Is it worth the money? The honest version.

What is well evidenced

That tutoring works. The largest synthesis is a meta-analysis of 96 randomised studies, and its pooled effects are large by the standards of education research — around 0.38 standard deviations for one-to-one, drawn from 67 separate studies. Effects of that size are uncommon; most schooling interventions struggle past 0.1.

One qualification belongs here rather than in a footnote, because it is the number closest to most families reading this. The same analysis reports its thinnest cell at the secondary level: across grades 6 to 11 the pooled effect is 0.16 standard deviations, from just eight studies (registered as TUT-17, the same entry Figure 1 plots). The headline figure is driven by younger students and by reading rather than writing. That does not make it wrong; it makes it a weaker basis for a promise about a Year 9 student than the 0.38 alone would suggest.

And here is the part the industry tends not to mention

Look at that third bar. If smaller groups were simply better, one to three would not beat one to two. The meta-analysis’s own table is not monotonic in group size, which means group size is not doing the work people assume it is doing. Something else — who the tutor is, what happens in the session, how often it runs — is carrying the effect.

There is a clean one-to-one result, and it is worth having, but it is narrower than it is usually quoted as being:

one-on-one virtual early literacy tutoring was nearly twice as effective as two-on-one tutoring for improving student learning… Tutors in two-on-one settings also spent 63 percent more time managing behavior and transitions

National Student Support Accelerator, Stanford University (2025), The power of personalized attention. Registered as TUT-12.

That is virtual early literacy tutoring. Not secondary English, not Australia, and not this. It is real evidence and it points the right way; it is not a warrant for claiming the same multiple applies to a Year 9 essay.

The Australian gap

Despite the popularity and success of one to one programs, there is little research that specifically examines the effectiveness of the one to one element of these programs in Australia.

Evidence for Learning (2024), Teaching and Learning Toolkit — one to one tuition. Registered as TUT-04.

That is the Australian mirror of the EEF Toolkit saying, in its own words, that the local evidence base for the specific thing being sold here is thin. Anyone quoting “+5 months’ progress” at you is quoting a pooled international figure, not an Australian measurement of their own programme.

And when Australia did measure it, twice

Since that toolkit entry, two Australian governments ran tutoring at scale and had it evaluated against matched comparison students. Both found what no tutoring business wants quoted.

Students who received tutoring in 2023 did not show greater learning gains than similar students who did not receive tutoring. When we compared similar students from each group, we found that students who received tutoring learnt less than those who did not receive tutoring.

Victorian Auditor-General’s Office (2024), Effectiveness of the Tutor Learning Initiative — a $1.2 billion programme across more than 1,500 government schools. Registered as TUT-18.

At a system level, participating students and similar non-participating students improved their standardised test scores by the same amount between 2021 and 2022. This improvement in academic outcomes cannot be attributed to the effects of the program as implemented, as students experienced the same degree of academic growth regardless of whether they participated in the program.

NSW Department of Education, CESE (2023), COVID Intensive Learning Support Program evaluation — 138,268 students, 17% of all NSW government school students. Registered as TUT-19.

Read them precisely, because the precision is the point. Both evaluate school-run small-group tuition at system scale — tutors recruited in thousands, often in groups, inside the school day, for a fixed number of weeks. Neither evaluates weekly one-to-one teaching with a named practitioner over a year, which is a different intervention with a different mechanism. So they do not show that this practice does not work; nothing published shows that either way.

What they do show is harder and more useful: tutoring delivered as a programme, at scale, can produce nothing at all. The pooled international effect sizes are averages over implementations, and a large Australian implementation landed at zero — twice. That is the strongest available argument against treating “we provide tutoring” as evidence of anything, including here.

So what should you actually ask a tutor?

If group size is not the mechanism, the useful questions are about what happens in the hour.

• Who teaches the lesson, every week? The meta- analysis found effects varied by who tutored. A roster where your child gets whoever is free is a different product from a named person who knows the last six pieces they wrote.
• What happens to the writing between lessons? Tutoring effects come from feedback on work, not from contact hours. Ask what is read, and what comes back.
• What are they measuring, and against what? A tutor who quotes an effect size for their own programme should be able to say who measured it. Almost none can, and the tutoring entries in the register are there so you can check any figure you are given.
• Beware the guarantee. Nobody can promise a band or a study score, and the year-to-year movement in national results is mostly noise the measuring authority itself declines to call a difference.

What this practice actually claims

One named practitioner teaches every lesson, weekly, and the work between lessons is reading the student’s writing closely enough to say what specifically went wrong in a sentence. That is the mechanism the evidence supports — feedback on real work by someone who remembers the last piece — rather than the ratio itself.

No outcome is guaranteed, no result is claimed on this site, and the register exists so that every figure used here can be checked against its source.