified into seven categories: clickbait (n = 12,955), conspiracy theories (n = 15,493), fake news (n = 16,158), hate content (n = 15,353), junk science (n = 16,252), rumors (n = 14,158), and factual information (n = 1743)”

  • The paper relies on this without support.

    Factual news articles in the dataset are reliable and objective.

    “In this case, I used factual news as the level of the outcome since I am interested how misinformation differs from this baseline using the following formula:”

  • The paper relies on this without support.

    Prior studies only focus on one or a few specific categories, usually fake news.

    “prior studies only focus on one or just a few specific categories, usually fake news.”
8 of 8

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?

  • Different examples of misinformation often make use of the same underlying tropes.

    “different examples of misinformation often make use of the same underlying tropes (18, 19).”

    From Psychological inoculation improves resilience against misinformation on social media (2022) · cites this paper as [18]

  • These tropes can be analyzed and used for inoculation without prior knowledge of specific misleading content, thus potentially providing broad resilience against social media or news content that draws on one or more of the techniques that someone has been inoculated against.

    “These tropes, which include manipulation techniques such as logical fallacies (20) and emotionally manipulative language (18), can be analyzed and used for inoculation without prior knowledge of specific misleading content, thus potentially providing broad resilience against social media or news content that draws on one or more of the techniques that someone has been inoculated against.”

    From Psychological inoculation improves resilience against misinformation on social media (2022) · cites this paper as [18]

  • The five manipulation techniques used in the videos are consistently identified as epistemologically dubious: emotionally manipulative rhetoric, incoherent or mutually exclusive arguments, false dichotomies or dilemmas, scapegoating individuals or groups, and ad hominem attacks.

    “These five techniques were taken from the broader literature on argumentation and manipulation strategies and are consistently identified as epistemologically dubious: (i) using emotionally manipulative rhetoric to evoke outrage, anger, or other strong emotions (18, 22), (ii) the use of incoherent or mutually exclusive arguments (23), (iii) presenting false dichotomies or dilemmas (24), (iv) scape”

    From Psychological inoculation improves resilience against misinformation on social media (2022) · cites this paper as [18]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • High-readability category probabilities sum to only 0.897

    Major

    The seven reported probabilities for a text with readability 22 sum to 0.897, rather than 1. They are presented as the probabilities of all seven mutually exclusive categories.

    Reported 1 · recomputed 0.897

  • Negative-sentiment difference is described as both 18% and 18-fold

    Major

    The introduction calls the fake-news/conspiracy-theory difference 18% greater reliance on negative emotions, while the discussion calls the fake-news difference 18 times greater than factual news. An 18-fold difference is not an 18% difference.

  • Six misinformation categories cannot yield 210,000 articles at 30,000 each

    Major

    The paper describes six misinformation categories. Sampling 30,000 articles from each would yield 180,000 misinformation articles, not 210,000; adding the stated 3,177 factual articles would yield 183,177 rather than the reported 213,177 total.

    Reported 210000 · recomputed 180000

  • Junk-science readability coefficient differs from Table 3

    Minor

    The results prose reports a junk-science readability coefficient of 0.04, whereas Table 3 reports 0.127. This is not a rounding difference.

  • Discussion gives two different readability differences for fake news

    Minor

    The same discussion sentence says fake news is both 3% and 13% easier to process than factual sources.

5 of 5

Cites

1 retracted

Flagged first, then by citations

Research works printed but not linked: 12

  1. Beheshti Z, Nejadansari D, Barati H (2020) The relationship between emotional intelligence, Lexical diversity and the syntactic complexity of EFL Learners’ written productions. J Mod Res Engl Lang Stud 7(1):133–161
  2. Berger JA, Milkman KL (2009) What makes online content viral? Ssrn 49(2):192–205
  3. Cinelli M, Quattrociocchi W, Galeazzi A, Valensise CM, Brugnoli E, Schmidt AL, Scala A (2020) The COVID-19 social media infodemic. Sci Rep 2020 10:1 10(1):1–10
  4. Eshet Y, Eshet Y (2004) Digital literacy: a conceptual framework for survival skills in the digital era. J Educ Multimedia Hypermedia 13(1):93–106
  5. Fowler JH, Christakis NA (2009) Dynamic spread of happiness in a large social network: longitudinal analysis over 20 years in the Framingham Heart Study. BMJ (Online) 338(7685):23–26
  6. Gamallo P, Campos JRP, Alegria I (2016) Comparing two basic methods for discriminating between similar languages and varieties*. Phys A: Stat Mech Appl 170–177. https://aclanthology.org/W16-4822/
  7. Goel S, Anderson A, Hofman J, Watts DJ (2015) The structural virality of online diffusion. Manag Sci 1909:150722112809007
  8. Gonzàlez M (2015) An analysis of Twitter corpora and the differences between formal and colloquial tweets. In Proceedings of the Tweet Translation Workshop 2015 (pp. 1−7)
  9. Nielsen FÅ (2011) A new ANEW: evaluation of a word list for sentiment analysis in microblogs. CEUR Workshop Proc 718:93–98
  10. Sun S, Nekmat E (2008) Learning through “prosuming”: insights from media literacy programmes in Asia. Technol Soc 13:259–278
  11. Tajfel H, Turner JC, Austin WG, Worchel, S (1979) An integrative theory of intergroup conflict. Organizational identity: A reader, 56(65), 9780203505984−16
  12. Wang WY (2017) “Liar, liar pants on fire”: a new benchmark dataset for fake news detection. ACL 2017-55th Annual Meeting of the Association for Computational Linguistics. Proc Conference (Long Papers) 2:422–426

Cited by

Sorted by citations

Related

Authors
Carrasco-Farré, C.
Affiliations
Esade (ES)
Corresponding author
Carrasco-Farré, Carlos
ORCID
Carrasco-Farré, Carlos
Venue
Humanities and Social Sciences Communications
Publisher
Springer Science and Business Media LLC
Published
2022-05-09
Volume
9
Issue
1
ISSN
  • 2055-1045 (ISSN-L)
  • 2662-9992
DOI
10.1057/s41599-022-01174-9
OpenAlex
W4229373744
Semantic Scholar
97a90b895fae626b75f51d309c4c751282956a71
Cited by
84
References
110
Citations by year
2026: 15 · 2025: 23 · 2024: 30 · 2023: 12 · 2022: 4
FWCI
24.82
Type
journal article
Language
en
Open access
Licence
Locations
  • Humanities and Social Sciences Communications (publishedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Ministerio de Ciencia e Innovación
  • European Social Fund
Topics
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
  • Physical Sciences › Computer Science › Information Systems › Spam and Phishing Detection
  • Social Sciences › Arts and Humanities › Philosophy › Linguistics and Discourse Analysis
Keywords
misinformation, cognitive effort, emotional appeal, clickbait, conspiracy theories, fake news, hate speech, junk science, rumors, lexical complexity, grammatical complexity, sentiment analysis

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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