In our Research Insights series, members of the Learning Design team connect areas of research expertise to current themes in online learning. This piece focuses on connections between research on exposure to inaccurate information and learning from texts.
Amalia Schwee is an Associate Director of Instructional Design and the Quality Assurance Principal in the Learning Design department at Everspring. She has a PhD in Learning Sciences from Northwestern University and lives in Chicago, Illinois.
Reading is an essential component of formal and informal learning experiences. The quality of the information we read can have a significant impact on learning outcomes. When texts are thoughtfully constructed and present accurate information, readers are better equipped to develop valid and useful knowledge. Conversely, when texts are vague, misleading, or incorrect, readers may develop misconceptions or inaccurate understandings. In contemporary learning environments, students may engage with a wide range of text-based content with varying degrees of accuracy, including generative artificial intelligence (AI) outputs. This can amplify the need for and the benefits of supporting students’ evaluative engagement with the materials they read. This piece provides an overview of research on the effects of exposure to inaccurate information in text content and presents recommendations for mitigating reliance on inaccuracies in coursework.
Reliance on Inaccurate Information
An extensive body of research has examined the influences of information accuracy on post-reading assessment performance (Fazio et al., 2013; Marsh et al., 2003; Rapp et al., 2014). In such studies, readers are often presented with texts containing accurate information about general knowledge topics as well as inaccurate information and “neutral” or unspecified references to general knowledge topics. For example, study participants might be assigned to read one of three versions of a text that includes information about the largest ocean in the world. One group of study participants would read that the largest ocean on Earth is the Pacific (the accurate condition), whereas another group of participants would read that the Atlantic is the largest ocean (the inaccurate condition). A third and final group of participants would see a reference to the largest ocean on Earth, lacking any mention of the ocean’s name (the “neutral” condition).
Findings from such studies have demonstrated that after reading texts containing falsehoods, readers are more likely to reproduce that false information in completing subsequent tasks (e.g., answering a free-response question of “What is the largest ocean in the world?” with the Atlantic rather than the Pacific) as compared to after reading texts with accurate information or “neutral” information. While these findings are most pronounced for “hard” items—items for which participants are unlikely to hold relevant prior knowledge that can be used to validate text information—they also emerge for “easy” items on familiar, well-known topics (Marsh et al., 2003; Marsh & Fazio, 2006). For a deeper methodological review, see Marsh et al., 2003.
Given the implications these findings could have for all types of reading experiences, researchers have investigated factors that contribute to reliance on inaccurate information and interventions that might reduce reliance. Plausibility, for example, has been shown to influence rates of inaccurate information use, with readers being less likely to reproduce implausible as compared to plausible falsehoods (Hinze et al., 2014). Interventions aimed at reducing reliance on inaccuracies have often yielded limited success. Examples of such interventions include the following:
- Instructing readers to re-read texts (Marsh et al., 2003)
- Presenting stories slowly (Fazio & Marsh, 2008)
- Highlighting inaccurate text information (Eslick et al., 2011)
- Offering warnings about potential falsehoods (Marsh & Fazio, 2006)
What might account for the ineffectiveness of these interventions? Simply put, these approaches can backfire by making inaccurate information more salient to readers. Highlighting inaccuracies alone, for example, can make inaccuracies stand out more. This can, in turn, make readers more likely to remember them and potentially rely on them later on.
While evidence of readers’ reliance on inaccurate information has been a cause for concern, there are situations in which readers are relatively less likely to encounter falsehoods. When engaged with texts curated for a course, for example, readers are likely to find accurate, reliable information corroborated by other learning content shared by their instructor. Compared to other reading-based activities, such as scrolling through social media posts or seeking out information online about current events (Pennycook & Rand, 2021; Suarez-Lledo & Alvarez-Galvez, 2021), course-based reading may warrant less concern about potential reliance on inaccurate information. However, increased usage of AI in courses—both sanctioned and unsanctioned—adds a layer of complexity. While AI can be helpful for both students and instructors alike, AI is also prone to hallucinating or generating inaccurate information (Augenstein et al., 2024). AI hallucinations are particularly concerning because AI-generated content may often exhibit features like plausibility that can encourage readers to rely on the information provided (Google Cloud, n.d.). Similar risks can arise from other digital sources of information students often rely on, including collaborative study documents and unvetted web resources that may not receive routine updates or verification. As noted previously, readers have been shown to exhibit reduced reliance on implausible inaccuracies as compared to plausible inaccuracies. Plausible inaccuracies, however, can go undetected—even by skilled readers (Hinze et al., 2014).
With these considerations in mind, what can be done to encourage effective validation of text content and mitigate concerns about reliance on inaccurate information? While many interventions aimed at reducing inaccurate information use have been unsuccessful, some studies have yielded more promising results and can usefully inform instructional design decisions. Additionally, patterns of inaccurate information use identified in this body of research can help instructors target their efforts to encourage information validation more effectively. The following recommendations, informed by both successful interventions and pervasive patterns, can be readily integrated into coursework across disciplines.
Recommendations
Integrate guidance and resources for information validation into activities and assessments.
As described previously, a range of interventions aimed at reducing reliance on inaccurate information have failed to yield substantial reductions. However, some interventions—namely those oriented around fact-checking—have been shown to attenuate rates of inaccurate information use. Donovan and Rapp (2020) found that when readers were given the option to search online for information—a common feature of naturalistic reading experiences, but a rarity in lab-based evaluations of reading comprehension—they ultimately reproduced fewer inaccuracies in completing post-reading tasks than did readers who were not given this option. The authors also found that warning readers about their potential exposure to inaccuracies in texts they’d read bolstered search rates, though more frequent fact-checking was not found to further reduce rates of inaccurate information use.
Overall, these results are encouraging, suggesting that readers engage in and benefit from online searches. However, rates of fact-checking observed were relatively low, and other research suggests that students often forgo fact-checking or lack the skills to do so successfully (Brodsky et al., 2021; Wineburg & McGrew, 2019). With this in mind, we recommend going beyond simply reminding students that they can or should engage in fact-checking by offering direct support and guidance. Demonstrating how to corroborate claims effectively by consulting trusted sources, for example, can increase the likelihood of students engaging in fact-checking efforts successfully.
Supplement activities and assessments with reflection and discussion.
Inaccuracies—particularly plausible ones—can go undetected as readers process texts (Hinze et al., 2014). In laboratory experiments, it can be feasible to include synchronous prompts to encourage the detection of inaccuracies during reading. In a classroom context, consider embedding reminders into course materials and assessments. For example, you might integrate suggestions and reminders for reviewing the accuracy of reading materials directly into course pages. This can be especially helpful for situations in which students are selecting their own materials or are likely to explore content beyond what’s directly included in the course. If AI usage is integrated into your assessments, you might ask students to share elements of AI content they generated and describe the steps they took to verify the accuracy of the content. This practice can both encourage more evaluative processing during reading and elevate the quality of assessment submissions. Similarly, incorporating follow-up discussions for activities and assessments can provide students with opportunities to reflect on when they have and haven’t successfully detected inaccuracies. You, as the instructor, can also leverage discussions to share your own experiences with encountering inaccuracies and strategies you’ve found helpful for validating information. In addition, you can model strategies for validating information, such as thinking aloud while working through a text or introducing simple checklists that help students practice evaluating sources.
Assess students’ levels of prior knowledge.
Unsurprisingly, readers are more likely to rely on inaccurate information when answering “hard” questions (those for which they likely lack prior knowledge) as compared to “easy” questions (those for which they likely possess relevant prior knowledge) (Donovan & Rapp, 2020). In studies on inaccurate information use, questions are categorized as hard or easy based on norming studies that evaluate general knowledge in the populations from which study samples are drawn (e.g., Nelson & Narens, 1980; Tauber et al., 2013). Incorporating prior knowledge assessments in the classroom can similarly help you determine which concepts or topics students know well and which they don’t. You can then tailor your instructional approach accordingly, such as by providing students with additional resources to support understanding of less familiar concepts or topics and validation of information students might encounter about those concepts or topics elsewhere.
Conclusion
Reading is a key component of many learning experiences. While much of what we read can help us, inaccuracies contained in texts can negatively impact post-reading outcomes. Even in learning experiences that feature reliable sources, learners may still be exposed to inaccurate information, such as through interactions with AI-generated content reflecting hallucinations. Such exposure can result in students unknowingly reproducing false information in assessment submissions, especially when students rely on AI-generated responses. This highlights the importance of designing assessments that both evaluate knowledge and provide opportunities to address and correct misunderstandings. As exposures to inaccuracies may be even more common in everyday reading experiences, actively supporting students’ validation of information they read can yield substantive benefits that extend beyond the classroom and prepare them for navigating the digital world successfully.
References
Augenstein, I., Baldwin, T., Cha, M., Chakraborty, T., Ciampaglia, G. L., Corney, D., DiResta, R., Ferrara, E., Hale, S., Halevy, A., Hovy, E., Ji, H., Menczer, F., Miguez, R., Nakov, P., Scheufele, D., Sharma, S., & Zagni, G. (2024). Factuality challenges in the era of large language models and opportunities for fact-checking. Nature Machine Intelligence, 6(8), 852–863.
Brodsky, J. E., Brooks, P. J., Scimeca, D., Todorova, R., Galati, P., Batson, M., Grosso, R., Matthews, M., Miller, V., & Caulfield, M. (2021). Improving college students’ fact-checking strategies through lateral reading instruction in a general education civics course. Cognitive Research: Principles and Implications, 6(1), Article 23.
Donovan, A. M., & Rapp, D. N. (2020). Look it up: Online search reduces the problematic effects of exposures to inaccuracies. Memory & Cognition, 48(7), 1128–1145.
Eslick, A. N., Fazio, L. K., & Marsh, E. J. (2011). Ironic effects of drawing attention to story errors. Memory, 19(2), 184–191.
Fazio, L. K., Barber, S. J., Rajaram, S., Ornstein, P. A., & Marsh, E. J. (2013). Creating illusions of knowledge: Learning errors that contradict prior knowledge. Journal of Experimental Psychology: General, 142(1), 1–5.
Fazio, L. K., & Marsh, E. J. (2008). Slowing presentation speed increases illusions of knowledge. Psychonomic Bulletin & Review, 15, 180–185.
Google Cloud. (n.d.). What are AI hallucinations?
Hinze, S. R., Slaten, D. G., Horton, W. S., Jenkins, R., & Rapp, D. N. (2014). Pilgrims sailing the Titanic: Plausibility effects on memory for misinformation. Memory & Cognition, 42(2), 305–324.
Marsh, E. J., & Fazio, L. K. (2006). Learning errors from fiction: Difficulties in reducing reliance on fictional stories. Memory & Cognition, 34, 1140–1149.
Marsh, E. J., Meade, M. L., & Roediger, H. L. (2003). Learning facts from fiction. Journal of Memory and Language, 49(4), 519–536.
Nelson, T. O., & Narens, L. (1980). Norms of 300 general-information questions: Accuracy of recall, latency of recall, and feeling-of-knowing ratings. Journal of Verbal Learning and Verbal Behavior, 19(3), 338–368.
Pennycook, G., & Rand, D. G. (2021). The psychology of fake news. Trends in Cognitive Science, 25(5), 388–402.
Rapp, D. N., Hinze, S. R., Kohlhepp, K., & Ryskin, R. A. (2014). Reducing reliance on inaccurate information. Memory & Cognition, 42(1), 11–26.
Suarez-Lledo, V., & Alvarez-Galvez, J. (2021). Prevalence of health misinformation on social media: Systematic review. Journal of Medical Internet Research, 23(1), Article e17187.
Tauber, S. K., Dunlosky, J., Rawson, K. A., Rhodes, M. G., & Sitzman, D. M. (2013). General knowledge norms: Updated and expanded from Nelson and Narens (1980) norms. Behavior Research Methods, 45(4), 1115–1143.
Wineburg, S., & McGrew, S. (2019). Lateral reading and the nature of expertise: Reading less and learning more when evaluating digital information. Teachers College Record, 121(11), 1–40.