Research Article
AI tutors vs. tenacious myths: Evidence from personalised dialogue interventions in education
Synthesis: Corbett & Tangen (2026) test whether personalized AI dialogue can correct deeply held misconceptions ("tenacious myths") in psychology and education. In a preregistered experiment (N = 375), participants holding strong misconceptions engaged in one of three interventions: personalized Misconception AI Dialogue targeting their specific belief, generic Textbook-style Refutation (a refutation text), or Neutral AI Dialogue (control). Personalized Misconception AI Dialogue produced significantly larger immediate belief reductions than both Textbook Refutation and Neutral AI Dialogue; this advantage persisted at 10-day follow-up but diminished by 2 months, where Misconception AI Dialogue and Textbook Refutation converged while both remained superior to control. Both AI conditions generated significantly higher engagement and confidence than Textbook Refutation reading, showing the motivational benefits of conversational interaction.
Key Findings
- Personalized AI dialogue beats generic refutation — initially. Personalized Misconception AI Dialogue produced significantly larger immediate belief reductions than both Textbook-style Refutation and Neutral AI Dialogue, disrupting the cognitive processes that maintain misconceptions.
- The advantage fades over time. The personalized-dialogue advantage persisted at 10-day follow-up but diminished by 2 months, where Misconception AI Dialogue and Textbook Refutation converged — both still superior to control. This suggests brief interventions require spaced reinforcement for lasting change.
- AI conditions are more engaging and confidence-building. Both AI conditions generated significantly higher engagement and confidence than reading a Textbook Refutation, highlighting the motivational benefits of conversational interaction.
- Refutation texts still work. Textbook-style refutation remained superior to control at all time points — confirming the effectiveness of the refutation text technique even when outshone on immediacy by personalized dialogue.
What this means for practice
- Instructors. Reach for personalized AI dialogue when the goal is immediate misconception correction: it produced significantly larger immediate belief reductions than both the textbook refutation and the neutral AI dialogue control.
- Designers. Build spaced reinforcement into the program rather than scheduling one-off chats, because the dialogue advantage persisted at 10 days but converged with textbook refutation by the 2-month follow-up.
- Designers. Keep the refutation text in the toolkit — it beat the neutral control at every time point — and choose dialogue where engagement and confidence matter, since both AI conditions generated significantly higher engagement and confidence than reading did.
- Instructors. Target the specific belief the learner holds: the dialogue condition addressed each participant's highest-rated misconception, and no intervention changed misconceptions it did not address.
- Researchers. Check the domain before generalizing: this experiment tested everyday psychology and education myths with adults, and the comparison reversed in at least one science-education study (Akdoğan 2025).
Limitations
- The experiment used N = 375 volunteer participants aged 19–78 (M = 40.25), 60.8% of them in the United Kingdom — not a classroom sample of the students these interventions would serve.
- The intervention was a single three-round dialogue with Claude 3.5 Sonnet; dropout by the 10-day follow-up was 5.6% for the dialogue condition, 6.4% for the neutral control, and 0.8% for the textbook condition.
- No process measures were collected during the intervention (response times, think-aloud protocols, linguistic analysis of responses), so the authors cannot distinguish reasoning-based from fluency-based mechanisms.
- The exploratory measures — confidence and trust in AI — were collected post-intervention only, with no pre-intervention baseline, and measurement invariance across time points and demographic groups was not established for the misconception scale.
Connected Concepts
Connected Articles
- Comparing the effectiveness of expert-written text, AI-generated text, and interactive AI dialogues on students' — Comparing expert-written vs. AI-generated conceptual change text vs. interactive AI dialogue
- Building AI Companions that Prioritise Learning over Performance — AI companions and misconception correction
- GenAI Knowledge, Epistemic Orientation, and Intellectual Values Predict Undergraduate Students' Critical GenAI Use — Recommends refutation texts to target conceptual misconceptions
Citation
Corbett, B. J., & Tangen, J. M. (2026). AI tutors vs. tenacious myths: Evidence from personalised dialogue interventions in education. Computers in Human Behavior, 175, 108828.