The clinical research industry is understandably excited about what artificial intelligence (AI) will bring to science. The computational power and scale of AI tools could advance research in previously impractical ways by accelerating drug discovery, improving diagnostic models, and identifying patterns across datasets that no human team could process at a comparable speed. That transformation is real, and the attention it receives is warranted.
There is a second benefit that AI will bring to clinical trials that will get a smaller share of the headlines: time. Specifically, time returned to the people running studies at the site level. And that benefit may do more for participant enrollment in the near term than any of the more ambitious applications the industry is currently debating.
When asked how AI will improve participant enrollment, the field's instinct is to look at the technology's most powerful capabilities: the ability to identify candidates at large, query records across patient populations, and automate outreach. These are likely focus areas where AI will eventually go. Today, there are regulatory considerations to contend with.
Regulatory Considerations for AI and LLMs
Recruiting a clinical trial participant requires access to personal medical information and direct outreach to individuals. That information is protected by a framework of privacy laws and regulations. In Europe, the General Data Protection Regulation sets a high bar for lawful processing of personal data and has strict limitations on how data can be used and by whom. In the United States, HIPAA establishes its own framework governing access to protected health information. Beyond these, clinical trial participants are further protected by IRB and ethics committee oversight and informed consent requirements that exist regardless of jurisdiction. AI cannot circumvent these frameworks, nor should it.
The ability to query medical records has existed for years. Sites can review their own patient populations today to find potential matches for a given trial. AI will improve the efficiency of that process, but it will not bring new access to external patients. The regulatory pathway governing who can access those records, under what consent, and for what purpose remains firmly in place. While a large language model (LLM) could certainly review records and surface matches, access to those records is regulated, and that has not changed because AI has arrived. If anything, regulators in Europe and elsewhere are actively examining how existing frameworks apply to AI, and further guidance or restriction, rather than relaxation, is likely. The outreach side faces a different but equally real constraint. An AI agent could theoretically contact prospective participants much faster than a person; however, that is not the direction sponsors are signaling today. The human relationship at the point of recruitment is not a connection point they want to automate out. For now, that preference holds.
The direct path to AI-driven recruitment is impacted by two substantive and intentional barriers: regulatory protection and the value sponsors place on the human connection at the point of participant engagement.
What is important to remember is that sites are not facing the same restrictions. A site can review its own patient records today, within the consent and governance frameworks currently in place. A study coordinator can call a potentially eligible patient to gauge their interest in a trial today. Neither requires new regulatory permissions, new technology, or a shift in sponsor direction. What is limiting sites is not access - it is time.
Administrative Burden on Site Staff
Study coordinators are not underutilising their recruitment tools by choice. They are managing an active protocol while simultaneously handling clinic operations, regulatory documentation, adverse event reporting, invoicing, source data verification, grant writing, and a range of administrative tasks that are real, necessary, and largely invisible to anyone outside the site. These are not inefficiencies that a motivated coordinator can simply eliminate. They are essential tasks to be completed.
Clinical trials involve people. Patient care is not a line item. It is the center of gravity for everything happening at a clinical site. Patients in a hospital or care center require time, attention, and presence that cannot be scheduled around a task list. A conversation that was supposed to take ten minutes takes 40 when a family member has questions, or a patient is anxious about what participation means. These are not interruptions to the work. They are the job, and they are unpredictable in a way that administrative tasks, for all their burden, are not.
When that combination of administrative load plus the genuine human demands of a clinical environment fills the day, the tasks most likely to get crowded out are the ones with softer immediate deadlines: the follow-up call to a participant who missed a visit, the systematic review of recently diagnosed patients who might qualify for an open study. Not because those tasks are unimportant, but because there is no time left.
The bottleneck is not access. It is bandwidth.
This is where AI is already making an impact, even if the story is less dramatic than the industry tends to prefer. An AI tool that drafts regulatory correspondence, prepopulates case report forms, summarises a protocol amendment, or handles a first pass on invoice reconciliation is not doing direct recruiting. What it is doing is returning time to the person who otherwise would have spent hours on those tasks, time that can be redirected toward recruiting.
Consider the process of writing this article as a direct example. A first draft that previously required uninterrupted hours now takes a fraction of that time. The completion, the editing, review, and judgment calls remain human. What arguments to make, what to leave out, where the logic holds and where it breaks, what tone aligns with the audience, and the multiple edits are all made by a human. It is the same quality of thinking applied to a more efficient process. However, the time staring at a blank page is gone. Multiply that kind of time recovery across the administrative life of a study site. Multiply it across documentation, grant writing, regulatory submissions, correspondence, reporting, and the aggregate effect becomes significant. Study teams are not suddenly going to have empty afternoons. Hopefully, they will find that the tasks that were consistently crowded out are no longer consistently crowded out.
How AI Can Enable Human Oversight
The unresolved question, and the one the industry needs to be asking now, is where that recovered capacity will go. There is no immediate answer, and answering it requires a more fundamental rethink than simply redirecting hours. It requires rethinking what daily work should look like in a world where AI handles a significant share of the computational and organisational burden.
The tasks that consume the most time in clinical operations are rarely the tasks that require the most human judgment. Documentation, reconciliation, scheduling coordination, grant writing, reporting; these are tasks that are essential but not where people bring their most complete or creative selves. They are also notable tasks where AI performs well. Speed, consistency, and organization at scale are exactly what these functions require, and exactly what AI delivers.
People, by contrast, are better at empathetic aspects: connection, judgment, and synthesis in complex and unpredictable situations. The conversation with a hesitant participant about trial involvement. The reason why a site is struggling doesn't show up in the data. The creative solution to an enrollment problem that no query would surface. These are human capabilities, and they are most available when the person isn't mentally occupied by a backlog of administrative work they haven't gotten to yet.
The opportunity AI presents is not simply efficiency. It is a rebalancing of where human energy goes. AI handles computation and organization faster than people can. People handle connection and creative problem-solving better than AI can, particularly when they are not exhausted by tasks that needed to be automated years ago. The combination, AI doing what AI does well, people doing what people do well, could be genuinely powerful in a clinical research context. The question is whether organizations are intentional enough about the division to realize it.
History offers a useful and cautionary frame here. Email was supposed to streamline communication. And in a narrow sense, it did. Messages that once required days of postal transit arrived instantly. Email did not simply make existing communication faster. It opened a channel that had never existed: asynchronous, immediate, essentially costless to send. Within two decades, that channel expanded to consume a significant portion of the professional workday.
Email is now widely cited as one of the largest single time-consuming tasks in knowledge work. Not because it failed, but because it succeeded in creating demand that hadn't previously existed. Mobile communication followed a similar pattern: sold as a tool to make professionals more efficient, but it largely resulted in the expectation of constant availability. Response times collapsed, the boundary between work and off hours dissolved, and the time nominally saved by faster communication was absorbed into perpetual availability.
AI will almost certainly save time. Significant amounts of time, across a wide range of tasks. That is not in question. What is genuinely uncertain is where that time goes. Sites operate under persistent and compounding resource pressure. When administrative burden decreases, the path of least resistance is absorbing recovered capacity into higher volume, more participants, more protocols, more tasks, rather than focusing deeper into the work that already exists.
The opportunity is real. AI will give study teams something many of them have not reliably had: enough hours in a day to make a phone call, review a patient record, or have an unhurried conversation with someone weighing enrollment in a trial. Whether that time gets used that way is a question of intention and planning.
The Real Headline for Recruitment
The industry headline about AI and clinical trials tends toward the ambitious - autonomous agents identifying potential participants, predictive models optimizing site selection, and end-to-end digital recruitment journeys. Most of those will become real, eventually. The regulatory and sponsor constraints that restrict AI's direct path to recruitment may shift over time. For the near-term gain, that is quieter and directly actionable. AI will provide overburdened study teams with some of their much-needed time back. The question will be whether that recovered time will be applied to those tasks that can be done today, and that will move studies forward.
Recovering time is not a small gain. Enrollment failure remains one of the most consistent and costly problems in clinical research. If the answer to that problem, or even a meaningful part of it, turns out to be study teams having more time, that is worth taking seriously. AI's immediate contribution may not be the headline breakthrough the industry is waiting for, but maybe something more real.
