Industries / Government, education and community / Education / Tutoring
Matching students and tutors
Finding a tutor for a student, checking that the tutor is safe and suitable, and deciding which tutor teaches which student.
6
Milestones achieved, from 1955 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1955
A worksheet center in Osaka
In 1954 Toru Kumon, a high school math teacher, wrote calculation problems on loose-leaf paper for his son, and in 1955 the first math center using his worksheets opened in Moriguchi, Osaka. Families no longer had to find a personal tutor: local instructors ran centers on the same graded worksheets. The first center outside Japan opened in New York in 1974, and by 2008 Kumon said it served over 4 million students in 45 countries through franchises, a figure the education researcher Mark Bray reports as Kumon's own claim.
Credit
Toru Kumon; Kumon Institute of Education
2005
Tutors picked from profiles
Wyzant, a US marketplace running since 2005, let families browse tutors' profiles and reviews, message them and book pay-as-you-go lessons, with the first hour free if the student is not satisfied; IXL Learning agreed to buy it in February 2021. Marketplaces of this kind are paid from tutors' fees. Preply keeps 100% of the fee for each trial lesson with a new student and 18% to 33% of later lessons, the share falling as a tutor's total hours on the platform rise.
Credit
Wyzant (chief executive Andrew Geant); IXL Learning; Preply
2022
England drops its tutor broker
England's National Tutoring Programme (NTP), begun in November 2020, first had schools hire tutors from an approved list of "tuition partners", organizations that supply tutors, managed in 2021/22 by the contractor Randstad. That year tuition-partner courses reached 45% of the Department for Education's target, while school-led tutoring, added in September 2021 because the logistics of outside providers put schools off, made up 81% of the 2.14 million courses started, by the National Audit Office's count. For 2022/23 the department did not renew Randstad's contract and paid all tutoring money to schools directly.
LessonSchools chose tutors they could arrange themselves over a central list; in NFER's survey, the share of schools using tuition partners fell from 39% in the first year to 18% in the fourth.
Credit
National Audit Office; Department for Education
2023
Drop-in tutoring sits unused
Many US school systems bought online on-demand tutoring, in which a student who logs in is matched with whichever tutor is free, usually a different one each session. The Hechinger Report found some systems reporting usage below 2%; at the partner schools of UPchieve, a free service run by volunteer tutors, 5% of students logged in at least once in 2022-23, its leader said. A Mathematica study of UPchieve with 89 students found a math gain that was not statistically significant, and a co-author warned against putting much stock in the number.
LessonWhen matching waits for the student to ask, the students who need help most ask least, as a 2022 Brown University study of 7,000 charter school students found.
Credit
Jill Barshay, The Hechinger Report; Greg Chojnacki and colleagues, Mathematica
2026
Parents told to check tutors
In England, tutors who work privately are not required to hold an enhanced Disclosure and Barring Service (DBS) check, the criminal record check that school staff need, so parents do their own vetting, as the school principal Mark Steed wrote in Tes in June 2024. Guidance from the DBS, live from 21 January 2026 and naming tutors as an example, says a self-employed person can obtain an enhanced check on themselves if eligible and a parent may ask to see the original certificate, but a parent cannot apply on the tutor's behalf. In Scotland, by contrast, self-employed people in "regulated roles" with children must by law join the Protecting Vulnerable Groups scheme, the NSPCC notes.
Credit
Disclosure and Barring Service; Department for Education
2026
Female tutors lift girls' math
Researchers randomly assigned 422 ninth-grade Algebra 1 students tutored by the nonprofit Saga Education to a tutor of the same or the opposite gender. Girls with female tutors reported higher interest in science and math, by 0.73 standard deviations (a measure of how spread out scores are), and were 3.9 percentage points more likely to pass the course with a C- or better; there was no effect on attendance, and the effects were stronger in person than online. The authors call it the first experimental evidence that female tutors can lift girls' view of themselves in science and math and their results.
Credit
J. Bleiberg, C. D. Robinson, E. Bennett and S. Loeb, American Educational Research Journal (2026); Saga Education
Watch for
Matching handed to software
Nerdy, the owner of Varsity Tutors, says its full-year 2025 headcount fell 22% through automation in tutor matching, tutor substitution, customer service and scheduling, and that new tutor incentives brought faster first sessions and lower replacement rates. Its third-quarter 2026 report shows active tutors down 17% from a year earlier, which it attributes partly to steering learners to its highest-quality tutors.
Achieved whenAn outside party publishes replacement or retention rates for software-made matches against matches made by staff.
Claimed by
Nerdy, February 2026
Open
01
A tutor-matching test set
A public set of 200 made-up requests from families, written the way parents write them, in several languages, some with the real need left unstated (a request for "a maths tutor" when the trouble is exam nerves), and a pool of 1,000 made-up tutor profiles. Three experienced tutoring coordinators each pick their best three tutors for every request. AI matchers and a simple baseline are then scored on how often their picks overlap the coordinators'.
The briefCredit · unclaimed02
Who checks tutors, by country
A public table, for 20 countries, of whether law requires a criminal-record check for private tutors and for tutoring centers, citing the statute; and, for the 40 largest tutoring platforms and agencies in those countries, which checks each says it makes on tutors (identity, qualifications, criminal record, references), taken from its own public pages and rechecked every six months.
The briefCredit · unclaimed
Open 01 · the brief
A tutor-matching test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of 200 made-up requests from families, written the way parents write them, in several languages, some with the real need left unstated (a request for "a maths tutor" when the trouble is exam nerves), and a pool of 1,000 made-up tutor profiles. Three experienced tutoring coordinators each pick their best three tutors for every request. AI matchers and a simple baseline are then scored on how often their picks overlap the coordinators'.
Why it is still open
Claims about matching have not been tested in public. Crimson Education's founder registered a randomized trial of matching students and tutors by personality test in 2019; the registry still lists it as ongoing, with an end date of December 2020 and no results posted. Nerdy's matching automation is described only in its own letters, and the matching code found on GitHub is student and hobby projects with no test set. The nearest benchmark, published in September 2025, gives AI models 369 college-admission matching problems with the preferences already ranked; it does not test reading a family's request.
Built with what exists
Claude or Gemini to draft the requests and profiles from a written specification; the Hungarian algorithm, a standard method for one-to-one assignment, as the baseline; current AI models as matchers. Ai2's XYBench, covered in the Note "Answering the question behind the question", measures whether an answer finds the user's real goal, the skill a matcher needs.
First step
Write 30 requests and 150 profiles and have two coordinators rank them, to measure how often people agree.
Done looks like
The set and a public scoreboard on which at least three AI matchers and the baseline are scored for top-three overlap with the coordinators, shown next to how often the coordinators agree with one another.
Open 02 · the brief
Who checks tutors, by country
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table, for 20 countries, of whether law requires a criminal-record check for private tutors and for tutoring centers, citing the statute; and, for the 40 largest tutoring platforms and agencies in those countries, which checks each says it makes on tutors (identity, qualifications, criminal record, references), taken from its own public pages and rechecked every six months.
Why it is still open
In England private tutoring "remains unregulated" and tutors are not required to hold an enhanced DBS check (Tes, June 2024), and UNESCO's 2021/2 education report found that less than half of countries regulate tutoring at all. Stanford's National Student Support Accelerator advises US school districts which checks to run on tutors and says the rules vary by state, without listing them. This research found no country-by-country table of what tutors must pass and what platforms check.
Built with what exists
An AI agent that reads laws and policy pages, such as Claude driving a browser through Browser Use (see the Note "Deleting 80% to start over"); a spreadsheet with a source link and a quoted sentence in every cell; a second human reader.
First step
Fill the table for England, Scotland, Hong Kong and Singapore, starting from the DBS guidance for private individuals and the NSPCC's page for tutors.
Done looks like
20 countries and 40 platforms filled in, every cell linked to a quoted source, and at least 95% agreement between the agent and a person on a 100-cell sample.