How to Run a Study Group With AI: A 90-Minute Session Plan That Actually Works

Published on: November 1, 2026 • Last Updated: November 3, 2026

Short answer: Run the 90-minute plan in five blocks: cold recall, group quiz, teach-back, source check, and a gap list with a solo task.

Most study groups do not fail because the students are lazy. They fail because nobody owns the session. One person talks, two people nod, someone checks their phone, and the group leaves with the same gaps it arrived with — only now it feels like studying happened. Artificial intelligence can fix the structure, not by doing the learning for the group, but by giving the group a job for every minute and a way to check whether the answers are actually right.

This guide gives you a complete 90-minute study-group session plan built around AI tools you probably already have: a general chatbot, a source-grounded tool such as NotebookLM (now called Gemini Notebook), and the quiz and flashcard tools you already use. It is written for students revising for exams, not for teachers running a classroom, and it keeps the academic-integrity line in plain sight throughout.

Why most study groups waste time

A study group without a plan defaults to re-reading. Someone reads a slide aloud, someone else says “yeah, that makes sense”, and the group moves on. Re-reading feels productive because it is familiar, but it does not test whether anyone could explain the idea without the slide in front of them. Research on learning has repeatedly found retrieval — trying to recall an answer before checking it — far more useful than re-reading, and a group session is an ideal place to practise it, because other people can catch your mistakes.

AI makes the failure mode worse if you let it. Ask a chatbot to “explain this chapter” and it will produce a smooth, confident summary that the whole group can agree with without anyone learning anything. Ask it instead to quiz the group, to argue against a wrong answer, or to check a worked solution step by step, and it becomes a sparring partner rather than a shortcut.

The plan below is built on three rules. First, nobody reads an AI answer aloud and moves on; every AI output gets tested against a source or against the group’s own reasoning. Second, every member has a role, so nobody can hide in the audience. Third, the session ends with solo work, because group understanding is not the same as individual recall.

Before the session: the 10-minute setup that saves 30

Agree on one topic, not a whole subject. “Cell respiration” is a session; “biology” is a month. Each member brings one source the group trusts — lecture notes, a textbook chapter, or a set of slides — and uploads the same sources to a shared, source-grounded workspace if you use one. When the AI answers from your own uploaded notes rather than from the open web, its answers are easier to check, and invented references are far less likely.

Pick the roles for the session in advance and rotate them next time:

• Facilitator: keeps time, reads the plan, and stops the group drifting into re-reading.
• Challenger: argues against answers, including AI answers, and asks “how do we know that?”
• Scribe: writes down every gap the group finds, in the group’s own words.
• Checker: verifies every disputed claim against the source material before the session ends.

Agree the integrity rule out loud before you start: AI may quiz you, explain a concept, and check your work, but it may not write work you will submit, and any answer you cannot explain in your own words does not count as learned. If your institution has an AI-use policy, follow it; where the policy is stricter than this guide, the policy wins.

The 90-minute AI study group plan

Run the session in five blocks. The timings are deliberately tight: a group that knows the clock is moving stays on task, and a group that finishes a block early can spend the spare minutes on the hardest gap, not on chatting.

Block	Time	What happens	AI’s job
1. Cold recall	0–15 min	Everyone writes answers to five questions from memory, alone, no notes, no AI.	None yet. The AI has nothing to grade until you have tried.
2. Group quiz round	15–40 min	Members quiz each other from the cold-recall answers; the group argues out the disagreements.	Generate follow-up questions and harder variants of the questions the group got wrong.
3. Teach-back	40–60 min	Each member teaches one concept to the group in two minutes, without notes.	Act as a confused student: ask the questions a real beginner would ask.
4. Source check	60–75 min	The checker verifies every disputed claim against the notes or textbook.	Point to the exact passage in the uploaded sources that settles each dispute.
5. Gap list and solo plan	75–90 min	The scribe reads back the gap list; each member picks their own solo task for the week.	Turn each gap into a flashcard set or a five-question quiz for solo review.

Block 1: cold recall, no exceptions

Start with five questions written before the session, ideally by the facilitator, covering the topic’s core ideas rather than trivia. Everyone answers alone, on paper or in a private document, for ten minutes. This feels uncomfortable, and that is the point: the discomfort tells you exactly where the gaps are. If the group skips this block and goes straight to AI explanations, you will spend the session nodding at material you cannot actually recall.

For ideas on turning notes into good quiz questions, see our guide to turning notes into a quiz with AI, and for the solo follow-up, active-recall prompts work well as the bridge between a group session and individual revision.

Block 2: argue, then ask the AI

Now compare answers. Where the group agrees, move on quickly. Where it disagrees, argue it out first, in your own words, before asking the AI anything. Only then ask the AI to adjudicate — and ask it to show its working, not just the answer. A useful prompt pattern is: “Two members disagree: one says X, the other says Y. Which is right for this topic, and what is the one sentence in our notes that settles it?” If the AI cannot point to a passage, the claim stays disputed until the checker settles it in block 4.

Ask for harder variants, not more explanations. “Give us a question that someone who only memorised the definition would get wrong” produces far better practice than “explain this again, more simply”. The group should get at least one question wrong in this block; if it does not, the questions were too easy.

Block 3: teach-back with an AI beginner

Each member teaches one concept for two minutes, with no notes. The rest of the group, plus the AI in the role of a confused beginner, asks questions. Teaching exposes shallow understanding faster than any quiz: you cannot teach what you only recognise. The AI beginner is useful here because it never gets embarrassed and never lets a vague sentence pass — it will ask “but why does that step happen?” until the explanation is actually complete.

Keep teach-backs short and rotate the topic, not just the speaker. If one member always teaches the easy concept, rotate the concepts too, so everyone eventually teaches the hardest one.

Block 4: source check, the block most groups skip

Every disputed claim from blocks 2 and 3 goes to the checker now. The checker opens the source — the notes, the textbook, the slide — and reads the relevant passage aloud. If the AI’s answer matches the source, it stays. If it does not, the source wins and the AI’s answer is written down as a cautionary example, not deleted: knowing how the AI misled you is itself useful revision, because the same mistake will tempt you again under exam pressure.

This is where a source-grounded tool earns its place. Our NotebookLM guide for students walks through building a workspace from your own sources, so the check in this block takes minutes rather than a hunt through PDFs.

Block 5: leave with a gap list, not a good feeling

The scribe reads the gap list back, in the group’s own words, and each member chooses one solo task before the next session: a flashcard set, a five-question quiz, or a one-page summary written from memory. The AI’s job here is purely mechanical — turn the gaps into practice material. The learning happens when you do that material alone, later, which is why our guide to using AI tools for studying treats group sessions and solo review as two halves of one system, not two separate activities.

An illustrative session, start to finish

Imagine, as an example only, a group of four students revising photosynthesis for a biology exam. In block 1, all four answer five cold-recall questions; three confuse the light-dependent reactions with the Calvin cycle. In block 2, they argue it out, then ask the AI for a question that a memoriser would get wrong — it asks where ATP is actually made, and two members change their answers. In block 3, one member teaches the Calvin cycle in two minutes and gets stuck on why it is called a cycle; the AI beginner keeps asking until the group can explain it. In block 4, the checker finds the textbook passage that settles the ATP question. In block 5, the gap list is three items long: ATP location, the cycle’s inputs and outputs, and the difference between the two stages. Each member leaves with a five-question quiz on exactly those gaps. The session worked not because the AI was clever, but because the group used it to find and fix specific gaps.

Prompts that help, and prompts that hurt

Good group prompts ask the AI to test, challenge, or verify. “Quiz us one question at a time and do not reveal the answer until we have all answered.” “Argue against our answer and tell us the strongest counter-argument.” “Find the passage in our uploaded notes that supports or contradicts this claim.” Bad group prompts ask the AI to replace the work: “Summarise the chapter for us”, “Write our revision notes”, or “Give us the answers to this problem set”. The first set makes the group think harder; the second set makes the group feel finished while learning nothing.

Watch for two failure modes. The first is AI agreement: chatbots tend to agree with the group’s last answer, so never ask “is this right?” without also asking “what would make it wrong?” The second is the confident wrong answer, which sounds most convincing on the topic the group knows least. That is why block 4 exists, and why it is not optional.

After the session: the 15-minute solo loop

Within a day, each member should spend fifteen minutes alone on their gap task, without the group and without the AI’s help at first. Try the flashcards or quiz from memory, then check. If you still cannot answer a gap question alone, it goes back on the list for the next session — that is not failure, it is the system working. Groups that repeat this loop weekly, with the same roles rotating, tend to find their gap lists getting shorter and more specific, which is the honest sign of progress: vague gaps (“I don’t get photosynthesis”) become precise ones (“I mix up where ATP is made”).

If your group meets online, the same plan works with one change: keep cameras on for teach-back, and put the gap list in a shared document during block 5, not afterwards from memory. Memory is exactly what the session is trying to improve, so do not rely on it to record the session.

What worked and what didn’t

What worked:

• Running the session in five tight blocks: cold recall, group quiz, teach-back, source check, and a gap list with a solo task, with a role for every member, so nobody can hide in the audience.
• Arguing disagreements out in your own words first, then asking the AI to show its working and point to the passage in your own notes that settles the claim, and ending with the 15-minute solo loop within a day.

What didn’t work:

• Letting the session default to re-reading, with one person talking while others nod or check their phones, and leaving with the same gaps but feeling like studying happened.
• Asking the AI to ‘explain this chapter’ or ‘summarise the chapter for us’, which produces a smooth, confident summary the whole group can agree with without anyone learning anything.
• Skipping the cold-recall block and going straight to AI explanations, so the group spends the session nodding at material it cannot actually recall.
• Asking ‘is this right?’ without also asking ‘what would make it wrong?’, since chatbots tend to agree with the group’s last answer, or trusting a confident wrong answer on the topic the group knows least.
• Skipping the source-check block, the block most groups skip, or leaving with a good feeling instead of a gap list, since group understanding is not the same as individual recall.

 

Frequently asked questions

How many people should be in an AI study group?

Three to five works best. With two, there is nobody to break a tie; with six or more, teach-back takes too long and quieter members disappear. If your group is larger, split into two groups for blocks 1–3 and rejoin for the source check.

Do we need a paid AI tool for this plan?

No. The plan uses a free general chatbot for quizzing and a free source-grounded workspace for the source check. Paid tools may be faster or allow more sources, but they do not change the structure, and you should check the vendor’s current limits before paying for anything.

What if the AI gives different answers to different members?

Treat that as a gift, not a bug. Different answers mean the question is ambiguous or the group is missing a distinction. Put the claim on the disputed list, and let the source check in block 4 settle it. Never settle a disagreement by majority vote among AI answers.

Is using AI in a study group cheating?

Using AI to quiz you, explain a concept, or check work you have already attempted is study help, in the same way a tutor is. Using it to produce work you submit as your own is not. When in doubt, check your institution’s policy and disclose the help where the policy asks you to.

How often should the group meet?

Once a week for a single topic, with the solo loop between sessions, beats three long sessions before an exam. Spacing the sessions out is what turns group understanding into individual recall.

For the solo side of the system, our comparison of AI flashcard apps can help you pick the tool that turns your gap list into a review habit you will actually keep.

Related: After a group session, log the questions the group got wrong — our AI error log method shows how to turn those misses into your next revision plan.