Beyond compliance: How European fact checkers correct their own errors

---

Implications

Accountability in fact-checking implies responsibility for the quality and consequences of published content. This is typically pursued through normative frameworks (codes) established by EFCSN and IFCN (Chaparro Domínguez et al., 2019; McQuail, 2003). Accountability in fact-checking is especially important because fact checkers hold a unique knowledge-verification position in the contemporary information ecosystem and are seen in society as a potential means of restoring trust in journalism. For that reason, EFCSN and IFCN have created Code of Standards and Code of Principles, which include recommendations regarding transparency, impartiality, fairness, and correction policies (Moreno-Gil & Salgado-de Dios, 2022).

Even though normative frameworks exist, there is a lack of empirical evidence on how these mechanisms are implemented in practice (Moreno-Gil & Salgado-de Dios, 2022; Singer, 2020). The aim of this study was to methodologically expand this evidence by assessing whether these commitments are symbolic or substantive. To assess it, we examined corrections as the most concrete and observable indicator of accountability. Unlike ethical codes and other accountability tools, corrections reveal more about the actual practice of accountability than codes or organizational policies, which represent a list of intentions (McQuail & Deuze, 2020). However, publishing a correction does not necessarily achieve accountability, as the correction should communicate the values of accountability (Kampf & Daskal, 2014). Corrections should clearly identify the error, provide accurate information, explain how and why the error occurred, explain how similar errors can be avoided, and include an apology to those affected by the error (Bugeja, 2007).

Article 3.3 of the EFCSN Code of Standards regulates the correction of errors. Under this article, signatories are obliged to correct every known mistake, excluding typographical or grammatical errors, in the form of a note in the same publication (A). Corrections should be disseminated separately if the mistake cannot be edited in the same format and channel (B). If the error is substantial, the correction note should contain accurate information, as well as an explanation of the circumstances (C). Even the addition of new information should be acknowledged in an update note (D). The correction policy should be easily findable on the website, including a record of corrections (E). There should be a direct channel for requesting corrections (G), the rules for assessing requests should be explained in the policy (F), and there should be an explanation of how to appeal to EFCSN, including a complaint form link (H).¹The letters A–H are abbreviations taken from Article 3.3 of the EFCSN Code of Standards.

This study connects provisions from EFCSN’s Article 3.3 requirements with Kampf and Daskal’s (2014) conceptual model to evaluate how EFCSN signatories implement accountability and transparency by examining corrections (visibility, placement, timing) and their discursive features (explanatory detail, attribution of error, acknowledgment of affected actors), as well as Code obligations through correction practices. Kampf and Daskal (2014) proposed assessing the values of accountability in corrections through textual dimensions: (a) the corrective marker (location of the error and the correction, time between publishing the error and its correction), (b) the offender (identification of who was responsible for the error), (c) the offense (how the error occurred), and (d) the offended party (who was affected by the error).²The letters a–d are abbreviations taken from Kampf and Daskal (2014).

As shown in Table 1, some EFCSN members’ corrections do not fulfill their accountability role by acknowledging errors, explaining their causes, or ensuring transparency about why and how the error has occurred. These results suggest that current accountability frameworks rely too heavily on declarative compliance in the form of policies, registries, and notes, and that the practices are not systematically supervised by standard-setting bodies. As such, correction practices and fact-checking accountability in the case of EFCSN members are more dependent on organizational culture rather than shared operational norms.

Area	Key finding
Correction activity	67.7% published corrections; 30.6% published none
Correction type	78.5% minor updates or typographical/stylistic changes
Substantive corrections	18.7% changed verdict or methodological basis
Visibility	50.5% labeled at the top and bottom
Responsibility	66.2% did not identify responsible party
Explanation	37.5% did not explain the cause
Formal compliance	61 policies published; 58 requests for correction
Policy detail	54.8% had detailed correction policies
Correction/update distinction	79% did not clearly distinguish them

These findings have practical implications for fact-checking organizations and policy frameworks. They suggest that the EFCSN Code should set more detailed requirements for visibility, explanation, and attribution of responsibility, while preserving methodological diversity and editorial independence, and considering the diverse political, cultural, and linguistic contexts of fact-checking in Europe. Although fact checkers argue that increased visibility of corrections may expose organizations to reputational attacks, these concerns also contribute to perceptions of arbitrariness or lack of accountability. Increased transparency should be seen as a long-term investment in their legitimacy rather than a vulnerability. Additionally, correction transparency may be used to assess the accountability of fact checkers and incorporated into auditing procedures by EFCSN and IFCN.

Suggestions that arise from findings concern standardizations of correction categories to formally distinguish substantial from minor corrections. Codes could also include a mandatory format of correction notes to include basic information about corrections: date, type of error, responsible party, and a brief explanation of the error. Additionally, there should be provisions on the location of corrections registry on the fact-checking website and the placement of correction notes on the original fact-check. Adoption of these recommendations could improve the level of public trust in fact-checking (Farkas & Schou, 2023).

Findings

Finding 1: EFCSN signatories publish corrections irregularly and most corrections address minor errors, with substantial variation after accounting for output volume.

The analysis identified a total of 1,555 corrections of errors published by EFCSN members between November 2013 and October 2025. When limited to the period during which all organizations were active EFCSN members from January 1, 2024, to October 15, 2025, 67.7% of organizations (N = 42) published 490 corrections. More than a third, 30.6% of organizations (N = 19), did not publish any corrections. The normalized correction rates showed that most organizations had low relative frequencies, with median values below 0.01, and only a limited number showed rates higher than 0.05. Extremely high values were found among organizations with relatively small fact-check production. Pearson correlation (r = −.1486, p = .249) showed no significant relationship between fact-check output volume and normalized correction rate. Regression analysis (R² = .02) indicated that output volume explains approximately 2.2% of the variance in normalized correction rate.

       These results suggest that higher production of fact checks does not result necessarily in more corrections. Some organizations have had higher correction rate than expected, relative to their production volume. Others had lower rates than expected and these differences could be explained with institutional transparency practices, internal editorial workflows, resources, and journalistic culture.

       The impact of the political environment on the production of corrections was analyzed by comparing the total monthly volume of corrections from January to October 2024 with that from January to October 2025, focusing on the U.S. presidential elections in November 2024. A paired t-test showed no significant difference between the compared periods, t(9) = 1.9, p = .086.

       The findings showed that more than 57.1% (N = 171) of corrections addressed factual updates that do not alter the original assessment or conclusion (minor errors and updates), 21.4% (N = 64) of corrections were typographical or stylistic changes, and 18.7% corrections (N = 56) were substantial corrections that fundamentally changed the verdict or methodological basis of a fact check.

Finding 2: The communication of accountability values is limited and inconsistent among EFCSN members.

 As prescribed by the Code, all analyzed corrections were added to the original content. Most of the corrections 50.5% (N = 166) are labeled at the top and bottom of the page. Usually, this means the label “updated” appears at the top, with a correction note at the bottom. Clearly labelled corrections in multiple positions accounted for only 3% (N = 9) of those analyzed, 24.1% of corrections (N = 72) were labelled only at the bottom of the page, while 15.7 (N = 47) had a label only at the top of the page.

In 66.2% of analyzed corrections (N = 198), it was not stated who was responsible for the error, while journalists (fact checkers) were responsible in only 0.6% of corrections (N = 2). The most common responsible parties were organizations, identified in 28.8% of corrections (N = 86). Fact checks were corrected or updated within 24 hours in 36.8% of analyzed corrections. Fact checks updated after seven days or more ranged between several days and several months after the original publication and account for 26.1% of the analyzed cases (N = 78).

Correct and erroneous information were present in 46.2% (N = 138) of the analyzed corrections, while only correct information was stated in 19.7% (N = 59) of the corrections. The causes of the errors were methodological in 6.4% (N=19), factual in 30% (N = 90), contextual in 5% (N = 15), technical in 23.4% (N = 70) and difficult to determine in 19.1% (N = 57) of the corrections, and a mix of the previous in 15.5% (N = 45) of the analyzed corrections.

In 42.4% of corrections (N = 127), the cause of error was explained, while 37.5% of corrections (N = 112) did not include this information. The offended party was not mentioned in 64.5% of corrections (N = 193), while in 16.4% of corrections (N = 49) the offended organization was mentioned, and in 15.7% of corrections (N = 46), individuals were mentioned.

Finding 3: Fact-checking organizations generally demonstrate formal compliance with EFCSN Article 3.3, but substantive implementation varies.

Almost all organizations publish corrections policies (N = 61) and have public registries of corrections, and the majority offer an option to submit a request for correction (N = 58). Information on how to complain to EFCSN is available for 70.1% of organizations (N = 44), while 29% (N = 18) do not provide information on the right to appeal to EFCSN. Correction policies are detailed in 54.8% of organizations (N = 34), while the correction policies are partial in 33.9% of organizations (N = 21). A detailed correction policy is one that includes information about the types of corrections, contact for claim submission, and the process for validating submitted claims. Partial policies are those that lack one or more of these elements. In 79% of the analyzed corrections (N = 49), a clear distinction between corrections and updates was not evident.

The paths and names of the pages where corrections policies and registries are published are very diverse. Thus, corrections policies and registries are hidden under methodology, about us, and similar pages (N = 37), while they are found under Corrections or Corrections Policies in one third of cases (N = 20).

Methods

The analysis of EFCSN signatories’ error correction practices was based on the frequency of corrections listed on their websites (N = 62) and on how the accountability, in the form of Code obligations and values of accountability (Brautović & Bebić, 2022; Kampf &  Daskal, 2014), was communicated in corrections.

Organizations were analyzed regardless of language or publication volume to cover the entire European fact-checking ecosystem represented in EFCSN. No active signatories were excluded from the analysis. The units of analysis, in addition to websites, were only visible instances of correction policies (N = 61) and corrections (N = 299) during the period from January 1, 2024, to October 15, 2025. One signatory did not have a correction policy nor list of corrections visible or accessible on their website. Although 490 corrections were identified during the analyzed period, 299 were randomly selected for content analysis to ensure manageable coding volume while preserving representativeness.

To enable comparison between organizations with different production volumes, the total production volume of published fact checks was primarily collected from organizations’ websites and, if unavailable, from the BENEDMO Factchecking Database or alternative sources as mentioned in Appendix A. Normalized correction rates were calculated as the ratio between the number of published corrections and the total number of published fact checks (corrections/fact checks). To assess the association between the volume of fact-check output, Pearson’s correlation was used, while to determine whether total production volume predicts normalized correction frequencies, linear regression analysis was performed using log-transformed output values.  

The temporal analysis of correction volume was conducted to assess the potential influence of political events on fact-checking practices. Data from January to October for 2024 and 2025 were compared by using parametric (paired t-test) and non-parametric (Wilcoxon signed rank) tests with significance level at p < .05. Pearson correlation, linear regression, paired t-test, and Wilcoxon signed-rank test were calculated using online calculators available at Social Science Statistics.³See https://www.socscistatistics.com/

Two coding schemes were deductively developed to analyze 1) compliance with the EFCSN Code of Standards (drawing on the Article 3.3 Honest Corrections) and 2) the communicative dimensions of corrections, based on the model of Kampf and Daskal (2014). The initial set of categories and coding rules was devised and tested on a small sample to further refine final categories and codes, based on coders’ feedback. The full codebook is provided in Appendix C.

Three coders conducted the content analysis, and inter-coder reliability was calculated with the online tool ReCal 0.1 Alpha for 3+ Coders. The Fleiss’ Kappa (κ = .709) and Krippendorff’s Alpha (α = .709) indicate a substantial level of agreement among coders and ensure methodological rigor and the validity of the content analysis.

This analysis was limited to correction lists and policies publicly available on websites of fact-checking organizations. These findings refer to the corrections’ transparency, not to the full volume of errors that may have occurred. Some semantic inaccuracies may have occurred because while Croatian, Bosnian, Montenegrin, Serbian, English, and French websites were analyzed in their original languages, other websites were automatically translated with Google Translate. This research included limited contextual variables. Future research should include political, cultural, or regulatory variables, since they may influence the transparency of fact-checking organizations. Additionally, research should be extended to error correction practices on social media platforms, as most fact-checking organizations disseminate their fact checks through multiple channels. Despite these limitations, the large-scale cross-national sample provides a comparative overview of correction transparency among EFCSN signatory organizations.