Using AI summaries in a reading group?Agree on the ground rules first

With AI summaries, every participant arrives at least minimally prepared. Fewer people saying "I never opened today's material" is an unqualified win for the flow of the meeting.

But there's a trap. A summary is so much easier to read than the original that reading it can feel like having read the original — and the urge to open the original evaporates.

The problem isn't AI summarization itself. It's the absence of an agreement about what the summary is for. Settle that first.

In the prep phase before a session, an AI summary does four jobs.

Sharing the big picture, so everyone knows what today's range covers. Clearing up terminology, so unfamiliar technical terms are sorted in advance.

A section map — a list of what's where. And candidate discussion topics — places where interpretations might reasonably differ.

What these have in common: they're all jobs that gather raw material for discussion. That far, you can delegate to AI with confidence.

Conversely, there are things the group should never delegate.

Substituting for the reading: the summary is a guide to the original, not a replacement for it. Judging validity: "can we trust this study?" gets settled with the original's methods and figures, not an AI's answer.

Arbitrating disagreements: when members read a passage differently, "the AI says so" must not be the tiebreaker. Preserving nuance: the authors' careful hedges and points of emphasis are exactly what summaries flatten.

When interpretations clash, the thing to go back to is the original — not the AI. "Because that's how the AI summarized it" is not an argument in a reading group.

When bringing AI summaries into a reading group, agreeing on just three things at the outset is enough.

Rule 1: the summary is a shared map; interpretation is each member's own work. Getting oriented via the summary is fine — but everyone reads the original for their assigned part or the part that interests them.

Rule 2: claims are grounded in the original page. Speak from "here's what the text says, right here," not "here's what the summary said." Numbers and figures always get checked at the source.

Rule 3: when the summary is wrong, fix it. That's not tedious housekeeping — it's how the group builds a record of its reading.

First, before the session, the leader uses AI to build the map — a summary of the range, shared with the participants.

Second, each member reads the original for the part that interests them. Not the whole thing. Even one section each, combined, covers a lot of ground.

Third, bring the discrepancies into the meeting. "The summary says X, but the original reads differently" is the best conversation-starter there is.

Fourth, put the discussion's conclusions into words: agreed interpretations, open questions, homework for next time. Fifth, save the corrected summary — a version that reflects the discussion, kept for future reference.

The goal isn't to eliminate the gaps between the AI summary and the original. It's to welcome them as sparks for discussion.

When the discussion turns up a spot where the summary is off, fix it on the spot.

AI draft, check against the original, discuss, revise — a summary that has passed through that cycle is no longer just AI output. It's a record of how the group read the text.

In Paperfy, the leader keeps that record on the paper's page and copies it into a shared memo if needed. Repeat this across a series, and you accumulate reading notes in the group's own words that you can return to later.

Upload the paper PDF to Paperfy and the AI summary, figures, tables, and original all sit on the same page — mapping and source-checking happen on one screen.

Crucially, the AI summary is hand-editable, so understanding revised during discussion can be written back immediately. The leader keeps it in Paperfy and shares it with the team by screen share or shared notes.

The radio script is editable and regenerable too. Build a review audio that covers this session's range plus the key points from last time, and catching up after a gap between sessions gets much easier.

If a reading group uses AI summaries, define them as a shared map — not a substitute for reading, and not an arbiter of interpretation.

Ground every claim in the original page. Welcome summary-versus-original gaps as fuel for discussion. Keep the corrected summary as the group's reading record, ready for next time.

Delegate the gathering of material to AI, and people can pour themselves into the discussion. Delegate the discussion too, and there's little point in meeting at all. Holding that line is what makes a reading group work in the age of AI.