■ MODERATE RISK ■ Healthcare
No — AI will take over the measurement side of speech therapy while leaving the therapy itself firmly human. Getting a frustrated four-year-old to attempt the /r/ sound for the fortieth time is not a signal-processing problem.
“AI speech analysis is precise, but therapy still needs human patience.”
Our AI replacement risk score — how we score jobs
A speech-language pathologist's week runs from articulation drills with preschoolers to swallowing evaluations for stroke patients, with a mountain of insurance documentation in between. The clinical toolkit is oddly physical: mirrors, tongue depressors, picture cards, and in medical settings, videofluoroscopic swallow studies read frame by frame. Caseloads in schools routinely hit fifty-plus students, which means the real scarce resource is minutes of skilled attention, not diagnostic accuracy.
AI is already good at the parts SLPs never loved. Acoustic analysis software can score articulation errors, track vocal pitch and jitter, and flag disfluencies more consistently than a tired human ear at 3 p.m. Speech-recognition-based practice apps let kids drill at home between sessions and report compliance back to the clinician. Report writing, session notes, and Medicaid billing narratives — the paperwork that eats a third of the job — are exactly what large language models chew through. Expect screening and progress measurement to become largely automated, with the SLP validating rather than administering.
What resists is the therapeutic core. Therapy works through rapport: a child who trusts you will attempt sounds they've been humiliated for; an aphasia patient grieving their lost fluency needs a clinician, not a metronome with a dashboard. Feeding and swallowing work carries genuine medical risk and liability that no vendor wants to own. And much of the job is negotiating with parents, teachers, and IEP committees — messy, political, human work. Our risk score of 35 reflects a role that gets a powerful measurement co-pilot and loses its paperwork, not its purpose.
Automatability: our editorial assessment of current and near-term AI capability
Through the late 2020s, expect AI to absorb screenings, progress tracking, and documentation while caseloads stay stubbornly human. By the mid-2030s, hybrid models — app-based drilling supervised by fewer clinicians handling more complex cases — will be standard in schools and teletherapy. The job transforms into a diagnostician-supervisor role rather than disappearing; demand from aging populations with post-stroke and neurodegenerative needs actually pushes the other way.
Yes, with eyes open. Demand is growing thanks to aging populations, autism diagnoses, and post-stroke rehab, and the hands-on therapy core resists automation. But the job is shifting: screening, drilling, and paperwork are being automated, so the value concentrates in complex clinical judgment and human rapport. Go in planning to work alongside AI tools, not to out-measure them.
They can replace some of the repetition, not the therapy. Speech-recognition apps are genuinely useful for home practice between sessions and for keeping data on progress. What they can't do is build the trust that gets a reluctant child or a grieving stroke survivor to keep trying, adapt mid-session to frustration, or manage swallowing disorders where mistakes are medically dangerous.
The measurement and paperwork. Acoustic analysis already scores articulation and voice parameters more consistently than the human ear, and language models are very good at drafting evaluation reports, session notes, and billing documentation. Screenings will increasingly be app-based with clinician review. That's arguably a gift — documentation is the part of the job most SLPs would pay to lose.
Move toward the medically complex end: dysphagia, AAC, neurogenic disorders — areas where risk and liability keep humans essential. Master the digital practice tools so you're the one prescribing and supervising them. And lean into the coordination work with families and schools, which is politically messy in a way no algorithm volunteers for.