Confronting Misinformation Produced with Generative AI

The document serves as a technical appendix to a meta-analysis on generative AI misinformation. It details the risk of bias assessment for various randomized controlled trials using the Cochrane RoB 2 protocol, finding that most studies scored 'low' to 'moderate' for bias. It also describes tests for publication selection bias—including Egger's test and the Duval and Tweedie trim-and-fill method—across samples focusing on visual misinformation, corrective information, and content labeling, concluding that while some small-study effects may exist, they do not fully explain the observed effects. Additionally, it provides supplementary figures on the perception of GenAI misinformation and the efficacy of corrective information.

• The risk of bias assessment using the Cochrane RoB 2 protocol indicated that most publications in the sample performed well, with scores ranging from "low" to "moderate" across most domains. A small number of publications showed a moderate risk of bias, primarily due to the randomization process, and one publication was removed entirely due to numerous concerns regarding random assignment.
• Analysis of publication selection bias for the post-2021 sample on visual misinformation perception showed no strong statistical evidence of funnel plot asymmetry via Egger's test (p = 0.17). However, the Duval and Tweedie trim-and-fill method suggested a mild asymmetry potentially consistent with small-study bias, though a 3-parameter selection model (3PSM) found no reliable evidence of publication bias.
• For post-2020 preventive correction information experiments, Egger's test and heterogeneity tests were not significant, and the 3PSM test did not detect publication bias (p = 0.85). Despite this, visual analysis of the funnel plot suggested that smaller studies reporting stronger negative effects might be disproportionately represented.
• In the full sample of content labeling studies, Egger's test indicated no significant evidence of small-study effects. While trim-and-fill analysis adjusted the overall effect size to g = -0.25 (p < 0.001), the 3PSM test confirmed the effect remained significant with a bias-adjusted effect size of -0.24, indicating that publication bias does not fully account for the observed effect.

APA

Herasimenka, A., Valenzuela, S., Boulianne, S., Esser, F., Given, L. M., Lewandowsky, S., Navarro-Lopez, E. M., & Howard, P. N. (2026). Confronting Misinformation Produced with Generative AI. International Panel on the Information Environment. https://cdn.prod.website-files.com/643e793799ecf3065c893f9d/6a15599a2cfb42fe2e6843d8_Online%20Appendix%20Final.pdf

Chicago

Herasimenka, A., S. Valenzuela, S. Boulianne, F. Esser, L. M. Given, S. Lewandowsky, E. M. Navarro-Lopez, and P. N. Howard. Confronting Misinformation Produced with Generative AI. International Panel on the Information Environment, 2026. https://cdn.prod.website-files.com/643e793799ecf3065c893f9d/6a15599a2cfb42fe2e6843d8_Online%20Appendix%20Final.pdf.

Wikipedia

{{cite report |last1=Herasimenka |first1=A. |last2=Valenzuela |first2=S. |last3=Boulianne |first3=S. |last4=Esser |first4=F. |last5=Given |first5=L. M. |last6=Lewandowsky |first6=S. |last7=Navarro-Lopez |first7=E. M. |last8=Howard |first8=P. N. |title=Confronting Misinformation Produced with Generative AI |publisher=International Panel on the Information Environment |date=2026 |url=https://cdn.prod.website-files.com/643e793799ecf3065c893f9d/6a15599a2cfb42fe2e6843d8_Online%20Appendix%20Final.pdf |access-date=23 October 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{herasimenka2026confronting, author = {Herasimenka, A. and Valenzuela, S. and Boulianne, S. and Esser, F. and Given, L. M. and Lewandowsky, S. and Navarro-Lopez, E. M. and Howard, P. N.}, title = {{Confronting Misinformation Produced with Generative AI}}, institution = {International Panel on the Information Environment}, year = {2026}, url = {https://cdn.prod.website-files.com/643e793799ecf3065c893f9d/6a15599a2cfb42fe2e6843d8_Online%20Appendix%20Final.pdf}, urldate = {2026-10-23}, note = {Indexed by Climate Insights Directory} }