Why the work stays in the room
Will AI replace secondary school teachers? Not in the main, and the reason is the shape of the day rather than the quality of the software. A high school teacher spends most of it with a live class: establishing and enforcing rules for behavior, adapting the explanation when half the room looks lost, and deciding in the moment whether to push a student or back off. Those judgments depend on knowing the fourteen-year-old in the third row, not on knowing the subject.
The paperwork around that is a different story. Preparing lesson plans, writing worksheets and quizzes, and grading written assignments are text tasks, and text is what language models do best. That is where the erosion shows up first: not a missing teacher, but a shorter evening.
Two more parts of the job resist handoff. Conferring with parents and guardians about a struggling student carries accountability a tool cannot hold. And preparing the classroom and supervising students in hallways, labs and field trips is physical presence. Our robotics read puts that side of the job in the dexterous humanoid tier, which is hardware that does not exist at school-district prices. Schools are also slow buyers by design; the wider picture for the schools sector is adoption through approved tools, not replacement of staff lines.
What AI drafts, what it assists, and what stays with the teacher
Our task split sorts this job’s work into three groups. The share AI can run on its own comes out at 8% of task time, and it sits on the document side of teaching: first-draft lesson plans aligned to a standard, practice question banks, reading-level rewrites of the same passage, and rubric-based marking of short written answers. A teacher still checks the draft before it reaches students.
The assist group is larger in practice. It covers 38% of task time: pulling patterns out of assessment results, suggesting which students need reteaching on a specific skill, drafting progress notes for a parent meeting, and translating a letter home. The teacher sets the goal and keeps the call; the tool shortens the typing. How we size this group is set out in how coverage is measured.
What is left needs a person, and it is the bulk of the role at 54% of task time: managing behavior in a room of thirty, running a discussion where a wrong answer has to be handled kindly, supervising a lab, and noticing that a usually reliable student has stopped turning things in. Teenagers also behave differently when an adult with authority is present. That is not a feature a model can supply.
What the evidence shows, and what is missing
This page carries an evidence grade of D, which means no published study has yet tested an AI system against a qualified secondary teacher doing the whole job. Tutoring trials and grading comparisons exist for narrow slices of it, but a slice is not a class period. Because the grade is at the lowest level, we publish no quality-parity number here: putting one up would imply a measurement nobody has made. How that grading works is explained in our quality-parity method.
What would settle it is specific: a controlled study over a full term comparing AI-led instruction with a teacher-led class on learning gains in the same subject, plus attendance, behavior incidents and completion rates, in ordinary public schools rather than volunteer pilots. Until that exists, the honest answer rests on the task mix above.
Labor market data is firmer. The Bureau of Labor Statistics counts about 1,065,210 jobs in this occupation, with median pay of $72,040, and projects employment roughly flat to 2035, a change of -0.2% over 2025 to 2035 (BLS, 2025). Enrollment and retirement drive that line far more than software does.
When the picture could change
Most likely between 2035 and 2049 (8 in 10 of our scenarios). What that window measures is set out in how we forecast replacement years, and the date is a median with a spread, not a prediction about any one school.
Two things could pull it earlier. Long-running teacher shortages in math, science and special-subject classrooms push districts toward blended models where one adult supervises more students with software doing the instruction. And if assessment moves fully onto platforms that grade, track and reteach automatically, the planning and marking load that currently justifies staffing could shrink.
Two things hold it back. Student supervision carries legal duty of care that sits with a licensed adult, and that is set in state law rather than in a product roadmap. Physical oversight is the other brake: labs, workshops, corridors and trips need a body in the space, and the hardware tier required for that is nowhere near classroom cost. Our headline figure for this job is 72 out of 100 (higher is safer), and you can see how every score is built on the methodology page.
How to stay needed
Lean into the parts of the job that stay with people. Classroom management and relationship-building with hard-to-reach students are the first. Live assessment, the running judgment about whether a class has actually understood something, is the second. Working with parents and guardians on a plan for one student is the third, because it needs accountability and follow-through.
Two skills raise your value either way. The first is designing assessment that survives AI use at home: oral defense, in-class writing, process evidence, project checkpoints. The second is using AI tools well enough to cut your own planning and marking hours, and to teach students where the output is wrong. The AI skills employers ask for overlap heavily with that.
Good to know: the task list above is the practical guide to where your time is safest, and it is more useful than any single headline figure.
Neighboring roles sit on similar ground and are worth a look if you are weighing a move: middle school teachers, special education teachers in secondary schools, and career and technical education teachers. You can see the wider group on the teachers job family page, put any two roles side by side with the job comparison tool, or see which roles depend most on presence and judgment in our list of jobs least exposed to AI.