Teachers

Will AI Replace Teachers? Here's What the Research Actually Says.

May 2026

By Mark Reeves — author of all 47 AI Field Guides. Written and tested in a working business.

AI-generated portrait. Pen name. Real work.

Not hype, not reassurance — just what we actually know.

Start With the Honest Number

A middle school English teacher spends Sunday night writing 28 rubrics, drafting parent emails, and generating substitute lesson notes. By the time she sits down to actually plan instruction, it's 10pm. All three tasks are automatable right now.

About a third of AI experts (31%) predicted AI could lead to fewer teaching jobs, according to Pew Research Center's April 2025 survey of AI experts. I'm not going to pretend that's nothing. It's not nothing.

But the conversation usually stops there, at the scary headline, without getting into what actually drives that prediction — and what it means for the teachers who are doing the job right now.

What AI Can Replace (And Is Already Starting To)

Let's be specific about what's actually at risk.

AI is very good at tasks that are structured, repeatable, and text-based. Lesson plan templates. Rubric generation. Boilerplate IEP documentation. First-pass feedback on formulaic essays. Progress report language. Substitute planning notes.

These are real parts of a teacher's week. In some cases, they're a significant part. RAND's research has consistently shown teachers spend a significant portion of their week on tasks outside direct instruction — paperwork, planning, communication. A lot of that is automatable right now, with tools that exist today.

There's a useful comparison from medicine here. When AI tools were introduced to automate clinical note-taking for physicians, a 2025 study published in JAMA Network Open (Olson et al., Yale School of Medicine) found burnout dropped from 51.9% to 38.8% within 30 days. The work that disappeared wasn't the doctoring — it was the documentation overhead that had piled up on top of the doctoring.

The same pattern is likely in education. The question is who captures that freed-up time.

What AI Cannot Replace

This is where the honest reassurance actually lives — not in "AI can't replace human connection," which is vague, but in the specific things teachers do that AI is structurally bad at.

Reading a room. Noticing that a student who's usually engaged went quiet this week. Adjusting a lesson in real time when the class's confusion is different from what you planned for. Building the kind of trust with a student that makes them willing to say "I don't understand" instead of pretending.

These aren't soft skills. They're the adaptive, judgment-intensive core of the job. A language model doesn't have a room to read. It doesn't have the longitudinal relationship with a student that tells you this quietness is different from last month's quietness. It generates the next token based on what's statistically likely — it doesn't make the call a good teacher makes.

The relational and adaptive work isn't just hard for AI. It's architecturally outside what current AI systems do. That could change — but it would require a fundamentally different kind of AI than we have now.

The Risk That's Actually Real

Here's the version of this that should make teachers uncomfortable, because it's the honest one.

The risk isn't AI replacing teachers wholesale. The risk is that teachers who don't learn to use AI get replaced by teachers who do — or by a school system that decides fewer teachers can serve more students if the paperwork load drops far enough.

If AI can cut administrative overhead significantly, a district facing budget pressure has a calculation to make. That's not hypothetical. We've already seen technology used to justify larger class sizes, fewer support staff, and reduced positions when the argument can be made that productivity has increased.

The teachers most exposed to that calculus are the ones doing the most automatable work — lesson templates from scratch every week, repetitive documentation that could be generated in minutes, feedback on assignments that follow a predictable format — without also doing the high-value relational and adaptive work that makes a smaller number of them irreplaceable.

A teacher who is primarily an information deliverer is more replaceable than a teacher who is primarily a learning coach, a trust-builder, an early-warning system for a kid in trouble.

What the Evidence Suggests Teachers Should Do

Not "embrace AI" as a vague prescription. Specifically: use AI to reclaim the hours that are currently going to paperwork, and redirect them toward the irreplaceable work.

Use it to generate a first-draft rubric, then spend the time you saved giving a student real feedback in conversation. Use it to draft the IEP progress language, then spend the time you saved actually talking to the parent.

The teachers I find most credible on this aren't the ones saying "AI is amazing" or "AI is a threat." They're the ones quietly using it for the grind work and showing up with more capacity for the human work.

That's not a guarantee of job security. Nothing is. But it's the most defensible position I can identify based on what the research actually says.

What This Means Right Now

If you're a teacher reading this, you probably didn't need someone to tell you AI is coming. You're already dealing with it in your classroom, in the work your students are submitting, in the policy vacuum your administration hasn't filled yet.

What I hope this does is give you a clearer picture of where the actual risk lives — not in a robot standing at the front of your classroom, but in the slower structural shifts that happen when automation changes what a district thinks it needs.

The teachers who don't build these workflows are the ones a district can point to when they're justifying larger classes. That's the actual risk — not a robot at the front of the room, but a budget meeting where your hours are the argument.

Try This Before You Close This Tab

Open ChatGPT and type: "Draft a 4-criterion rubric for a persuasive essay, 9th grade, first attempt at the assignment."

Read what comes back. That rubric took 20 seconds. It needs your judgment on top of it — the criteria should reflect what you've been teaching, not a statistical average of all rubrics — but the blank page is gone. Now consider how many Sunday nights that represents.

That's the concrete version of the argument this article is making.

This article covers what the research actually shows about replacement risk. It doesn't cover three things you'll need once you've decided to act on it:

How to use ChatGPT for the specific paperwork tasks this article mentions. Lesson plans, rubric drafts, IEP language, parent emails — the guide covers the prompting approach for each, including what to do when the output sounds like it came from a textbook.

How to handle AI in your classroom — policy, detection, and assessment redesign. Students are using these tools. The guide covers how to build a response that's defensible without criminalizing AI use that's genuinely ambiguous.

What to do when your school hasn't built an AI policy yet. Most teachers are absorbing every judgment call without institutional backing. That's the structural problem underneath every individual decision.

Related reading

• A student used AI and you can't prove it. Now what?
• You're already behind on AI. Here's how far — and how to close the gap.

Mark Reeves is a pen name. AI Field Guide publishes role-specific, practical guides for using AI tools in real work.