AI Is Splitting Tutoring in Two. Here's Which Side You're On

Two tutors. Same year. One is watching his business collapse in real time.

He's not a newcomer. He's been one of Wyzant's top-ranked tutors in his subject for years, commanding a high hourly rate, averaging around $10,000 a month. Then, last October, something shifted. By his own accounting, revenue fell roughly 75% — not because his reviews dropped, not because the platform changed its algorithm, but because his students started asking ChatGPT first. "I'm left with the conclusion that this is almost certainly due to ChatGPT," he wrote in a post on Reddit's tutoring forum, "and is the new state of affairs."

Four thousand miles away in Didsbury, Manchester, Phil Birchenall had a different October. He's an AI trainer by profession, and one evening his daughter Daisy was stuck on a Year 5 maths problem. Rather than sit through a homework argument, he built a custom ChatGPT tutor for her — one that explained fractions and word problems in the voice of their family's cocker spaniel, Izzy. Daisy, who idolizes that dog, engaged immediately. She worked through her revision. She smashed her exams. OpenAI later turned the story into a film on their homepage.

Same technology. Same window of time. Completely different outcomes. The difference wasn't credentials or luck. It was which part of the tutoring job each person was doing. To understand why, you need to see what's actually happening to the tutor market right now — at the institutional level, where the numbers are hardest to argue with.

The Disruption Is Real, and It's Already Institutional

Chegg was the dominant homework-help platform for a decade. On October 27, 2025, it announced 388 layoffs — 45% of its entire workforce — explicitly blaming "the new realities of AI" and diminished Google search traffic. This followed a 22% workforce cut just five months earlier in May. Roughly 57% of Chegg's employees were gone within a single calendar year. The company's stock, which peaked near $113 in 2021, has since lost 99% of its value.

The mechanism is worth understanding, because it applies beyond Chegg. Google's AI-powered search now answers the homework question directly, before the student clicks through to any platform. ChatGPT does the same. Chegg's business was built on being the fastest, cheapest source of content explanation on demand. That value proposition evaporated the moment free AI tools could do the same thing, instantly, from any browser.

The official labor data tells a similar story, at lower volume. The U.S. Bureau of Labor Statistics projects just 1% growth in tutor employment from 2024 to 2034 — slower than the all-occupations average of 3.1%. Median tutor pay sits at $40,090 a year. Net new jobs over the entire decade: roughly 1,300. Meanwhile, the AI-tutoring market itself is valued at $2.1 billion in 2025 and is projected to reach $17.7 billion by 2033.

That last number isn't a contradiction. It's the point. The disruption compressing human-tutor employment is the same force funding an explosion in AI-tutor products. The market is bifurcating — and the jobs that look most like Chegg's business model are the most exposed, regardless of the individual tutor's performance ranking.

For anyone tutoring through a platform whose core promise is "fast content explanation on demand," this data is a direct signal about structural viability, not a distant abstraction. The Wyzant tutor was still top-ranked. His income fell 75% anyway.

Which Tasks Are Automating — and Which Are Gaining Value

Not all tutoring is the same job. Three categories of tutoring tasks are moving in opposite directions right now, and which category dominates your practice largely determines your exposure.

The first category is automating now. Homework explanation, practice-problem generation, initial concept walkthroughs — these are the tasks where students already have a free, always-available alternative. One six-year math and economics tutor on Reddit described the shift precisely: students now ask ChatGPT to "generate 20 practice problems on whatever topic he's weak on, and then ask it to grade and explain" their mistakes — before contacting their human tutor. This is exactly the task category that drove the anonymous Wyzant tutor's 75% decline, and it's the category Chegg's entire business was built on. If your sessions are primarily structured around explaining content a student could find elsewhere, the substitution has likely already begun.

The second category is shifting to higher value. Socratic questioning, misconception diagnosis, motivation scaffolding, and exam strategy are becoming more defensible, not less. A 2025 Carnegie Mellon study of more than 350 seventh graders found that students receiving human plus AI tutoring ended the year 0.36 grade levels ahead of students receiving AI tutoring alone. The human contribution in that study wasn't content delivery — it was the motivational scaffolding and misconception correction that AI tutors generated errors in and human tutors caught. That 0.36 grade-level gap is where the defensible tutor practice lives.

Looking back, being suspended from Columbia for creating Interview Coder was a pivotal moment. It forced me to step outside the traditional path and burned all other bridges except entrepreneurship.
— Chungin "Roy" Lee, CEO of Cluely

The third category is protected for now. Accountability, oral language practice, high-stakes coaching, and special-needs support remain difficult to automate in any meaningful sense. This is where Birchenall's story lands. He didn't compete with ChatGPT on content. He wrapped the technology in something ChatGPT couldn't supply on its own: knowledge of his daughter, a relationship with her anxiety about maths, and the judgment to know that Izzy the spaniel was the hook that would get her to open the worksheet. That kind of situational reading isn't in the software.

This framework applies across every subject. A music theory tutor who spends most sessions explaining chord structure is more exposed than one who spends most sessions coaching performance anxiety before a recital. The automating category is defined by task type — explain, generate, summarize — not by subject.

The immediate audit question: in your last five tutoring sessions, how many minutes did you spend delivering explanations a student could have gotten from a search or a chatbot? That ratio is a first-order estimate of your exposure — and it's actionable, not fixed.

But knowing the framework is necessary without being sufficient. There's a temptation — especially after reading about Carnegie Mellon data and hybrid-tutoring models — to assume the augmentation argument is just optimism dressed up in research. The person best positioned to correct that assumption is the one who had the most to gain from overstating it.

The Most Honest Case Doesn't Come from Critics of AI

In April 2026, Sal Khan — founder of Khan Academy and architect of Khanmigo, the most prominent AI tutor in education — told Chalkbeat that for most students, Khanmigo's impact has been "a non-event." His Chief Learning Officer Kristen DiCerbo was specific about why: students "struggle to ask the AI the right questions." A revolution in education, she said, has not yet occurred.

The engagement data explains the gap. Only about 15% of students with Khanmigo access actually use the tool, even though the platform has logged more than 108 million total interactions since 2023. The AI tutor exists. Most students, most of the time, don't consistently choose it.

Tools that claim they can detect AI writing are snake oil. They have 40 to 50 percent false positive rates.
— Sal Khan, Founder of Khan Academy

The constraint on AI tutoring expansion isn't technical. It's human. Students need motivation to open the tool, persistence to phrase their confusion as a question, and accountability to return when they'd rather quit. None of that is in the software. Khan's concession isn't a reason to relax — it's a precise diagnosis of where human tutors remain structurally necessary.

This reframes the augmentation argument entirely. It's not "don't worry, AI can't replace the human touch." It's that the data shows AI fails at the point of engagement — and engagement is the skill tutors should be sharpening, documenting, and pricing. The ESL tutor who keeps a student showing up every Thursday despite a full-time job. The STEM tutor who can tell when a student is about to quit and adjusts in real time. The AI tutor can't do either of those things. The question is whether the human tutor is explicitly building their practice around doing them.

Which Side Are You On

The Wyzant tutor and Phil Birchenall weren't separated by credentials — his were excellent — or by luck, or even by how much they knew about AI. They were separated by which problem they were solving. One was selling explanations. The other was solving engagement. Those are different businesses now, and they have different futures.

The research from Carnegie Mellon and the admission from Sal Khan point to the same conclusion: AI tutoring fails at the moment a student stops showing up, stops asking questions, or stops trying. That moment — the human-shaped gap — is where skilled tutors will live or die professionally over the next five years.

Here's a concrete audit you can run today, with no tools required. Take your last five tutoring sessions. For each one, mark every minute you spent explaining something a student could have gotten from a search or a chatbot. Then mark every minute you spent diagnosing a misconception, keeping a student from quitting, holding them to a commitment they'd made, or adjusting your approach in real time because you read the room. The ratio you get is a first-order estimate of your exposure. If the first category dominates, that's not a verdict. It's a direction. The second category is where you build.

AI hasn't made tutoring less valuable. It's made the wrong kind of tutoring very easy to skip.

Recommended Tools & Resources