Digital research tasks can filter participants before a single intervention begins
Digital health studies often present enrollment as a clean binary: eligible or not, consented or not, randomized or not. This qualitative observational study shows a messier reality. Among 50 adults with lower digital literacy, pre-intervention research tasks such as reviewing AI-avatar study information, completing electronic consent, and entering electronic patient-reported outcome data were not simply administrative steps. They were lived experiences shaped by confidence, device access, family support, interface design, and the degree to which a person could navigate the system without help.
The result was not just a question of whether someone finished the task. It was a question of how completion happened, who made it possible, and what it cost emotionally.
Participation was relational, not purely individual
Interviews and think-aloud sessions revealed that participation was often a shared activity rather than a solitary one. Supporters reassured participants, handled devices and accounts, and in some cases completed parts of the process on a participant’s behalf. Completion therefore took several forms: independent, assisted, shared, or proxy.
This matters clinically because digital trial infrastructure often assumes a stable, self-directed user. In practice, many patients enter digital studies through social scaffolding. A spouse may help with passwords. A family member may read instructions aloud. A supporter may steady the process when anxiety rises. Those interactions can expand access, but they can also blur the line between the participant’s own understanding and someone else’s interpretation of the study materials.
The study design captured this dynamic in real time by combining verbal accounts with screen interactions and contextual observations. That method matters because completion statistics alone would have missed the amount of invisible work required to move through the digital pathway.
Interface friction accumulated across linked tasks
The burden did not come from one isolated screen. It built across a sequence of linked activities, each one small enough to seem manageable in isolation and exhausting when stacked together. Study information, electronic consent, and ePRO completion each introduced opportunities for confusion, delay, or loss of confidence. A participant who could manage one step might still struggle when the next required a different login, a new piece of language, or an unfamiliar way of moving through the interface.
Distress, frustration, and reduced confidence emerged as repeated themes. These were not trivial emotional reactions. In a digital trial, confidence can determine whether a person proceeds, asks for help, or drops out before receiving the intervention under study. For people already operating near the threshold of digital exclusion, friction at the consent stage may function as an unrecognized gatekeeper.
One step can end a study before treatment ever begins.
AI-avatar acceptability depended on trust, disclosure, and human backup
The work also explored reactions to AI-avatar study information. Acceptability depended on several conditions: transparent disclosure, clear institutional provenance, and access to human support. Participants were more likely to engage when the technology was presented as coming from a recognizable research institution and when a person remained available to answer questions.
This is a useful signal for designers of digital trials using artificial intelligence–mediated communication. An AI-avatar may reduce staff burden and standardize delivery, but it cannot substitute for trust. For participants with lower digital literacy, the credibility of the messenger and the visibility of human backup are part of the intervention environment, not extras layered on afterward. If the technology appears opaque, impersonal, or disconnected from an accountable institution, participation may narrow to people already comfortable with digital systems.
The implication extends beyond convenience. Informed consent is only meaningful when participants can understand what they are agreeing to and can seek clarification without penalty. AI-driven presentation tools may improve reach, but only if they preserve the participant’s ability to pause, question, and re-engage with a human when needed.
The first hurdle may determine who enters later digital treatment studies
The investigators argue that these procedures may select participants before intervention exposure. In other words, the population that reaches the actual digital health intervention may already be filtered by the requirements of the research interface itself. People who can move through multiple digital tasks independently may be overrepresented, while those who need support may be undercounted, partially assisted, or excluded entirely.
That has direct implications for representativeness. If digital health interventions are evaluated mainly in participants who are comfortable with electronic systems, the resulting evidence may overstate usability and understate burden for groups with lower digital literacy. The concern is not theoretical. It affects who gets counted in the evidence base, whose outcomes are measured, and which patients are invisible before the intervention even starts.
The study also raises a second issue: participants who do continue may carry emotional burden into later use of the intervention. Frustration during onboarding is not erased once a person reaches the intervention phase. A participant who has already felt confused, dependent, or embarrassed may approach the digital tool with less confidence and less patience. For some, the research process itself may shape subsequent adherence and engagement.
Recording how completion occurs could change digital trial design
A more complete account of digital participation would document not just whether a participant finished, but how completion occurred. Independent completion, assisted completion, shared completion, and proxy completion are not interchangeable states. They represent different levels of autonomy, different risks of misunderstanding, and different degrees of burden on the participant and their network.
Capturing those distinctions could help researchers identify where digital research systems systematically exclude people. It could also guide the redesign of study pathways to make them recoverable and participant-controlled. If a participant loses their place, forgets a password, or becomes confused by a disclosure screen, the system should allow them to return without starting over or surrendering control to someone else.
For digital health interventions aimed at underserved groups, this is not a minor usability issue. It is a matter of who enters the evidence base in the first place. The study makes clear that the architecture of recruitment, consent, and electronic data capture can shape the final trial population as powerfully as the intervention itself. Any evaluation of differential intervention effects across disadvantaged groups will be incomplete unless researchers also examine the digital pathway that determined who was able to participate at all.