Why tutoring stays close to a person
Tutoring is not mainly about having the right answer. It is about finding the one explanation that lands for this student, on this afternoon, after a bad test. That is the core of the work: explaining a concept a second and third way, teaching study and test-taking habits, then watching whether any of it sticks. Software can supply explanations endlessly. The judgment about which explanation to try next, and when to stop explaining and rebuild confidence, sits with the tutor.
The second reason is accountability. A tutor checks progress, adjusts the plan, and reports back to parents and classroom teachers. Someone has to answer for whether a student is improving. Parents hire a person partly for that answer. A chat tool responds to the question a student asks; much of tutoring is spent on the student who will not ask, who guesses, or who quietly skips the hard part.
Market conditions matter too. The Bureau of Labor Statistics counts about 175,070 tutors in the US, with median pay near $43,350, and projects employment roughly flat, about -0.2% from 2025 to 2035 (BLS). None of the work needs machinery, so hardware cost is not a barrier here either way. That makes this a software question, not a robotics one.
What AI handles, assists with, and leaves to tutors
The paperwork side is where tools already carry the load. Generating practice sets at a set difficulty, drafting session notes, and pulling a progress summary together are now quick jobs for a language model. Our read of the task time AI can handle largely on its own for this job is 14%.
A larger slice is shared work. Explaining a topic, suggesting worked examples, and building a study or revision plan all go faster with a draft the tutor then edits for the student in front of them. Assisted task time comes out at 49%. The pattern is familiar across education roles: the prep shrinks, the session does not.
What is left is the part parents are actually paying for. Reading a discouraged student and changing tack mid-session, holding a teenager to a schedule, and negotiating goals with parents and teachers stay with the tutor. Task time that still needs a person is 37%. The headline Still needs a human score for tutors is 66 out of 100 (higher is safer), and you can read how that figure is built on the Still needs a human page.
What has been tested, and what has not
Here the honest answer is thin. Our evidence grade for quality parity in this job is D, which means no study in our evidence set has tested an AI tutor head to head against a trained human tutor working with the same students toward the same goals. Because of that, we publish no parity number for tutors at all. Plenty of trials test tutoring tools against no tutoring, or against ordinary classwork. That is a different question.
What would settle it is specific: a randomized trial over a full term, comparing AI-only tutoring with trained human tutoring and with a blended version of both, measuring learning gains, attendance and drop-off, with results published in full. Until something like that exists, claims in either direction are opinion. Our grading scale and what each letter requires are set out under quality parity.
Coverage is on firmer ground, because it asks a narrower question: how much of the job’s task time can AI handle today? For tutors that comes to 39 out of 100, built from the task list above rather than from any single study. The method behind it is on the coverage page.
When the picture could shift
Most likely between 2034 and 2046 (8 in 10 of our scenarios). We do not restate what that window measures here; the replacement-year method explains how the range is produced and why it is wide.
Two things could pull it earlier. Software is cheap next to a person’s hourly rate, so homework-help subscriptions can undercut private tutoring on price alone. And district or university purchases move fast once a tool is approved, which puts a free or low-cost option in front of students who would otherwise hire someone.
Two things hold it back. Trust is the first: parents and schools want a named adult responsible for a child’s progress, and a chat log does not carry that weight. The second is the relational half of the job, which is in-person or at least live, and does not get easier as models improve at explaining. Our general approach is set out in the methodology.
How tutors stay needed
Lean into the tasks the tools do worst. Diagnose the misunderstanding rather than answering the question asked. Own progress reporting to parents and teachers, with specifics and next steps. And take the students who need pacing, structure and someone to show up for, not just content.
Two skills are worth real time. First, assessment: writing and reading diagnostics so you can say exactly where a student is stuck. Second, using AI tools well in prep, so you arrive with better materials than a student could generate alone, and can explain honestly what the tool is good for.
What to do: Pick one subject where you can show measured gains over a term, and make that your pitch.
Close roles are worth a look if you are weighing options: adult basic education and ESL instructors, self-enrichment teachers, and short-term substitute teachers. The wider other teachers and instructors family and the education sector page show how the scores move across teaching work.
Next step: put tutoring side by side with one of those roles on the compare tool, or see where teaching jobs sit among the jobs that mostly need a person and search the full set in the rankings.