Choosing the Right AI Channel for Patient Engagement

Why AI-Assisted SMS Often Offers a Better Path to Patient Action Than AI Voice

An independent, use-case-based assessment for healthcare technology buyers

Joshua Lowentritt, MD, FASN, FNKF

Chief Medical Officer,  Generated Health

Past President, Orleans Parish Medical Society

Healthcare organizations are rapidly adopting artificial intelligence to automate patient outreach, close quality gaps, improve appointment completion, and extend care-team capacity. Two increasingly common approaches are conversational AI voice agents and AI-assisted text messaging. Both can automate outreach and interpret patient responses, but they are not interchangeable technologies. The communication channel itself can determine whether a patient answers, remains engaged, and ultimately completes the desired healthcare action.

Recent healthcare and technology research points to several principles that should guide technology buyers. Successful patient-facing AI depends not simply on how sophisticated an agent sounds, but on trust, convenience, accessibility, workflow integration, clinical governance, and measurable patient outcomes. For population-health programs—such as annual wellness visits, well-child visits, cancer screening, chronic-disease monitoring, and post-discharge follow-up—AI-assisted SMS has important structural advantages.

Text messaging is asynchronous, persistent, familiar, and relatively low-friction. Patients can engage when convenient rather than at the precise moment a call arrives. AI voice, by contrast, requires patients to answer an incoming call, rapidly establish trust in the caller, and remain available for a synchronous conversation. Voice can be valuable for selected clinical situations, but it may be a less effective first-line channel for large-scale outreach to people who have already been difficult to reach.

A de-identified real-world implementation by one large health plan illustrates this distinction. A proprietary AI-assisted SMS solution generated engagement rates approximately three to four times those observed by the same organization with AI voice outreach. Across three SMS pilots involving nearly 50,000 members, 19% engaged with the platform. Beyond initial responses, the program was associated with higher completion of adult wellness and well-child visits, answered thousands of member questions, identified social and logistical barriers, and produced actionable information for care teams. These observational results do not establish universal superiority, but they provide a useful case study in how channel design can influence patient action.

The AI Agent Is Only Part of the Engagement Strategy

Much of today’s healthcare AI discussion focuses on what an agent can do: understand natural language, personalize a conversation, answer questions, schedule appointments, or transfer a patient to a care-team representative. Those capabilities are important. But they come after a more fundamental question: Will the patient remain in the interaction long enough for the technology to help?

Industry research increasingly emphasizes the need to reduce friction across the healthcare consumer journey and provide personalized, multimodal communication. Deloitte has identified patient engagement, administrative simplification, and workforce productivity as major opportunities for AI. McKinsey has similarly emphasized continuity across channels, while Salesforce has described AI agents as a way to handle routine interactions so healthcare staff can focus on people who require human judgment or more complex support.

These findings support AI-enabled engagement broadly, but they do not establish that voice is the optimal channel. Buyers should evaluate the entire pathway from successful contact to completed clinical action. A compelling voice demonstration may show an agent conducting a remarkably natural conversation. Yet if most targeted patients do not answer—or disconnect shortly after realizing they are speaking with an automated system—the sophistication of the downstream conversation has limited practical value.

Why AI Voice Can Struggle With Unsolicited Outreach

AI voice has genuine strengths. It can support immediate dialogue, clarify questions in real time, and create a more conversational experience than traditional interactive voice response systems. Published studies have demonstrated promising results in narrowly defined clinical situations. In one small randomized trial, conversational AI voice helped patients with type 2 diabetes complete repeated insulin-titration tasks and improved medication adherence. That type of use case matters, but it differs substantially from population-level outreach to people overdue for preventive services.

In care-gap outreach, the patient frequently did not request the interaction. The caller may be unknown. The patient may be driving, working, caring for children, or otherwise unable to talk. Automated calls must also overcome increasing consumer skepticism toward unfamiliar numbers, robocalls, fraud, and synthetic voices. Before a voice agent can create clinical or operational value, the patient must answer, remain on the line, quickly understand who is calling and why, and be available to converse at that moment. Speech recognition must also interpret the patient accurately despite accents, background noise, hearing impairment, or language variation. If escalation is required, the patient and a care-team representative may need to be available simultaneously.

Trust is especially consequential in healthcare. Recent reports from Accenture and EY/MIT Technology Review Insights emphasize that patients expect privacy safeguards, clear disclosure of AI’s role, transparency about data use, and appropriate human oversight. Trust tends to decrease as AI assumes activities patients perceive as more human or clinical. These concerns do not make voice inappropriate; they increase the importance of matching the channel to the task and setting.

In the de-identified health-plan experience reviewed for this article, an AI voice pilot targeted mammography and cervical cancer screening. Members often answered but disconnected before meaningful engagement occurred, and hang-ups were reported as the most common outcome. The vendor’s principal performance measure was successful transfer to a care-team representative. Transfers can be useful, but they are an intermediate operational metric. They do not necessarily show that a patient scheduled a test, attended a visit, or closed a quality gap.

Why AI-Assisted SMS Changes the Engagement Equation

SMS starts from a different premise: the patient does not have to reorganize the day around an outreach attempt. A message arrives and remains available. The recipient can read it immediately or later, check a work schedule, consult a family member, confirm whether an appointment already exists, or gather information before replying.

Practical advantages of asynchronous messaging

•  Lower interaction burden. A patient who cannot take an unexpected five-minute call may still be able to reply “yes,” “already scheduled,” “need a doctor,” or “call me tomorrow” in seconds.

•  Persistence over time. A voice call disappears when it ends. A text remains on the phone, can serve as a reminder, and allows outreach to continue across days or weeks.

•  Two-way problem solving. Patients can report transportation problems, insurance issues, scheduling conflicts, language needs, confusion about preventive care, or difficulty reaching a clinician. The platform can categorize those responses and escalate when human intervention is needed.

•  Scalable education. Routine questions can be answered without requiring a care manager to return every call. Patients can ask what a well-child visit includes, why an annual wellness visit matters, or how to prepare for an appointment.

•  Structured operational data. Conversational SMS can convert natural-language replies into usable categories such as appointment scheduled, service already completed, needs new primary care clinician, requests callback, reports transportation barrier, or needs rescheduling.

These advantages matter because engagement is not simply a response. It is movement from contact to understanding to action. Systematic reviews of digital interventions have found that text messaging can improve participation in preventive services, although effects vary by population, message design, and comparison condition.

A De-Identified Case Example: From Outreach to Patient Action

Across three pilots conducted by one large health plan, 49,973 members were enrolled in protocols delivered through a proprietary AI-assisted SMS platform; 9,365 members engaged, producing an overall engagement rate of 19%.

Among people specifically considered “lost” or previously difficult to reach, an adult annual wellness visit program achieved a 10.1% deduplicated engagement rate. An early well-child visit cohort achieved approximately 7% deduplicated engagement while still progressing through a 42-day outreach sequence.

Engagement also extended to measurable clinical action. Among adults with wellness-visit gaps, SMS-engaged members closed 58.9% of identified gaps, compared with 50.0% among non-engaged members—an 8.9-percentage-point difference. For well-child visits, engaged families closed 79.3% of gaps, compared with 74.5% among non-engaged families—a 4.8-percentage-point difference. Because the comparison was observational rather than randomized, the results should be interpreted as an association; people who engage may differ from those who do not.

The system also answered 4,277 questions from 2,844 unique members. Topics included appointment scheduling and confirmation, cancellations, insurance, transportation and social barriers, requests for human help, preventive-care education, and clinical concerns. In a separate chronic-care pilot involving hypertension and diabetes, participants responded to 59% of requests for blood-pressure or glucose readings.

This case example illustrates why buyers should avoid defining engagement only as calls placed, calls answered, or transfers completed. A clinically meaningful funnel is: eligible population → outreach delivered → meaningful two-way engagement → barrier or intent identified → appointment scheduled → visit completed → quality gap closed. Vendors should report the same denominator and outcome definition at every stage.

Matching the Channel to the Use Case

AI voice and AI-assisted SMS should not be viewed as mutually exclusive. Voice may be preferable when a patient expects the interaction, needs immediate clarification, has difficulty reading or texting, or requires a nuanced synchronous conversation. It may also be useful after a text exchange identifies a high-risk concern or a preference for spoken assistance.

For high-volume population-health outreach, SMS is often the stronger first-line modality. Relevant applications include preventive-care outreach, cancer screening, medication adherence, chronic-disease check-ins, collection of blood-pressure or glucose readings, emergency department and hospital follow-up, appointment reminders and preparation, and re-engagement of patients disconnected from primary care.

A practical sequence is SMS first, voice or human intervention second. SMS can establish contact, determine intent, resolve routine questions, and identify which patients need synchronous assistance. Voice or staff outreach can then be directed toward the smaller group for whom real-time discussion adds value. This channel orchestration may improve patient convenience while preserving scarce care-team capacity.

What Healthcare Buyers Should Measure

Technology selection should be based on standardized outcomes rather than conversational realism or outreach volume. At minimum, buyers should compare meaningful engagement, appointment scheduling, completed visits, closed quality gaps, opt-outs, complaints, staff workload generated, performance by language and demographic group, and cost per completed clinical action.

Clinical governance also matters. Buyers should determine whether outbound clinical content is based on reviewed protocols, how patient intent is interpreted, when escalation occurs, whether the system can preserve context over time, and what role human clinicians have in oversight. A platform that generates more responses but overwhelms staff with unstructured follow-up may simply move the bottleneck. The goal is not maximal automation; it is reliable patient action with appropriate human attention.

Procurement teams should also insist on comparable pilots. Voice and SMS cohorts should target similar populations, care gaps, and outreach periods, with pre-specified definitions of delivery, meaningful engagement, and completion. Results should separate people who were newly activated from those who already had an appointment or completed service. Reporting should include failed contacts, hang-ups, opt-outs, safety escalations, and unresolved messages—not only successful conversations. Whenever feasible, randomized or well-matched comparison groups should be used. This discipline helps prevent a polished demonstration or a favorable denominator from being mistaken for evidence of clinical effectiveness.

Conclusion

AI voice and AI-assisted SMS both have legitimate roles in healthcare. Voice can work well for expected, focused interactions that benefit from immediate dialogue. It should not, however, be assumed to be the best channel simply because it sounds more human.

For large-scale outreach to patients who are overdue for preventive care, disconnected from primary care, or managing chronic conditions between visits, asynchronous conversational SMS offers several structural advantages: lower interaction burden, persistence, convenience, scalable education, structured barrier identification, and a direct path from outreach to measurable action.

Published evidence and the de-identified health-plan experience described here support a use-case-specific conclusion: when the primary challenge is initiating and sustaining engagement across a broad population, AI-assisted SMS should generally be considered the preferred first-line technology, with voice and human outreach reserved for patients and situations that benefit from synchronous conversation.

Selected Sources

•  Accenture. Technology Vision 2025: Healthcare perspectives on trust, transparency, privacy, and human oversight in AI-enabled care.

•  Cohen-Cline H, et al. Direct-to-member interactive voice response outreach for colorectal cancer screening. Medical Care. 2014;52(5):451–457.

•  Deloitte. 2025 Global Health Care Executive Outlook. Deloitte Insights; 2025.

•  EY and MIT Technology Review Insights. Powering Next-Generation Services With AI in Regulated Industries. 2025.

•  Liu Y, et al. Electronic interventions to improve cervical cancer screening participation: systematic review and meta-analysis. 2024.

•  Marcotte LM, et al. Automated telephone outreach strategies for mammography completion. JAMA Internal Medicine. 2023.

•  McKinsey & Company. Engaging the evolving US healthcare consumer and improving business performance.

•  Milne-Ives M, et al. The effectiveness of artificial intelligence conversational agents in health care: systematic review. Journal of Medical Internet Research. 2020;22(10):e20346.

•  Nayak A, et al. Conversational AI for basal insulin initiation and titration in adults with type 2 diabetes: randomized clinical evaluation. 2023.

•  Salesforce. State of Service: Healthcare and Life Sciences Edition. 2025.