SMEs rated ChatGPT's content accuracy acceptable but inconsistent across questions and significantly below human TAs' accuracy¶

Claim · All claims
Evidence · 1 study (3 entries) · causal r1 · q2
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¶

q2 i? All ChatGPT responses to the eight laboratory questions scored at or above the neutral point of a 7-point Likert accuracy scale, but none reached the strongest ratings of 6 or 7. → M. Said Doğru & Emily K. Faulconer 2026 q2 i? SME-evaluated content accuracy of ChatGPT responses differed significantly across the eight questions, indicating the AI did not respond with consistent content accuracy. → M. Said Doğru & Emily K. Faulconer 2026 (2) q2 i? SME accuracy ratings significantly differed between ChatGPT and human TA responses for 5 of the eight questions, with human TAs considered more accurate on each of those questions. → M. Said Doğru & Emily K. Faulconer 2026 (3)

Evidence¶

M. Said Doğru & Emily K. Faulconer 2026¶

M. Said Doğru & Emily K. Faulconer. (2026). ChatGPT as a Virtual Laboratory Teaching Assistant in Undergraduate Biology. Research in Science Education. https://doi.org/10.1007/s11165-025-10271-z

q2 · i? · design · r2

Descriptive SME accuracy ratings (7-point Likert) of ChatGPT responses to eight student-generated enzyme-laboratory questions, from the RQ1 results. SMEs judged accuracy acceptable while noting "no questions showed stronger responses (6 or 7 on the Likert scale)".

"All ChatGPT responses for the eight questions scored at or above the neutral point of the Likert scale, but no questions showed stronger responses (6 or 7 on the Likert scale)."

M. Said Doğru & Emily K. Faulconer 2026 (2)¶

M. Said Doğru & Emily K. Faulconer. (2026). ChatGPT as a Virtual Laboratory Teaching Assistant in Undergraduate Biology. Research in Science Education. https://doi.org/10.1007/s11165-025-10271-z

q2 · i? · causal · r1

Two-factor repeated measures ANOVA over SME accuracy ratings of ChatGPT responses in the RQ1 analysis. The article reports significant variation across questions (F = 4.88, p = 0.000) and concludes "AI did not respond with consistent content accuracy".

"there are statistically significant differences in the SME-evaluated content accuracy across the eight questions ( F = 4.88, p = 0.000); AI did not respond with consistent content accuracy, according to the SMEs."

M. Said Doğru & Emily K. Faulconer 2026 (3)¶

M. Said Doğru & Emily K. Faulconer. (2026). ChatGPT as a Virtual Laboratory Teaching Assistant in Undergraduate Biology. Research in Science Education. https://doi.org/10.1007/s11165-025-10271-z

q2 · i? · causal · r1

Blinded repeated measures ANOVA comparison of SME accuracy ratings of AI- versus human-authored responses (RQ3), with Bonferroni-corrected post-hoc tests. On the questions differing significantly, "SMEs considered human TAs more accurate than ChatGPT".

"SME content accuracy ratings significantly differed between ChatGPT and human TA responses for 5 of the eight questions used in this study. For each of these questions, SMEs considered human TAs more accurate than ChatGPT."

Discussion¶

Related Claims¶

• Human TAs demonstrated stronger instructor voice, understanding, and helpfulness than AI in responding to student laboratory questions — related
• Students rated human TA responses more favorably overall but preferred ChatGPT responses 40% of the time — related
• Students found ChatGPT responses useful but insufficient in depth and detail compared with TA responses — related