The Parent Funnel Fork: How AI Homework Tools Are Splitting Tutoring Demand

Two Funnels, One Age Group

By the time a parent asks their teenager how homework went, there's a good chance the answer runs through a chatbot the parent has never heard mentioned at the dinner table. Two things are true about the tutoring and homework-help market at the same time right now, and most coverage of AI in education picks one and misses the other.

The first: AI chatbot use for schoolwork has gone from a curiosity to the norm among US teens in about two years, and the businesses built to answer homework questions one at a time, priced per subscription, are losing ground fast. The second: paid, human-delivered tutoring, the kind a parent researches, calls, and pays for directly, has not shrunk to make room. School-funded high-dosage tutoring kept growing even after pandemic relief funding expired. The overall tutoring industry's revenue estimate held steady in the high teens of billions of dollars.

Those two facts only look contradictory if you assume AI homework help and paid tutoring are competing for the same purchase decision. They aren't. That's the fork this study traces: not one market shrinking while another grows, but two different buying processes running in parallel, one largely invisible to the person who'd normally be the buyer.

How Fast Teens Adopted This

The adoption numbers are, at this point, well established rather than speculative. Pew Research Center, surveying 1,458 US teens ages 13 to 17 in late 2025, found that 54% have used an AI chatbot to help with schoolwork, and 10% say they do all or most of their schoolwork with AI help. That is a national, peer-reviewable survey with a published margin of error, not a vendor's marketing claim.

College Board's research brief, drawing on its own survey program through mid-2025, found an even higher and still-rising figure among high schoolers specifically: 84% reported using generative AI for schoolwork as of May 2025, up from 79% just four months earlier in January. Sixty-nine percent named ChatGPT by name. Two independent, methodologically transparent surveys, run by different organizations, landing on the same basic picture: a majority of US teens now use AI chatbots for schoolwork, and the share is still climbing month over month.

One figure worth handling with more care is the often-cited 86% of students globally using AI in their studies, from the Digital Education Council's 2024 global survey. That number is real, but it describes university-level students worldwide, not US K-12 students, and it measures general study-related AI use, not homework help specifically. It's a useful signal that the underlying behavior is global and not confined to any one country's education system, but it should not be quoted as a US teen or K-12 statistic. We tier it emerging, on the strength of a single, if large, international survey.

A Gap Most Parents Don't See

Here is the finding that gives this study its name. Common Sense Media surveyed 1,045 teens and 1,045 parents in parallel in 2024 and found that among teens who use generative AI, 53% say they use it most often for help with homework. So far, unsurprising. But when the same organization asked parents whether they knew their teen had used generative AI, only 37% of parents whose teen actually had used it said yes.

That's not a small awareness gap, it's a majority of parents in the dark about a behavior that's already the norm for their kids. And it reframes what the AI-homework-help layer of this market actually is: not a product category with a buyer and a purchase funnel, but something closer to a private habit, adopted directly by the student, funded by nothing (most of the tools are free), and largely invisible to the person who would traditionally be the one making a purchasing decision on the child's behalf.

We tier this finding emerging rather than established, since it rests on a single survey without independent replication. But it's consistent with the broader pattern in the adoption data: teens are the ones doing the choosing here, not parents.

Only 37 out of every 100 parents whose teenager had used a generative AI tool for schoolwork actually knew it had happened.

The Commodity Layer Buckles

If there's a canary in this market, it's Chegg. The company built its business on exactly the transaction generative AI chatbots now perform for free or near-free: a student has a homework question, and gets an answer. Chegg's full-year 2024 revenue fell 14% year over year to $617.6 million, its fourth-quarter 2024 revenue fell 24% year over year, and its subscriber base fell 21% year over year to 3.6 million. Those are the company's own reported earnings figures, as covered by EdTech Innovation Hub, and they describe a business in real, ongoing decline, not a one-quarter dip.

At the same time, tools built on the same underlying technology but distributed free or cheap are scaling quickly. Khan Academy's Khanmigo grew from about 68,000 users to more than 700,000 within its first year after its March 2023 launch, reaching partnerships with over 380 school districts. That figure is self-reported by Khan Academy and hasn't been independently audited, so we tier it contested, but directionally it tells the same story as Chegg's decline told in reverse: when a homework-help transaction can be answered for free by a general-purpose chatbot, or by a free tool built for that exact use case, the paid, single-purpose incumbent loses ground.

This is the part of the fork that looks the most like classic technology substitution, a scaled free alternative displacing a paid one for the same narrow task. It's also the part that, on its own, would support the wrong conclusion: that AI is simply eating the tutoring market. The next section is why that conclusion doesn't hold.

The Premium Layer Isn't Going Anywhere

IBISWorld estimates the US Tutoring & Driving Schools industry generated $18.9 billion in revenue in 2025, growing at a 2.6% compound annual rate over the prior five years, across roughly 176,000 establishments. We flag this as emerging rather than established, and as directional rather than precise, because tutoring-market sizing varies by roughly 40 times across research vendors depending on how each one defines scope: online-only versus all channels, tutoring-only versus tutoring-plus-driving-schools bundled together, current revenue versus forward-looking growth forecasts. The $18.9 billion to $20 billion range is the most defensible current-year anchor available, not a precise figure to build a forecast on.

What's more telling than any single market-size number is what happened when a major funding source disappeared. States had dedicated an estimated $4.2 billion in pandemic-era ESSER relief funds to tutoring and accelerated learning. That funding cliff hit on September 30, 2024, and only 23 states have any ongoing funding mechanism to replace it, Tennessee is the only one with a formula-funded program. Despite that, high-dosage tutoring offerings in US public schools grew about 5% between 2024 and 2025, per Stanford's National Student Support Accelerator analysis. As of October 2024, 37% of US public schools offered high-dosage tutoring specifically, and 78% offered some form of tutoring, according to the federal government's own School Pulse Panel survey, a figure we treat as established given its source and methodology.

That's demand surviving a funding cut, which is a stronger signal than demand simply existing. A market that shrinks the moment free money disappears is a market propped up by the money. A market that keeps growing after the money disappears is a market with underlying demand. At the far premium end, the college admissions consulting market, valued by one lesser-known research vendor at $7.2 billion globally in 2025 with individual packages running from roughly $200 to as much as $500,000, points in the same direction, though we tier that single-vendor figure contested and would not lean on it heavily on its own.

Why: What Practice Scores Hide

The most rigorous piece of evidence in this study isn't a market-size figure at all, it's a peer-reviewed field experiment. Researchers led by Bastani and colleagues, published in the Proceedings of the National Academy of Sciences, gave roughly 1,000 Turkish high school math students either unrestricted access to a GPT-4-based math tutor or no such access, and measured both practice-problem performance and closed-book exam performance. Students with unsupervised AI access scored roughly 48% higher on practice problems, but roughly 17% lower on the closed-book exam that followed, relative to students without AI access.

That's a real, measured mechanism, not a hunch: unsupervised AI help can make a student feel and appear more capable while they're doing the work, without producing durable learning that shows up once the AI is taken away. It's the closest thing in this evidence base to a causal, not just correlational, explanation for why a parent who understood what was happening might choose to keep paying for outcome-verified human tutoring even as free AI homework help becomes universal among teens.

Two caveats belong right next to that finding. First, the study is set in Turkish public high school math classrooms, and no comparably rigorous, US-specific study exists yet testing AI tutoring against human tutoring in an American K-12 population, so the exact magnitude may not transfer directly. Second, it measures one subject, math, under one specific condition, unrestricted, unsupervised access, not AI tutoring in general or AI tutoring designed with guardrails. What it does establish, credibly, is that the assumption that AI homework help is a good substitute for human tutoring is not something to take on faith.

Unsupervised AI access raised practice-problem scores by roughly 48%, and lowered closed-book exam scores by roughly 17%, in the same students.

Two Purchases, Not Two Market Segments

Put the pieces together and the sharper version of this story isn't that the tutoring market is splitting into an AI segment and a human segment. It's that two entirely different buying processes are running at the same time, for the same age group, and they barely interact.

On one side: a teen, alone with a device, adopts a free or nearly free AI chatbot for homework help, without a purchase decision, without a sales process, and largely without their parent's knowledge. That's the layer collapsing Chegg-style incumbents and scaling free tools like Khanmigo. On the other side: a parent, initiating and funding the decision themselves, researches, evaluates, and pays for a human tutor, an admissions consultant, or a school-funded high-dosage program, because they're buying a verified outcome, not just an answer to tonight's assignment.

The practical implication for a business in this category is that you are not competing with a general-purpose chatbot for the same customer. You're competing for a parent's attention and trust, in a purchase decision the teen isn't the one making, against other paid options and against a parent's own uncertainty about whether paying is even necessary. The PNAS finding matters here specifically because it gives a parent a concrete, evidence-backed reason, not just a marketing claim, that the free option their teen is already using may not be producing the learning they think it is.

You are not competing with a free chatbot for this customer. You are competing for a parent's attention in a decision their teenager isn't making.

The Unmeasured Frontier

One layer of this thesis is real in concept and genuinely not yet measured: whether AI answer engines like ChatGPT, Perplexity, Google AI Overviews, or Gemini function as a discovery or recommendation surface when a parent researches tutoring options. As of this study, no independent, methodologically transparent census exists on that question. Any claim about a tutoring company's visibility inside the answers those engines give would be speculation dressed as data, so we're not making one.

What that means practically is that the parent-facing side of this market still runs largely through the channels it always has: search, referral, school relationships, and reputation, with AI-engine discovery as an emerging but unproven surface worth watching rather than a channel you can already build a strategy around. Given that teens are already living inside AI interfaces for their half of this fork, and parents are the ones actually initiating and funding the premium purchase, the near-term opportunity sits squarely on winning visibility with the parent, across the surfaces where their research and decision-making actually happens today. That's the terrain worth mapping before you assume you already know where it is.