Large language models achieve a pooled accuracy of 63.69% (95% CI 60.26%-67.06%) in answering dental examination questions¶

Claim · All claims
Evidence · 1 study · quant-synthesis r3 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q3 i? Across 39 included studies, the random-effects pooled answer rate of LLMs on dental examination questions was 63.69% (95% CI: 60.26%-67.06%). → Dashti Mahmood 2026

Evidence¶

Dashti Mahmood 2026¶

Dashti Mahmood, Khosraviani Farshad, Meyari Atieh, Amirzade-Iranaq Mohammad Hosein, Chaurasia Akhilanand, Hefzi Delband, Ghadimi Niloofar, Tichy Antonin, Khurshid Zohaib, Schwendicke Falk. (2026). Accuracy of Large Language Models in Answering Dental Examination Questions: A Systematic Review and Meta-Analysis. International Dental Journal. https://doi.org/10.1016/j.identj.2026.109609

q3 · i? · quant-synthesis · r3

Meta-analysis of 39 studies of LLMs answering dental examination questions, using a random-effects model with Freeman-Tukey double arcsine transformation. The pooled answer rate was "63.69% (95% CI: 60.26%-67.06%)" with I² = 91.5%. No effect size beyond the pooled proportion is printed.

"Under the random-effects model, the meta-analysis of 39 studies yielded an overall pooled answer rate of 63.69% (95% CI: 60.26%-67.06%), with substantial between-study heterogeneity (I² = 91.5%, τ² = 0.0287; P < .0001)."

Discussion¶

Related Claims¶

• Subgroup analyses show significant between-model differences, with Copilot (75.25%) and ChatGPT-4 (72.97%) achieving the highest pooled accuracies — related
• Between-study heterogeneity in LLM dental-question accuracy is very high (I² = 91.5%), with a prediction interval spanning 29.65% to 91.71% — related
• Direct comparisons confirm ChatGPT-4 significantly outperformed earlier versions and some competitor models on dental examination questions — related
• Question provenance and format vary widely across studies: author-constructed sets dominate (53.8%), and 97.4% of studies used text-only questions — related
• Newer LLM versions generally outperformed older versions on dental examination questions — related