Why the work keeps a person in it
Will AI replace community health workers? Look at the task mix, not the tools. Much of this job happens at someone’s door: a home visit, a conversation about why a prescription sat unfilled for three weeks, a ride arranged to a clinic that opens at 7 a.m. A model can draft the reminder text. It cannot earn the trust that makes a second visit possible.
The other anchor is advocacy. Community health workers speak for clients inside systems that were not built for them — a benefits office, a landlord, a billing department, a clinic that needs a translation in both language and culture. That work runs on relationships and local knowledge, and it changes with every family.
Some pieces do move. Case notes, intake logs, program reports, grant data, flyers and outreach scripts are all text, and text is where current systems are strongest. The honest story is task erosion rather than a job going away: the clerical layer thins while the visits stay. You can see how that logic turns into numbers on our methodology page.
What software can take, assist with, or leave alone
Start with the tasks AI can run with little supervision. Writing up visit notes, pulling program numbers into a report, and drafting health education material in plain language sit here. Share of task time in that group: 0%.
Then the assisted middle, where a person still leads and the tool saves time. Translating a handout, building a call list from screening data, summarizing a client’s history before a visit, flagging who missed an appointment. Our split puts the assisted share at 54%. Across every task together, the coverage score for this job — how much task time AI can handle today — comes out at 29 out of 100.
The third group explains the headline answer above. Home visits, crisis support, building trust with people who have good reasons to distrust institutions, and physically getting someone to care are not text problems. Share of the job that still needs a person: 46%.
What has actually been tested
Very little, in this job specifically. Our evidence grade for quality parity — is AI better than a person? — comes out as D. A D grade means there is no direct test of AI against community health workers on their own tasks, so we publish no parity number here. See how the grades work on the quality parity method page.
What would settle it is a field trial: AI-assisted outreach against standard outreach with comparable clients, measured on outcomes people in the job already track. Appointments kept. Screenings completed. Follow-up held over months, not one call. Chatbot studies on general health questions do not answer that question, because the hard part of this job is not answering questions.
Official data gives the other half of the picture. BLS counted about 61,660 community health workers in the United States, with median pay of $51,850, and projects 12.7% employment growth from 2025 to 2035 (BLS, 2025). Funding decides how many of these roles exist, so budget cycles can move headcount faster than any model does.
When this could shift
Most likely between 2036 and 2052 (8 in 10 of our scenarios). That window is wide because two forces pull against each other.
Two things could pull it earlier. First, cost: the yearly software figures on this page are small next to a salary, so a squeezed program may push reminders, screening follow-ups and basic education into automated messaging. Second, channel drift. Where outreach is already done by phone and text, a model can carry more of the exchange, and entry-level openings are usually the first to thin.
Two things hold it back. Part of this job is physical — traveling to homes, delivering supplies, accompanying someone to an appointment — and the robotics tier that would be needed for that is mobile robots, which are nowhere near doing it in a stranger’s apartment. Consent and privacy around health data in a community setting is the other brake: programs answer to funders and public agencies before they answer to a vendor. The replacement-year method explains how we build the range itself.
How to stay needed in this job
Lean into the tasks in the needs-a-person group. Field visits and door-to-door outreach, where presence is the service. Advocacy and system navigation, where you argue a case nobody else will. Trust and crisis work with clients who have disengaged from care, which is slow by nature and hard to hand off.
Two skills raise your floor. One is interviewing that changes behavior — motivational interviewing, de-escalation, working with interpreters. The other is plain data fluency: knowing what your program reports, and using documentation tools well enough that notes and reports stop eating your afternoons. People who can show outcomes get kept when budgets tighten.
What to do: pick one recurring report or note template this month and let a tool draft it, then spend the saved hour on a visit you keep postponing.
If you are weighing a nearby role, the closest work sits with health education specialists, social and human service assistants and healthcare social workers. You can put any two side by side on our compare tool, see the wider community and social service family, check pressure across the healthcare sector, or scan the jobs that most need a person.