Study groups have a split reputation: students swear by them, and the research suggests most of them don't work. Both are right. A structured study group is one of the most effective learning formats available; an unstructured one is a social hour with notes open. The difference isn't the group — it's whether the session makes everyone retrieve or lets everyone watch someone else retrieve.

Here's how to build the kind that works.

Why groups fail

The default study-group format — one person explains, everyone else nods — is rereading with company. The explainer gets the benefit (teaching is retrieval), but the nodders get the illusion: they follow the explanation, feel like they understand, and mistake that fluency for their own ability to produce the material.

That's the core failure: retrieval is being done by one person and observed by the rest. The observers get the comfort; only the retriever gets the learning. A group where everyone watches the strongest member work is a lecture, not a study group.

The rules that make it work

Retrieval first, always. Start every session with a solo attempt — 5 minutes, closed book, everyone answers the same prompts silently. This is non-negotiable: the group's value is built on top of individual retrieval, not instead of it. Skip the solo quiz and you've skipped the learning.

Everyone teaches something. Assign each member a topic to explain — not to read aloud, to teach from memory. Teaching is retrieval plus organization plus error-detection; it's the highest-yield activity the group enables. Rotate who teaches what so every member teaches across the semester.

Feedback is the group's job. The person explaining gets corrected, questioned, and pushed by the others — that's the mechanism. "Explain it again, I'm not convinced" is worth more than an hour of solo review, because it forces the explainer to find the actual gap, not just re-say the familiar phrasing.

End with the error list, not the good vibes. Every session produces a written list of what the group got wrong — the disagreements, the stumbles, the "wait, how does that work again." That list is next session's opening quiz. The errors are the syllabus; the group is just the mechanism for finding them.

A 60-minute structure that works

• 5 min — solo quiz. Same prompts for everyone, closed book, silent. Establishes the baseline retrieval before anyone can coast.
• 20 min — round-robin teach-back. Each member teaches their assigned concept from memory. The group interrupts, questions, corrects. The interruptions are the point — they expose the gap between "followed it" and "understood it."
• 15 min — transfer scenario. A novel problem or case nobody's seen — the group works it together, applying what was just taught. This is the step that turns "we each know our part" into "we can use any of it."
• 15 min — error harvest. What did the group get wrong? Write it down. These become the prompts for the next meeting's solo quiz and each member's spaced review.
• 5 min — exit ticket. Each person writes 3 things they now know cold and 1 thing still shaky. The shaky one is their homework.

Where the agent fits — and where it doesn't

An agent in a study group is a prompt generator and an arbiter, not a participant:

• Before the session: use it to generate the solo-quiz prompts and the transfer scenario. The group's bottleneck is usually "who makes the questions" — an agent solves that in seconds, from your actual course materials.
• During the session: use it to resolve factual disputes — "is it X or Y?" — not to explain concepts. The explaining is the group's work; outsourcing it to the agent removes the peer-teaching that is itself the learning mechanism.
• After the session: use it to turn the error list into next week's quiz prompts and each member's spaced re-tests. The errors are already identified; the agent just schedules their return.

When a study group is the wrong tool

• When you don't know the basics yet. A group is for sharpening, not for first exposure. Learn the material solo first; bring the questions the solo work couldn't answer.
• When the group's level is too uneven. If one member is always the teacher, the rest are in a lecture. Pair levels closer, or split the teaching so everyone has something to explain.
• When it becomes a standing social appointment. A weekly group that hasn't produced an error list in a month is a hangout. The error list is the output; no output, no point.

Where LearnX fits

LearnX's Study Squad is built for exactly this — a shared Exam Sprint gives the group aligned material, the question generation produces the solo-quiz prompts and transfer scenarios from your actual course, and the per-member performance tracking surfaces the error list automatically. The agent handles the prompt generation and the factual arbitration; the group does the teaching and the correcting. That's the division of labor that makes the format work.