Generative AI · Education leadership

Actionable companion · Assessment and evidence of learning

From evidence to district action: Assessment of drug harm by final-year medical students and ChatGPT – a comparative analysis

AI-assisted editorial implementation guide. The district actions are proposed applications, not findings or recommendations attributed to the original authors. No additional peer review is asserted.

Read original paperSearch paper agentCC BY 4.0

Evidence and context

Health-professions education context: test transfer to K–12 rather than assuming it.

Original authors’ abstract, reproduced under the linked source license:

Abstract Background Artificial intelligence (AI) is becoming increasingly integrated into medical education and clinical practice. Junior medical staff often act as first-line responders in substance-related presentations, yet their perceptions of drug-related harm are not well characterised. It is unclear how their assessments compare with harm ratings generated by emerging AI tools. This study aimed to compare final-year medical students’ harm ratings of recreational drugs and alcohol to users and society with those generated by ChatGPT. Methods 110 final-year medical students (59.1% response rate) rated the harmfulness of 28 recreational drugs to individual users on a 1–10 scale and ranked seven drug classes for societal harm. ChatGPT-4 (paid subscription version) generated 40 response sets in five conversations through repeated administration of the same survey. Discrepancies in harm ratings to users and rankings of societal harm were analysed descriptively. Results Students rated their drug harm knowledge at 5.8/10 (SD 1.7) and their education on drug harm at 5.3/10 (SD 1.8). ChatGPT rated its drug harm knowledge at 8.3/10 (SD 1.2). The mean harm-to-user scores across the 28 substances were 6.7 (SD 1.8) for students and 6.4 (SD 0.8) across the repeated ChatGPT response sets. Both groups identified heroin, crack cocaine, and fentanyl as the most harmful substances. Differences were observed for less harmful substances: students rated snus (smokeless tobacco), laughing gas, and cannabis least harmful, whereas ChatGPT rated magic mushrooms, DMT/ayahuasca, and LSD least harmful. ChatGPT ratings were more similar to those of previously published expert panels. Conclusions Students and ChatGPT strongly agreed on the most harmful substances but differed in their assessments of moderately and minimally harmful drugs. These differences suggest that perceptions of drug harm are shaped not only by pharmacology and toxicity but also by social and cultural context, local patterns of use, clinical exposure and experience, and medical training. AI may be of limited value when the evidence base is sparse, contested, or difficult to trace. The lower confidence among students suggests that further research is needed to better understand educational needs in drug-harm assessment and the critical appraisal of AI-generated evidence. Supplementary Information The online version contains supplementary material available at https://doi.org/10.1186/s13011-026-00762-1. Keywords: ChatGPT, Medical students, Recreational drugs, Substance use disorders, Artificial intelligence, Illicit drugs, Alcohol

A 90-day district pilot

  1. Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
  2. Days 16–30 · Design. Redesign one low-stakes assessment: require a first attempt, an AI-assisted revision, a source-check log and a short oral explanation. Use the same rubric for the comparison task. Name an accountable owner and set the baseline, success threshold and budget before implementation.
  3. Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
  4. Days 61–90 · Decide. Blind-score reasoning, factual accuracy and unaided transfer on a fresh task. Record suspected misconduct separately from demonstrated evidence. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Isak Näslund; Michael Ott; Alexander Vucins; Ursula Werneke. Assessment of drug harm by final-year medical students and ChatGPT – a comparative analysis. Substance Abuse Treatment, Prevention, and Policy (2026-09-12). https://doi.org/10.1186/s13011-026-00762-1

Companion prepared 2026-10-30. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.