Can AI do clinical research coordinator work?

Not the whole job: AI can help draft and check records, but ICH E6(R3) keeps consent and other delegated trial tasks with qualified people overseen by the investigator (ICH E6(R3) 2.3.1).

On our published benchmark, CRLN-AIE-001, some models scored the same as practitioners and some above them, which measures judgment on written scenarios, not the job.

What stays with people

  • The consent discussion is conducted by the investigator or by qualified staff the investigator has delegated it to (ICH E6(R3) 2.8.5).
  • Every delegated task goes to a person who is qualified and adequately informed about the protocol and the task (ICH E6(R3) 2.3.2), and the delegation is recorded (ICH E6(R3) 2.3.3).
  • Trial-related medical decisions stay with a qualified investigator (ICH E6(R3) 2.7.1(a)), and assessing and reporting safety events is the investigator's responsibility (ICH E6(R3) 2.7.2(b)).
  • Visit work in the room: taking samples and measurements, answering a participant's questions, and noticing that someone does not look well.

Where AI can help today

  • Drafting visit checklists, reminders and notes for a coordinator to check before use.
  • Comparing a visit record with the protocol's schedule and flagging missing items, signatures or dates.
  • Summarizing a protocol amendment so the team can see what changed.
  • Practice: rehearsing the decisions a coordinator makes, with feedback, before facing them at a visit.
  • Any AI tool used for trial activities is a computerised system, so it needs documented procedures, trained users and validation scaled to its intended use and risk (ICH E6(R3) 4.3.1; ICH E6(R3) 4.3.2; ICH E6(R3) 4.3.4).

What the published benchmark shows, and what it does not

CRLN-AIE-001 is our evaluation protocol and response dataset. It puts AI models and human clinical research practitioners on the same scored decisions about how a trial is run, and it is published on Zenodo under CC BY 4.0 with its own DOI.

Our first version concluded that a frontier model and practitioners could not be told apart. The next version compared like with like and reversed that conclusion, and the earlier record stays online beside the correction. Compared like with like, some models from Anthropic, Google and OpenAI were statistically indistinguishable from the practitioners, and others scored above them.

The study states its own limit: the instrument's ceiling is too high to detect whether a model flips between a compliant and a non-compliant answer. It does not certify any model or person, and it is not a claim that a model can run a trial or do a coordinator's job.

Questions to answer before handing a coordinator task to AI

  • Which decision does the tool support, and which person checks its output?
  • What happens to a participant or to the data if it is wrong?
  • Has it been tested on decisions like yours, against people who do this work?
  • Does any participant information leave the site, and under what agreement?
  • Have you described the use to the sponsor and, where needed, the IRB?

Questions people also ask

Will AI replace clinical research coordinators?

Not under the current rules. ICH E6(R3) places consent, delegated tasks and their oversight with qualified people (ICH E6(R3) 2.3.1), so AI can change parts of the work but not who is responsible for it.

Is AI better than people at GCP decisions?

On one published instrument, some models scored the same as practitioners and some above them. That describes written scenarios with fixed options, not a real visit, and the instrument cannot detect every kind of error that matters.

Can AI obtain informed consent?

No. The consent discussion is conducted by the investigator or delegated, qualified staff (ICH E6(R3) 2.8.5). Digital materials can support it (ICH E6(R3) 2.8.1(c)).

Where can I see the benchmark results and corrections?

On the AI scoreboard and the CRLN-AIE-001 research page, both linked below, which show the published results, each version's DOI and both public corrections.

Primary sources

  • ICH E6(R3) Guideline for Good Clinical Practice, final version adopted 6 January 2025 (ICH): https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Step4_FinalGuideline_2025_0106.pdf
  • Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, draft guidance (FDA): https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-artificial-intelligence-support-regulatory-decision-making-drug-and-biological
  • CRLN-AIE-001: Evaluating AI systems on clinical research competency against a human baseline, concept DOI 10.5281/zenodo.22708751 (Zenodo, CC BY 4.0): https://doi.org/10.5281/zenodo.22708751

Last reviewed

2026-11-14