Industries / Government, education and community / Education / Tutoring
Online tutoring
Tutoring over the internet or by phone, with tutor and student in different places, sometimes in different countries.
6
Milestones achieved, from 1998 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1998
Homework help through libraries
George Cigale started Tutor.com in 1998 to connect students one to one with a live tutor over the internet. Public and academic libraries bought annual subscriptions and offered its Live Homework Help free to their communities; in 2006 Cigale said it worked with more than 1,500 libraries in over 40 US states, expected to serve more than a million students through them that school year, and used more than 1,300 background-checked tutors.
Credit
George Cigale, Tutor.com
2016
Remote tutors, no measured gain
In 2014–15 Third Space Learning gave Year 6 pupils in English primary schools weekly one-to-one maths lessons over the internet from maths graduates working in India and Sri Lanka. In a randomized trial with 64 schools and 600 pupils over 27 weeks, funded by the Education Endowment Foundation (EEF), independent evaluators at the University of York found no evidence of an effect on the national Key Stage 2 maths test; the program cost £378 per pupil.
LessonTeachers were largely positive about the lessons, but that did not show in test scores; the evaluators noted the trial ran while the company was at an early stage.
Credit
Third Space Learning; evaluators Carole Torgerson, David Torgerson and colleagues
2020
Tutoring by phone call
During COVID-19 school closures, a trial with 4,550 Botswana households compared weekly text messages of number problems with the same texts plus a 15 to 20 minute weekly phone call walking a child through them. The calls plus texts raised numeracy by 0.12 standard deviations after 3 hours of teaching over 8 weeks, which the authors put at 0.63 to 0.89 standard deviations per US$100. Follow-up trials in India, Kenya, Nepal, the Philippines and Uganda found average effects of 0.30 to 0.35 standard deviations, with government teachers as effective as nonprofit staff.
LessonThe text messages alone had no measurable effect; the live call did the work.
Credit
Youth Impact (Moitshepi Matsheng); Noam Angrist and Peter Bergman
2020
Student volunteers tutor online
Italy's Tutoring Online Program matched university students, as unpaid volunteers, with disadvantaged middle-school students for lessons held entirely online. Random assignment showed that three hours of one-to-one tutoring a week raised math scores by 0.23 standard deviations in 2020 and 0.20 in 2022; group tutoring had smaller effects.
LessonResults differ by design; a US pilot that paired college volunteers with middle-school students online found positive but statistically insignificant effects of 0.07 standard deviations in math and 0.04 in reading (Kraft, List, Livingston and Sadoff, 2022).
Credit
Michela Carlana and Eliana La Ferrara
2021
Cross-border lessons shut off
VIPKid, a Beijing company started in 2013, sold live one-to-one English lessons from native speakers, many of them American, to children in China, and had about 100,000 teachers at its peak. After China's State Council banned hiring tutors based abroad in July 2021, VIPKid stopped selling those classes in August, and from 5 November 2021 students in mainland China could no longer take them; 51Talk, with about 30,000 Filipino teachers, halted its children's courses, and ByteDance's GoGoKid cancelled all classes abruptly.
LessonA market of cross-border tutors depended on one government's rules, and families' prepaid packages, some booked a year ahead, were cut short.
Credit
VIPKid; China's State Council (the rules)
2024
Owner's ties end contracts
In March 2024 Santa Rosa County Schools in Florida cancelled its contract with Tutor.com after parents raised concerns about the company's control by Primavera, a firm owned by Chinese nationals, and Florida's education commissioner told school districts and colleges not to contract with companies tied to "foreign countries of concern". Tutor.com said no student information was shared with Primavera; it now describes itself as 100% US-owned and operated.
LessonFor an online tutor that holds children's data, who owns the company can decide whether schools may use it.
Credit
Santa Rosa County School District; Florida Department of Education
Watch for
A spoken AI maths tutor
Third Space Learning now sells Skye, an AI maths tutor that pupils talk to. It reports that Educate Ventures Research, which it calls independent, reviewed session data and found pupils answered 30% of check-in questions correctly at the start of a session and 70% of check-out questions at the end; the reviewers also found that pupils learning English as an additional language, and pupils whose speech the system does not pick up reliably, had more trouble using it.
Achieved whenA trial with a comparison group finds gains on a test taken outside the sessions, such as the Key Stage 2 maths test.
Claimed by
Third Space Learning, reporting Educate Ventures Research, September 2026
Open
01
Planted errors for speaking tutors
A repeatable test of paid AI conversation tutors for language learners. Consenting adult volunteers, or synthetic voices, speak scripted lines with planted mistakes in grammar, word choice and pronunciation; the test records which mistakes each tutor notices, whether its correction is right, and whether it "corrects" things that were said correctly.
The briefCredit · unclaimed02
Prepaid lessons lost to closures
A public ledger of online tutoring and coaching companies that have closed, stopped a service or entered insolvency since 2020, recording for each what happened to families' prepaid lesson packages (refunded, moved to another service, or lost), with a source for every row. No data about individual students.
The briefCredit · unclaimed
Open 01 · the brief
Planted errors for speaking tutors
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable test of paid AI conversation tutors for language learners. Consenting adult volunteers, or synthetic voices, speak scripted lines with planted mistakes in grammar, word choice and pronunciation; the test records which mistakes each tutor notices, whether its correction is right, and whether it "corrects" things that were said correctly.
Why it is still open
The studies found compare a few tools on a handful of recordings, not a count of planted errors across products. A 2025 Hanoi University study compared feedback from ChatGPT-4o, ELSA Speech Analyzer and three trainee teachers on one student's IELTS speaking test, and found ChatGPT missed pronunciation entirely and "did not identify audio input as human speech". Cambridge researchers' work on correcting spoken grammar tests research systems on learners' natural errors, not the tutors people pay for.
Built with what exists
Text-to-speech in several accents, such as ElevenLabs; speech-to-text to log the tutors' replies, such as ElevenLabs Scribe v3 (see the Note "Fewer word errors"); a script of lines with known errors, sorted by an error list from second-language research; subscriptions to the tutors tested.
First step
Write 30 lines with one planted error each and run them by hand through two tutors.
Done looks like
300 planted errors run twice through at least five paid tutors; for each tutor, the share of errors caught and correctly fixed and the number of false corrections, with scripts and recordings published.
Open 02 · the brief
Prepaid lessons lost to closures
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public ledger of online tutoring and coaching companies that have closed, stopped a service or entered insolvency since 2020, recording for each what happened to families' prepaid lesson packages (refunded, moved to another service, or lost), with a source for every row. No data about individual students.
Why it is still open
Packages were cut short when Chinese platforms ended foreign-taught classes in 2021, and GoGoKid cancelled its classes from one day to the next. Byju's, an Indian learning company once valued at $22 billion, was put into insolvency proceedings in July 2024, with creditors invited to file claims. This research found no record, across companies and countries, of what families got back.
Built with what exists
An AI model that reads court and regulator filings and news in Chinese, Hindi and English, such as Claude or Gemini; orders of India's National Company Law Tribunal; consumer-complaint statistics where regulators publish them; a spreadsheet with a source link per row.
First step
Code the ten best-documented closures, starting with VIPKid, GoGoKid and Byju's.
Done looks like
At least 40 closures in at least five countries, each with its date, the number of customers affected where a source gives one, and the outcome for prepaid fees, every row checked by a second person.