Research

Teachers, platforms and AI, by the research

You build it. Who keeps it?

Our headline asks a question: what if you owned it? This piece is the research behind it. It covers who makes the value on a course platform, who keeps it, why AI raises the stakes, and what research says happens when people own part of their work.

One of the largest online course marketplaces

The shrinking slice: teachers’ share of its subscription revenue

15 cents of each subscription dollar goes to the teachers who make the courses, down from 50¢ before 2018.

That’s a slice 70% smaller, cut in four steps.

The 2018 cut applied to the marketplace’s business subscriptions: its own letter to instructors dates it to 2018, correcting an earlier version that said 2019,3 and Class Central reports it as 50% to 25%.4 The next three cuts were announced in November 2023 and took effect each January from 2024 to 2026.5 Columns are drawn to scale.

You build it

Everything a learner finds was made by someone who teaches.

And who teaches matters. A study of more than one million children found that students assigned to more effective teachers were more likely to go to college and less likely to have children as teenagers.1

1,000,000+

children in one study of what teachers change
  • More likely to go to college
  • Less likely to have children as teenagers

Students assigned to more effective teachers, followed into adulthood.1

Someone else keeps it

The platform sets the terms, and it can change them.

Teachers make the courses. One of the largest online course marketplaces paid instructors 50% of the sales it brought in itself until 2021. That year it cut the share to 37%.2 Instructors’ share of its subscription revenue has fallen further: from 50% to 25% in 2018,3, 4 then in three more steps to 15%.5

The teacher’s share, before and after the cuts
  • Course sales the marketplace brings in itself, before 2021

    50% to the teacher2

  • The same sales, from 2021

    37% to the teacher2

  • Subscription revenue, before 2018

    50% to the teacher34

  • Subscription revenue, from January 2026

    15% to the teacher5

TeacherPlatform

AI raises the stakes

AI learns from what people make. The work people make is the anchor.

Trained over and over on its own output

In a 2024 study, models trained over and over on the output of the model before them developed defects that didn’t go away, and the rare parts of the original material disappeared first.6

Illustration of the effect, not the study’s data: the rare cases at the edges go first.

  • No collapse

    when researchers kept the original human-made data and added the machine-made data alongside it.7

  • 28%+

    of the most actively maintained sources in a widely used collection of web text became fully restricted from use for AI in one year, from 2023 to 2024. That anchor is getting harder to reach.8

  • 14,000

    websites in that audit.8

So the people who keep making new lessons, examples and explanations matter more, not less.

What the research says

What happens when people own part of their work

The best-studied case is employees who own part of the company they work for. The evidence is real, and it’s modest.

  • 102

    samples, about 57,000 firms

    A meta-analysis found that employee ownership goes with a small but positive gain in company performance.9

  • 76%

    as likely to disappear

    US public companies whose employees owned 5% or more were only 76% as likely as other companies to disappear between 1988 and 2001.10

  • 2001 & 2008

    recessions

    Companies with employee ownership kept jobs steadier through both.11

  • Lower

    staff turnover

    It works best alongside a real say in how the work is done. Firms that paired shared rewards with employee involvement had lower staff turnover.12

The limits matter just as much

  1. The effects are small

    Companies that choose employee ownership differ from those that don’t, so some of the gain may come from who chooses it.

  2. Bigger isn’t automatically better

    One large study found the clearest gains came from small stakes. Large ones had smaller effects on average, and were often used to replace pay or to block takeovers.13

  3. Grants can delay leaving

    Stock grants can delay people leaving rather than stop them.14

  4. Nobody has studied our case yet

    All of these studies are about employees. We found no study yet of what ownership does for creators or teachers working online, and none of a platform owned mostly by its users.

Why it’s a question

Paying the people who use a platform in ownership is new ground.

  1. November 2020

    The US Securities and Exchange Commission proposed a temporary rule that would have let platform companies pay workers partly in stock: no more than 15% of a worker’s pay in a year, and no more than $75,000 over three years.15

  2. May 2021

    It was withdrawn.16

  3. Today

    The rules for paying a platform’s users in ownership aren’t settled yet.

That’s why our headline asks a question instead of making a promise.

What we’re building toward

One lesson, many teachers.

ClimbHigh AI is building a course platform where AI maps each lesson and many teachers teach it in their own voice. The model we’re building toward reserves 67% of company ownership for creators, earned as pay for their teaching.

Proposed model

Where this stands: This is the model we’re designing, not a final program. Nothing is final until counsel-approved agreements are signed, and every creator sees every term in writing before signing. Joining the waitlist is free and doesn’t earn, reserve or guarantee any ownership.

Ownership awards are compensation for content contributions, subject to program terms, written agreements, eligibility, and applicable law, and may vary or be unavailable in some countries. This is not an offer to sell or a solicitation to buy securities.

Sources and method

  • 70% smaller = 1 − 15% ÷ 50%, the subscription share from January 2026 against the share before 2018.
  • Company figures are from the marketplace’s own letters and terms; studies are cited by their published versions.
  • The charts name no platform; the sources below identify the documents. Every link and figure was checked on November 19, 2026.
  1. Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014). Measuring the impacts of teachers II: teacher value-added and student outcomes in adulthood. American Economic Review, 104(9), 2633–2679. doi.org/10.1257/aer.104.9.2633
  2. Udemy. Business terms update, March 2021. The share for sales not driven by an instructor's own promotion became 37%. teach.udemy.com/business-terms-update-march-2021/
  3. Udemy. Updating subscription terms to enable key investments and business growth, November 2, 2023. Dates the earlier cut to the Udemy Business revenue share to 2018, correcting an earlier version that said 2019. teach.udemy.com/enabling-investment-subscription-terms-update/
  4. Shah, D. Udemy earns more, pays instructors less. Class Central, February 17, 2025 (updated February 27, 2025). Reports the earlier cut as 50% to 25%, dated 2019. www.classcentral.com/report/udemy-broken-promise-instructor-payouts/
  5. Udemy. Letter to instructors, filed with the SEC as Exhibit 99.1, November 2, 2023. The subscription revenue share falls to 15% in January 2026. www.sec.gov/Archives/edgar/data/1607939/000160793923000181/exhibit991-xinstructorle.htm
  6. Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., & Gal, Y. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759. doi.org/10.1038/s41586-024-07566-y
  7. Gerstgrasser, M., Schaeffer, R., Dey, A., et al. (2024). Is model collapse inevitable? Breaking the curse of recursion by accumulating real and synthetic data. arXiv preprint 2404.01413. doi.org/10.48550/arXiv.2404.01413
  8. Longpre, S., Mahari, R., Lee, A., Lund, C., et al. (2024). Consent in crisis: the rapid decline of the AI data commons. Advances in Neural Information Processing Systems 37 (NeurIPS 2024). doi.org/10.52202/079017-3431
  9. O'Boyle, E. H., Patel, P. C., & Gonzalez-Mulé, E. (2016). Employee ownership and firm performance: a meta-analysis. Human Resource Management Journal, 26(4), 425–448. doi.org/10.1111/1748-8583.12115
  10. Park, R., Kruse, D., & Sesil, J. (2004). Does employee ownership enhance firm survival? Advances in the Economic Analysis of Participatory and Labor-Managed Firms, 8, 3–33. doi.org/10.1016/s0885-3339(04)08001-9
  11. Kurtulus, F. A., & Kruse, D. L. (2018). An empirical analysis of the relationship between employee ownership and employment stability in the US: 1999–2011. British Journal of Industrial Relations, 56(2), 245–291. doi.org/10.1111/bjir.12254
  12. Blasi, J., Freeman, R., & Kruse, D. (2016). Do broad-based employee ownership, profit sharing and stock options help the best firms do even better? British Journal of Industrial Relations, 54(1), 55–82. doi.org/10.1111/bjir.12135
  13. Kim, E. H., & Ouimet, P. (2014). Broad-based employee stock ownership: motives and outcomes. Journal of Finance, 69(3), 1273–1319. doi.org/10.1111/jofi.12150
  14. Aldatmaz, S., Ouimet, P., & Van Wesep, E. D. (2018). The option to quit: the effect of employee stock options on turnover. Journal of Financial Economics, 127(1), 136–151. doi.org/10.1016/j.jfineco.2017.10.007
  15. U.S. Securities and Exchange Commission. Temporary rules to include certain "platform workers" in compensatory offerings under Rule 701 and Form S-8. Proposed rule, Release No. 33-10892, November 24, 2020. www.sec.gov/rules/proposed/2020/33-10892.pdf
  16. Office of Information and Regulatory Affairs. Unified Agenda, Spring 2021, RIN 3235-AM79. Listed as withdrawn, May 11, 2021. www.reginfo.gov/public/do/eAgendaViewRule?pubId=202104&RIN=3235-AM79