Does teaching by questions actually work?

Teaching by asking · Part 2 of 2

Part 1 made a confident case that asking beats telling. This part is where I show my work — including the parts that don't flatter us.

That's not modesty for its own sake. BrainScribe rests on a single claim: that a coach who only ever asks questions produces something better than a tool that hands over the answer. If that claim is weak, you should be able to find that out from us rather than from someone else.

The strong evidence

Generating an answer beats reading one. A meta-analysis pooling 445 effect sizes across 86 studies found a consistent advantage — around half a standard deviation — for information people generated over information they simply read (Bertsch, Pesta, Wiscott & McDaniel, 2007, Memory & Cognition).

Prompting explanation beats providing it. Ask learners to explain why something is true or how it connects, and they outperform learners given the explanation. A 2018 meta-analysis of 69 effect sizes from 64 studies found a moderate-to-large benefit that held across most school subjects, and across both conceptual and procedural knowledge (Bisra, Liu, Nesbit, Salimi & Winne, Educational Psychology Review).

Scaffolding is a real construct, not a metaphor we invented. The term comes from Wood, Bruner and Ross (1976), describing support that lets a learner accomplish what would be "beyond his unassisted efforts" — and that is progressively withdrawn.

Caveat one: the lab is not a classroom

The generation-effect studies are largely laboratory memory tasks — generating a word from a cue, then being tested on recall.

"People remember words they generated" is a considerably narrower claim than "talking through your essay makes you a better writer." Anyone citing the generation effect as direct proof of an essay-coaching product is over-reaching, and I'd rather say so than let you notice it yourself.

The self-explanation work is much closer to the real thing — actual students, actual school tasks, actual subject matter — so that's the finding doing the real load-bearing here. The generation effect supports the underlying principle, nothing more.

Caveat two: the field's flagship review rates this approach moderate

This is the one that would be easiest to leave out.

The most authoritative survey of learning techniques is Dunlosky, Rawson, Marsh, Nathan and Willingham (2013), published in Psychological Science in the Public Interest. It evaluates ten study techniques and assigns each a utility rating.

The two techniques nearest to what a Socratic coach does — elaborative interrogation ("why is this true?") and self-explanation — both received moderate utility. Not high.

High utility went to two other things entirely: practice testing and distributed practice. Retrieval and spacing.

The authors' stated reason is that questioning techniques "have not been adequately evaluated in educational contexts." So it's a verdict about thin evidence rather than a finding of no effect — but it is emphatically not an endorsement, and you'd find it in about one search. Better you find it here.

Caveat three: "Socratic" specifically is the weakest link

There's a difference between the mechanism and the brand name.

The mechanism — generation, self-explanation, scaffolding — is well studied. The named practice of "the Socratic method" is studied much less cleanly. Results vary by subject, age group, and outcome measure, and a good deal of the research assesses the quality of classroom discussion rather than whether anyone learned more, months later.

We use the word because it's the clearest available label for asking rather than telling. We don't lean on it as though it were itself an evidence base.

The limit that shaped the product

Across the classroom-questioning literature, one condition recurs: questions help when the learner has enough knowledge to answer from something real. When they don't, questioning degrades into guessing, and guessing under pressure teaches very little.

That is a genuine boundary on the method, and it's the one we designed to rather than around.

BrainScribe's coaches work on material the student already holds — their own reading, their own argument, their own assignment. The coach is not there to teach your kid a subject they haven't encountered. It's there to get thinking they already have out of their head and onto the page, which is the exact task the evidence supports.

It also happens to be why the coach can't write for them. It has no content of its own to supply. Take away the student's answers and there's nothing left.

What we actually claim

Not "proven best." Not "backed by science" as a decorative phrase.

Something narrower and, I think, more useful:

• Getting a student to produce and explain their own reasoning is well supported, most directly by the self-explanation research.
• The specific packaging — a voice-based AI coach that only asks — is new, and not itself the subject of a literature. Nobody has run that trial. We won't pretend otherwise.
• The method has a known limit, and we built inside it deliberately.

If that's less thrilling than a claim of proof, good. Overclaiming is how education products lose parents, and it's the fastest way to deserve it.

Request an invite — BrainScribe is invite-only while we're in early access.