What the research says actually makes tutoring work

Tutoring produces about a 0.29 standard deviation gain in learning, according to a 2024 meta-analysis of 89 randomised experiments, which makes it one of the best-evidenced interventions in education. The effect is not uniform: trained tutors, at least three sessions a week, earlier grades and during-school delivery all predict larger gains.

That last list matters more than the headline, because the tutoring most families actually buy - one session a week, after school, from a non-specialist - sits outside most of it.

The headline finding

Nickow, Oreopoulos and Quan reviewed every randomised tutoring experiment they could find and published the result in the American Educational Research Journal. Their pooled effect was 0.288 standard deviations: 0.290 for literacy, 0.268 for maths.

Their earlier working paper, covering a slightly different set of studies, reported 0.37 SD and described it as moving a median student to roughly the 66th percentile.

Either figure is large by the standards of education research, where most interventions that survive a randomised trial produce much smaller effects or none.

What predicts a bigger effect

Who does the tutoring. Programmes using teachers and paraprofessionals outperformed those using non-professional volunteers or parents. In the working paper's breakdown, teacher-led programmes pooled around 0.50 SD, paraprofessionals around 0.40, non-professionals around 0.21.

This is not about credentials for their own sake. A trained tutor hears a half-right answer and knows which follow-up question to ask. That diagnostic skill is the thing being bought.

How often. At least three days a week was associated with the strongest effects. This is the finding behind the term "high-dosage tutoring".

When. During the school day beat after school. The likeliest reasons are mundane: timetabled sessions are attended, and attention at 10am beats attention at 6pm.

Which grade. Effects were strongest in earlier grades, with reading tutoring doing better early and maths tutoring better later.

The uncomfortable implication

Line that up against how tutoring is usually purchased by a family:

Two of those four are usually missed, and frequency is the one most driven by price. At $60 an hour, three sessions a week is $7,000 a year, which is out of reach for most households. So families buy the pattern they can afford, which is the pattern with the weakest support.

This is the real argument for cheaper tutoring: not that cheap is as good per hour, but that frequency is purchasable at a lower rate and frequency is what the evidence rewards.

What the research does not say

Being careful here matters, because this literature is widely overstated.

It does not say every tutoring programme works. It is a pooled average over programmes that varied enormously; some individual studies found little.

It does not say the effects persist indefinitely. Most studies measure at or shortly after the end of the programme.

It does not speak to AI tutoring. These are human tutoring programmes. Anyone citing this meta-analysis as evidence for an AI product, ourselves included, is reasoning by analogy rather than from evidence.

It does not establish a dose-response curve. The working paper noted that four to five sessions a week helped the youngest children but was associated with weaker effects for grades 2 to 5 - more is not uniformly better.

The five features that predict a result

Drawn from what the meta-analysis found varied with effect size, in the order they are worth insisting on.

1. Frequency. At least three sessions a week. The single most actionable finding, and the one most often traded away for a more impressive tutor.

2. A trained tutor. Teachers and paraprofessionals outperformed volunteers and parents. The mechanism is diagnostic: a trained tutor hears a half-right answer and knows which of three misunderstandings produced it.

3. The same tutor each time. Continuity is a feature of nearly every programme that worked. Relationship does part of the work, and so does not re-explaining the student's history weekly.

4. A structured curriculum. A plan across sessions rather than whatever the student brought today. Reactive tutoring is expensive homework help.

5. Timetabled rather than optional. During-school programmes beat after-school ones, and the likeliest reasons are attendance and attention rather than pedagogy. At home the equivalent is a fixed slot that does not move.

Why high-dosage tutoring is hard to buy

Everything above describes school intervention programmes, which is where most of the research was done. A family buying privately faces a different market:

• Hourly pricing makes three sessions a week cost several thousand dollars a year.
• After school is the only time available, which is the weaker slot.
• Agencies rotate tutors when schedules change, which breaks continuity.
• Few private tutors work from a structured curriculum across sessions.

So the private market tends to supply the opposite configuration from the one with the best evidence. That is not dishonesty; it is what an hourly market produces. It does mean a family should read "tutoring works" as "tutoring run a particular way works" and ask accordingly.

What this means for school and district choices

If you are choosing a programme rather than a person: the findings point at small-group tutoring embedded in the timetable, delivered by trained staff, three or more times a week, with a structured curriculum and consistent pairings. Volunteer programmes still show positive effects and are worth running; they should not be expected to match trained delivery.

The honest caveat on all of it: these are averages across programmes that varied enormously, and most studies measured outcomes at or shortly after the programme ended. Durability beyond that is much less well established than the headline effect.

What to do with this

If you are buying tutoring:

• Favour frequency over prestige. Three cheaper sessions beat one expensive one.
• Ask about training in the subject specifically.
• Keep the same tutor. Continuity is doing work in every one of these programmes.
• Ask for structure: a plan across sessions, not a list of whatever the student brought.
• Agree in advance how you will know in six weeks whether it is working.

Where TruLearn fits

We designed around that list deliberately. TruLearn arrives with a plan built before the session rather than waiting for the student to bring a problem; it is the same tutor every time; it holds a rubric and a set of known misconceptions written by a person for each concept; and because the cost of a session is not an hour of someone's time, three or four sessions a week is an ordinary way to use it rather than a luxury.

What we cannot claim is a place in the evidence above. Those are human tutoring trials, and we have no outcome data of our own, which we state plainly. The design follows the evidence; the result is not yet demonstrated.

More: what a tutor costs, and AI tutor versus human tutor.