AI Can Help Design Better Trials, But It Still Can't Tell You Whether Patients Will Join Them

By Ross Jackson, Ross Jackson Consulting

AI is already changing the way clinical trials are discussed, planned, and potentially designed.

Much of the conversation understandably focuses on speed and efficiency. AI may help sponsors analyze large data sets more quickly, identify patient subgroups, model protocol scenarios, optimize endpoints, and make better-informed design decisions before a trial reaches the operational stage.

These are all meaningful developments. If AI can help sponsors avoid poorly designed studies, improve scientific decision-making, and reduce unnecessary complexity, then it should be welcomed.

But there is a danger in assuming that an AI-optimized trial is automatically a recruitable trial.

A protocol can be scientifically sound, statistically robust, and operationally well modelled, yet still struggle once it starts looking for real patients, real sites, and real coordinators. The question is not only whether AI can help design a better trial. It is whether the resulting trial is one that patients will understand, sites can deliver, and clinical teams can recruit for.

That distinction matters because many recruitment problems are not created after a trial launches; they are baked in much earlier.

Recruitment Failure Often Starts Before Recruitment Begins

When a trial is under-recruiting, the response is often to look downstream.

Sponsors may ask whether the advertising campaign is working, whether sites are making enough calls, whether digital vendors are generating enough referrals, or whether patient materials need to be refreshed. Those questions can be important, but they often come too late.

By the time a trial is open to recruitment, some of the most important recruitment decisions have already been made.

The inclusion and exclusion criteria are set. The visit schedule is defined. The endpoint strategy is locked. The geography has been selected. The sites have been chosen. The patient-facing burden has been created. The assumptions about prevalence, motivation, competition, and conversion have often been accepted.

If those assumptions are wrong, recruitment will almost certainly be compromised before the first patient is approached.

That is why AI-driven trial design needs to include the ability to recruit as a design consideration, not merely as a downstream operational problem.

An AI-Optimized Protocol May Still Be Hard To Recruit For

AI could help sponsors refine protocols in ways that are scientifically or statistically beneficial. It could identify a more precise patient subgroup, suggest additional data collection, recommend endpoints, or model different scenarios against historical and real-world data.

But optimization always depends on the question being asked.

A design may be optimized for statistical power, regulatory confidence, endpoint sensitivity, or scientific clarity. But that does not necessarily mean it has been optimized for patient participation.

For example, a narrower patient subgroup may improve scientific precision but reduce the available recruitment pool. Additional assessments may strengthen the data set but increase participant burden. More frequent visits may create better monitoring but make the trial unrealistic for patients who work, travel, care for others, or live far from the site. A more selective site list may look efficient on paper but fail if those sites are already saturated with competing studies.

This is where the AI discussion needs to become more practical.

The question is not simply “Can AI help us design a better trial?” The question is “Better for whom, and against which definition of success?”

A trial that can't recruit is hardly "better" in any meaningful operational sense.

The Tension Between Scientific Optimization And Real-World Feasibility

Clinical development has always involved trade-offs. AI will not remove those trade-offs. In some cases, it may make them easier to see. In others, it may make them easier to ignore.

There is often a tension between a trial that looks best from a scientific or statistical perspective and a trial that works best in the real world.

But here’s the thing that often gets overlooked by people within the industry: Patients do not experience a trial as a protocol. They experience it as time, travel, uncertainty, procedures, disruption, possible side effects, and conversations with their family, physician, and clinical team.

Sites do not experience a trial as a strategic asset. They experience it through feasibility questionnaires, competing workloads, staff capacity, training requirements, screening logs, queries, visit complexity, and the reality of whether their patient population matches what the sponsor believes is available.

Coordinators do not experience a trial as a sophisticated design. They experience it as the practical challenge of explaining the study, managing expectations, scheduling visits, following up with patients, handling screen failures, and keeping people engaged.

If AI is used primarily by teams focused on design, modelling, and development strategy, it is unlikely to capture those human and operational realities.

That does not mean AI is the problem. It means the inputs, prompts, assumptions, and success criteria need to be broader.

Ability To Recruit Should Be Pressure-Tested Earlier

One of the most valuable uses of AI may be to help sponsors pressure-test the ability to recruit before timelines start slipping.

That could include asking much more practical questions during protocol development:

• Would a patient understand why this trial is relevant to them?
• Would the visit schedule feel manageable?
• Do the eligibility criteria exclude large numbers of otherwise interested patients?
• Are the most burdensome procedures essential or merely desirable?
• Would a busy site see this as an attractive study to prioritize?
• How does the trial compare with competing studies in the same indication?
• Where might coordinators struggle to explain the trial clearly?
• What objections might patients, caregivers, or referring physicians raise?
• What parts of the patient pathway are likely to create delay, confusion, or drop-off?

These are not peripheral questions. They are central to whether the trial can succeed.

AI may be useful in this regard, not because it replaces recruitment expertise but because it allows sponsors to interrogate the trial from multiple perspectives earlier and more systematically. A protocol could be previewed through the lens of a patient, caregiver, principal investigator, study coordinator, referring physician, site director, or recruitment strategist before the operational damage is done.

That kind of pressure-testing should not be treated as a soft exercise. It should be part of the overall risk management process.

AI Can Surface Problems But It Can’t Own The Judgment

There is another risk in the current AI discussion — the temptation to treat AI-generated recommendations as more objective than they really are.

AI can help identify patterns. It can challenge assumptions. It can summarize complex information. It can compare scenarios. It can generate questions that teams may not have considered. Used well, it can be a powerful aid in decision-making.

But it doesn’t truly understand the lived reality of trial participation.

It doesn’t sit with a patient who is unsure whether another biopsy is worth it. It doesn’t hear the hesitation in a caregiver’s voice when travel burden is discussed. It doesn’t see how a coordinator reacts when yet another complex study is added to an already overloaded site. It doesn’t know which supposedly high-enrolling sites are genuinely motivated and capable and which are simply optimistic during feasibility.

Those realities still require human judgment.

This is especially important when a trial is already falling behind. By that stage, the problem is rarely solved by one more dashboard or another layer of analysis. The sponsor needs to understand what is happening in practice — which sites are active, which are stuck, where referrals are being lost, whether patients are declining, whether the trial is being explained effectively, and whether the assumptions underpinning the recruitment plan were ever realistic.

AI may help diagnose some of that. But experienced human interpretation is still needed to decide what to do next.

Can AI Help Us Ask Better Recruitment Questions?

The most useful role for AI in patient recruitment may not be to promise certainty. It may be to improve the quality of the questions sponsors ask before and during a trial.

Instead of asking, “How quickly can we recruit?” sponsors could ask:

• What assumptions would have to be true for this recruitment plan to work?
• Which of those assumptions are weakest?
• Where are we relying on hope rather than evidence?
• How does this trial look from the patient’s perspective?
• How does it look from the coordinator’s perspective?
• What would make a site prioritize this study over another?
• What would make a patient decline this study even if they appear eligible?
• What early signals would tell us the strategy is not working?

These questions may be uncomfortable to consider, but they are certainly valuable. They force a trial team to look beyond the protocol as a scientific document and consider it as a real-world participation proposition.

That is where AI could make a genuine contribution. Not by pretending to guarantee recruitment success but by helping sponsors identify weaknesses before they become expensive delays.

Trial Design And Recruitment Are Interlinked

The industry often talks about trial design and patient recruitment as if they are separate phases.

First, the trial is designed. Then it is handed to operations. Then recruitment begins. Then, if enrollment disappoints, recruitment specialists are brought in to fix the problem.

That sequence is part of the problem.

Recruitment is not just an executional activity. It is influenced by design decisions made months or even years earlier. If AI is going to reshape clinical trial design, then ability to recruit needs to be part of that reshaping from the start.

A better-designed trial is not only one that answers the scientific question. It is one that can be delivered by sites, understood by patients, supported by caregivers, and completed with the least unnecessary burden possible.

AI may help us get closer to that, but only if we resist the temptation to view recruitment as a downstream marketing or site performance issue.

The real opportunity is to use AI to connect scientific design with operational reality, because the ultimate test of a clinical trial is not whether it looked good in a simulation. It is whether the right patients joined, stayed, and generated the evidence the trial was designed to produce.

About The Author:

Ross Jackson is a patient recruitment specialist and author of the books The Patient Recruitment Conundrum and Patient Recruitment for Clinical Trials using Facebook Ads.

Having started out with digital marketing in 1998, Ross quickly developed a specialty in the healthcare niche, evolving into a focus on clinical trials and the problems of patient recruitment and retention.

Over the years Ross branched out from the purely digital and now operates in an advisory capacity helping sponsors, CROs, sites, solutions providers, and others in the industry to improve their patient recruitment and retention capabilities — having advised and consulted on over 100 successful projects.