Why this work keeps a person in the room
Most of this job is judgment made face to face. A psychologist runs a diagnostic interview, decides what is going on, picks a treatment and then changes it when the first plan does not work. That loop depends on reading a pause, a flat tone, a story that shifts between sessions. Software can produce a careful-sounding reflection. It does not carry the license, the duty of care, or the call about whether someone is safe to go home.
The second reason is accountability. When a psychologist assesses suicide risk, consults with a physician, or signs an assessment report used in school placement or court, a named professional stands behind it. That is a legal and ethical structure, not a technical gap, and it moves slowly. Our task split gives the group of work that needs a person a share of 32%.
Demand matters too. The Bureau of Labor Statistics counts about 75,990 US jobs in this occupation, with employment projected to grow 11.7% from 2025 to 2035 and median pay of $100,580 (BLS, 2025). A growing occupation with a waitlist problem tends to absorb tools rather than shed people.
What AI drafts, what it assists, what stays with the clinician
The clearest wins are paperwork and scoring. Writing progress notes and session summaries, and scoring and summarizing standardized questionnaires, are the kind of structured, text-heavy tasks models already handle end to end. The share of task time in that group prints as 0%. None of it is the clinical decision; it is the typing around it.
A bigger slice is assisted work. Drafting a treatment plan, pulling together research on a condition or a protocol, and sorting intake information before a first appointment all go faster with a model, then get checked and rewritten by the clinician. That assisted group sits at 68% of task time. Our coverage score, which measures how much task time AI can handle today, reads 33 out of 100; how coverage is measured explains what counts.
What does not move is the session itself and the risk call. Delivering psychotherapy over months, and deciding whether a client needs hospitalization or a referral, both stay with the person holding the license. Supervising trainees and consulting with doctors, teachers or family members sit in the same place.
What the evidence does and does not show
There is no scored head-to-head test of AI against licensed psychologists on this job’s core work in our data yet. Our “Is it better than a person?” measure carries an evidence grade of D, and a grade at that level gets no parity number at all. Saying otherwise would be guessing.
What would settle it is specific: randomized trials comparing AI-delivered therapy with licensed clinicians on symptom change, dropout and relapse; blind review of AI-written assessment reports by supervising psychologists; and audits of how a system handles disclosed self-harm, abuse or psychosis. Until results like those exist and are published, the honest read is uncertainty on quality, not parity. You can see how we grade evidence on the quality parity page, and the full method sits at needsahuman.com/methodology.
When the picture could shift
Most likely between 2037 and 2050 (8 in 10 of our scenarios). For what that window measures and how it is built, see the replacement-year method.
Two things could pull it earlier. First, none of this job needs robots: the physical share of the work is 0, so there is no hardware barrier to clear. Second, the cost gap is wide, with annual software costs in our estimates running from $70 to $6,840 against $18,150 to $59,540 in human cost for comparable task time, which gives payers a reason to push screening and follow-up onto tools.
Two things hold it back. Licensure, liability and reimbursement rules all assume a named clinician, and those rules change slowly. And safety is unsettled: a system that mishandles one crisis disclosure can set back adoption across a whole service. Tracked movement in hiring and adoption shows up in our AI adoption tracker.
How to stay needed as a psychologist
Lean into the parts of the job that sit in the human-only group. Three worth building deliberately:
- Complex diagnosis, where the presenting problem is not the real problem and the interview has to go somewhere unplanned.
- Risk assessment and crisis decisions, including the documentation and consultation that go with them.
- Supervision and consultation: training clinicians, and advising schools, courts, medical teams and employers.
Two skills raise your value rather than your workload. One is measurement-based practice: using outcome data well enough to argue what is working. The other is tool oversight, meaning you can read a draft note or an AI screening summary and say exactly where it is wrong and why.
What to do: keep a short written record of the clinical calls a tool got wrong in your setting, so your service has evidence when it decides what to automate.
Nearby work scores differently because the task mix differs. Compare this page with Clinical Neuropsychologists, School Psychologists and Psychiatrists, or put any two side by side on our compare tool. The wider group sits in social scientists and related workers, and demand drivers for this role mostly come from healthcare. Our headline figure here reads 69 out of 100 (higher is safer); to see where that sits against every other job, open the full rankings.