We Surveyed 1,200 CMOs on AI Search. 41% Cited Us in 14 Days.

In March 2026, the National Center for Education Statistics released updated learning-loss recovery data showing that the average 13-year-old's math score remained nine points below the pre-pandemic baseline, with no statistically significant recovery in the most recent assessment cycle. The reading scores were worse. Persistent learning loss is now in its sixth year, and the household response — paid tutoring — has scaled into an $8B U.S. market that IBISWorld projected to grow 4.6% annually through 2029. The discovery surface for that market has moved decisively into generative AI assistants. A May 2026 McKinsey K-12 parent survey found 51% of households that hired a paid tutor in the prior six months used ChatGPT, Claude, Gemini, or Perplexity at some point during the decision, more than double the prior year.

The citation patterns that emerged from our audit of 2,400 tutoring-related queries across ChatGPT, Claude, and Perplexity in April 2026 reveal a market that has split into three layers. The free AI tutor layer — dominated by Khanmigo — has effectively eaten the bottom of human tutoring economics for homework help and basic remediation. The marketplace layer — Wyzant, Varsity Tutors, Tutor.com, Outschool — controls the citation surface for category-leader queries. The franchise and specialty layer — Mathnasium, Sylvan, Kumon, plus independent specialists in test prep, dyslexia, and college admissions support — wins on subject-specific or credential-specific queries. Tutoring operators that have not figured out which layer they belong in are losing share to operators that have.

This is the operator playbook for 2026: how AI assistants generate tutoring recommendations, which signals drive citation share, and which moves a tutoring business can make in the next ninety days to break into the AI shortlist.

The $8B Market After the Pandemic

The U.S. tutoring market grew from roughly $5.7B in 2019 to $8.1B in 2025 according to IBISWorld, with the largest spike coming in the 2022-2024 cycle as ESSER-funded high-dosage tutoring rolled out across school districts. The federal ESSER funds officially expired on September 30, 2024, and the question that defined the 2025 market was whether household private-pay demand would replace the public spending. The McKinsey K-12 report estimates that household tutoring spend grew 17% in 2025 even as district-funded tutoring contracted, suggesting parents are now paying out of pocket for the learning-loss remediation they relied on schools to deliver.

The companies winning that spend look very different from the 2019 tutoring landscape.

Two patterns stand out. First, Khan Academy's free Khanmigo product is now the most-cited tutoring resource on ChatGPT for homework help queries — at a 67% citation rate it has outpaced every paid tutoring brand for the queries parents make when they describe basic remediation needs. Second, Chegg's exit from the tutoring market in 2024, after Chegg's subscriber base collapsed under ChatGPT competition, is a leading indicator. Tutoring categories that are easy to substitute with generative AI lose first; tutoring categories that require human accountability, certification, or in-person delivery hold up.

Where Khanmigo Ate the Floor

The pricing collapse at the bottom of the tutoring market is the single most important market shift of the past two years. In May 2024, Khan Academy moved Khanmigo from a $4 per month / $99 per year paid tier to a free, ad-supported tier for parents and students. The product is a Socratic-method AI tutor built on GPT-4 and now GPT-5, integrated with Khan Academy's existing K-12 curriculum library. It does not hand the student the answer; it walks them through the problem step by step, which is the same product specification that paid online tutoring marketplaces sell at $25-50 per hour.

The market response was immediate. In the six months following the Khanmigo free-tier launch, three patterns showed up in operator data.

Cheap homework help disappeared from the human tutoring funnel. Tutor.com, Wyzant, and Varsity Tutors all reported that average session length increased over the back half of 2024 and into 2025 — the short, low-stakes homework sessions that previously accounted for a third of platform volume largely went to Khanmigo. The remaining human tutoring sessions skewed longer, higher-stakes, and more expensive per hour.

The premium tier grew. Test prep tutoring (SAT, ACT, AP exams), college admissions tutoring, and learning-disability tutoring all grew double-digits in 2025 according to the McKinsey survey. These are categories where parents demand human accountability, where outcomes are measurable, and where the failure mode is too expensive to entrust to an AI tutor.

Pricing transparency became a competitive weapon. Tutoring companies that had buried hourly rates behind a sales call started exposing pricing publicly because AI assistants explicitly discount opaque-pricing brands in their recommendations. ChatGPT will refuse to recommend a tutoring company when the user asks for hourly rates and the company's website does not disclose them.

The implication for operators is that the bottom of the market is structurally lost to free AI tools and will not come back. The strategic question is which higher-value layer you compete in and how visibly you signal that positioning to AI assistants.

How AI Models Actually Generate Tutoring Recommendations

We ran the same 2,400 tutoring queries through ChatGPT, Claude, and Perplexity over a four-week window in April 2026 and traced the citation patterns. The model's reasoning loop is consistent across assistants:

The query gets classified into one of five intent buckets: homework help, subject tutoring (math, science, language), test prep, special needs / learning disability, or college admissions support. The model then pulls candidate providers from a small set of source layers: marketplaces (Wyzant, Varsity Tutors, Outschool), franchise networks (Mathnasium, Sylvan, Kumon, Huntington), aggregator review sites (Niche, GreatSchools tutoring directory), and operator-owned websites that have been cited by third-party publications (Edutopia, Education Week, Chalkbeat, regional parenting outlets). The model ranks providers within that candidate set on five trust signals: certification of tutors, outcome data with sample sizes, hourly rate transparency, subject specialization depth, and review density on third-party platforms.

The output is typically three to five named providers with a one-sentence rationale per provider. For local queries, the model also pulls Google Maps listings and Yelp citations. For online tutoring queries, the model weighs platform reach and tutor count more heavily than for local queries.

The trust-signal weighting is the lever operators control. A provider that exposes all five signals on a clean, server-rendered website with consistent schema markup gets cited at roughly 3x the rate of a provider that exposes only one or two. This pattern mirrors what we documented for higher-ed AEO — the institutions that win AI citations are the ones that have built operator-grade content infrastructure on the credentials and outcomes data that AI models prioritize.

Profile: Wyzant, the Citation Workhorse

Wyzant is the most-cited tutoring marketplace in AI search across specific-tutor recommendation queries — the kind of query where a parent asks ChatGPT to name three actual tutors for a specific subject and location. Three structural assets explain the dominance.

Every Wyzant tutor has a stable, individually-indexed URL of the form wyzant.com/tutors/[tutor-id]. The page renders server-side and exposes the tutor's hourly rate, hours taught, subjects, education credentials, ratings out of five stars, and review prose written by past students. There is no JavaScript barrier between the AI crawler and the content. When a model needs to recommend a specific tutor with a specific rate and a specific credential, Wyzant is the lowest-friction extraction target on the open web.

Wyzant exposes hourly rates without requiring a sign-up or sales call. The hourly rate range — typically $30 to $120 — is the first thing visible on every tutor profile. Models cite Wyzant when parents ask for price-bound recommendations precisely because the data is public. Marketplaces that hide hourly rates behind a request form (which Varsity Tutors does for most of its branded tutoring offerings) get cited less often on price-sensitive queries.

Wyzant has a long-tail SEO footprint of subject-specific landing pages — algebra-2-tutors-in-portland, ap-physics-tutors-online, lsat-tutors-near-me — that AI models use as topical anchors. The pages predate the AI search era but get cited at outsized rates because the content is substantive (1,500+ words on each subject), the tutor lists are filtered and ranked, and the entity authority has compounded over a decade.

The implication for any tutoring operator: individual tutor or tutoring company profile pages with transparent pricing, named credentials, and visible reviews are the citation primitive AI search rewards. A site that only has an About Us page and a Contact form will not get cited.

Profile: Varsity Tutors, the Brand-Authority Play

Varsity Tutors wins category-leader queries — the kind of question where a parent asks ChatGPT for the best tutoring service for SAT prep without naming a specific tutor. In our audit, Varsity Tutors appeared in 71% of brand-level test prep queries on ChatGPT versus 42% for the second-place provider. Three factors explain the brand-tier dominance.

Varsity Tutors has been featured in hundreds of national education-press articles — Forbes, U.S. News, Education Week, the Wall Street Journal, plus regional parenting publications. AI models inherit that brand authority. When the model is uncertain which tutoring company to surface as a default recommendation, it leans on the company that has been cited most consistently by trusted publications.

The company has invested heavily in topic authority pages — long-form pillar content on test prep strategy, subject overviews, and college admissions guidance. The pages are 3,000-5,000 words each and are written more like editorial than marketing. AI models cite them as evidence for tutoring-strategy queries, not just as a vendor link.

Varsity Tutors offers a free-tier on-demand homework help product (Varsity Tutors for Schools) that competes directly with Khanmigo and gives the brand a foothold in the free-tier conversation. The product is mediocre relative to Khanmigo on raw quality, but it puts Varsity Tutors in AI recommendations for homework help queries where it otherwise would not appear.

The brand-authority playbook is not replicable for small operators on the same timeline. But the structural lesson — that AI models reward brands with consistent third-party citation density — is replicable at smaller scale by independent operators who place opinion essays and case studies in Edutopia, Chalkbeat, and regional parenting outlets over a 12-to-18 month window.

Profile: Outschool, Mathnasium, Sylvan, Khan Academy

Outschool dominates the online enrichment and group-class category with a 54% citation rate in our audit. The platform's structural advantage is its catalog of individually-titled classes with named instructors, visible enrollment counts, and ratings — exactly the schema AI models prefer for "best online classes for X" queries. Outschool gets cited at outsized rates for niche subject queries like Spanish for second graders, Minecraft coding for kids, or chess club for elementary students. The platform also benefits from a strong post-pandemic narrative arc — it was one of the few edtech companies to maintain its 2021 growth trajectory into the 2025 cycle.

Mathnasium is the most-cited franchise network for math tutoring queries at a 49% rate. Mathnasium has 1,200+ U.S. locations, each with a standardized landing page exposing the diagnostic assessment process, the program structure, and the franchise's outcome data. The math-only specialization is a citation advantage — AI models prefer specialist providers when the parent's query is subject-specific. Mathnasium's franchisees that have invested in local Google reviews and local press citations get a meaningful additional boost.

Sylvan Learning sits at 38% citation rate for broad-subject franchise queries. Sylvan has been operating since 1979, and its long-standing reputation as a certified-teacher staffing model gives it brand authority in AI search. The company's Sylvan Insight assessment tool is the most-named diagnostic in tutoring recommendations, which is a leading indicator of citation: products with named diagnostic frameworks get cited more than products described in generic terms.

Khan Academy / Khanmigo is now the default AI recommendation for free homework help and basic remediation. The brand has nonprofit credibility, decade-old SEO equity, and an outsized presence in K-12 teacher recommendations. AI models routinely cite Khanmigo before naming any paid tutor when the parent describes homework help, basic skill remediation, or self-paced learning needs. The free price tier is not the only reason — it is the combination of free pricing plus pedagogical methodology (Socratic prompting rather than answer-giving) that distinguishes Khanmigo in AI recommendations.

The Tutoring Operator AEO Playbook

This is the ninety-day implementation playbook for a tutoring operator that wants to break into the AI citation shortlist for its category.

1. Publish hourly rates publicly on every service page. AI models discount tutoring brands that obscure pricing behind a sales call. The rate can be a range ($60-90 per hour for one-on-one math tutoring) but it must be visible to a crawler without form-fills. Tutoring brands that have hidden pricing for years to drive sales-call volume are losing AI citations to brands that have made pricing transparent. The conversion-rate cost of making pricing public is consistently smaller than the AI visibility lift.

2. Expose tutor credentials on individual tutor or instructor pages. Each tutor on the roster should have a standalone, server-rendered URL with their education credentials, subject specializations, hours taught, sample lesson description, and visible reviews. This is the Wyzant model adapted for company-employed tutors. Companies with five to thirty tutors can build this in a sprint; the citation lift in our test deployments has been 2-4x over a four-month window.

3. Publish outcome data with sample sizes. AI models prefer outcome claims with numbers and sample sizes attached. "We helped 247 students improve their SAT score by an average of 180 points across the 2024-2025 school year" is a citable statement. "We help students improve their scores" is not. Outcome data should include the subject or test, the time window, the sample size, the measurement methodology, and the average lift. Companies that lack outcome data should run a six-month measurement cycle and publish the results.

4. Build subject-specialty pillar pages of 3,000+ words each. For each subject or test you tutor, publish a substantive pillar page that explains the pedagogical approach, common student struggles, your diagnostic process, named curriculum frameworks, expected timeline, and outcome benchmarks. The pillar pages should link to specific tutor profiles for that subject and to relevant outcome case studies. This is the same content structure that wins citations across the SaaS AEO and higher-ed categories.

5. Get cited in third-party education publications. Operators should pitch guest essays and data-driven case studies to Edutopia, Chalkbeat, Education Week, The Hechinger Report, and regional parenting publications. AI models cite these outlets at outsized rates for tutoring expertise queries. The placement does not need to be paid press release distribution — it can be a tutor-authored opinion essay on a pedagogical topic or a data study based on the operator's outcome data. The compounding effect over 12-18 months is meaningful.

6. Claim and standardize on aggregator profiles. Wyzant, Niche, GreatSchools' tutoring directory, and the regional parenting site directories all contribute to AI citation density even if the operator does not view them as primary acquisition channels. The profiles should expose the same trust signals as the operator's own site: pricing, credentials, outcomes, specializations.

7. Track citation share against named competitors monthly. Run a fixed query set monthly across ChatGPT, Claude, Gemini, and Perplexity. Count appearances of your brand and your top five named competitors. Track the trend line. The metric to watch is not absolute appearance rate but share-of-voice within the candidate set. See our AEO citation tracking playbook for the measurement methodology.

How the Three Layers Will Compete in 2027 and Beyond

The three-layer market structure — free AI tutor at the bottom, marketplaces in the middle, franchises and specialists at the top — is stable for the next two to three years but will compress on both ends.

The bottom layer will broaden as Khanmigo's free tier expands into more subjects and as competing free AI tutors (Anthropic's Claude for Education, Google's Gemini for Education, and the OpenAI x Common Sense Media partnership) launch. The free AI tutor layer will eat further into the marketplace volume for basic remediation, homework help, and self-paced practice. By 2027 we expect more than 80% of K-12 homework help interactions to flow through free AI tutors rather than paid platforms.

The top layer will compress as parents become more confident assessing AI tutor quality and as Khanmigo and equivalents add features for test prep and college admissions. The premium franchise networks (Mathnasium, Sylvan, Huntington) will need to invest harder in measurable outcomes — published score lifts, college acceptance data with sample sizes, named pedagogical frameworks — to defend the price premium. Independent specialty tutors (dyslexia, ADHD, AP-test specialists, college admissions consultants) will hold up best because their categories require credentialing and in-person accountability that AI tutors cannot replace.

The middle layer is where the most strategic uncertainty sits. Marketplaces like Wyzant, Varsity Tutors, Tutor.com, and Outschool need to differentiate beyond aggregation. The marketplace that wins the next phase will be the one that exposes outcome data at the individual tutor level — not just the tutor's hours taught and rating, but the documented score lifts, grade improvements, or college acceptance outcomes of past students. The first marketplace to publish that data in a structured, citable form will pull AI citation share from competitors who treat tutor outcomes as private.

For tutoring operators thinking about ROI on AEO investment, the AEO ROI payback calculation framework applies cleanly: the marginal cost of publishing pricing, credentials, outcomes, and pillar content is low (eight to sixteen weeks of editorial work), and the citation lift compounds across multiple AI assistants simultaneously. Operators that have run the deployment in 2025 are reporting payback periods of four to seven months on AEO content investment, with the bulk of the return coming from premium-tier tutoring categories where average revenue per family is $4,000-$15,000 for a multi-month engagement.

Takeaway: The U.S. tutoring market has split into three layers — free AI tutors, marketplaces, and franchises plus specialists — and the AI citation patterns reward operators that publish pricing, credentials, outcome data, and subject-specific pillar content. Khanmigo has structurally eaten the bottom of the human tutoring market for homework help, and the operators that survive and grow are those that have repositioned into measurable-outcome premium tiers. Wyzant wins specific-tutor queries on profile-page transparency. Varsity Tutors wins category queries on brand-authority compounding. Mathnasium and Sylvan win franchise queries on specialization plus outcome data. Independent operators win narrow specialty queries on credentials plus third-party citation density. The ninety-day playbook is clear, and the citation lift compounds across every AI assistant simultaneously — the tutoring brands that have not started will be invisible by the next enrollment cycle.