Meeting notes are not memory: from transcripts to decisions
Meeting notes record what was said, not what was decided. How to turn AI transcripts and summaries into decisions your team and AI tools can find.
Meeting notes record what was said. Team memory records what was decided, why, and what changed as a result, in a place people and AI tools can find later. AI note-takers have made transcripts and summaries cheap, but a folder of summaries is still not memory until you pull out the decisions and put them where future work happens.
Most teams now have more meeting notes than ever and still ask "wait, didn't we decide this already?" This post explains why, and gives a simple process for turning transcripts into decisions people can use.
Why more notes didn't fix the problem
AI note-takers are now built into common meeting tools. Google Meet's "Take notes for me", for example, writes notes into a Google Doc that is saved in the organizer's Drive and shared with invitees based on the organizer's settings (Google support).
That's useful. But look at where the output lands: one document per meeting, in one person's drive, shared with the people who were already in the room. Three months later, the question is "what's our current policy on refunds for annual plans?" and the answer is spread across six meeting docs, two of which contradict each other.
Notes fail as memory for a few reasons:
They're organized by meeting, not by topic. Nobody looks for "the 14 March sync." They look for "refund policy."
They mix decisions with discussion. The decision is one sentence in a page of back-and-forth.
They don't show what changed later. A note from March doesn't know about the reversal in May.
They reach the wrong people. The people who need the decision most were often not in the meeting.
Atlassian reports that 25% of the workweek is spent searching for information. Piling up transcripts can add to that if nobody distills them.
Notes vs memory
Meeting notes | Team memory | |
|---|---|---|
Unit | One meeting | One decision or topic |
Content | What was said | What was decided and why |
Lifespan | Accurate on the day | Updated when things change |
Audience | Attendees | Anyone affected, plus AI tools |
Found by | Date and meeting name | Topic, project, or question |
What's worth keeping from a meeting
Most of a meeting doesn't need to be remembered. Pull out four things:
Decisions. What did we agree, and what's the reason?
Changes to earlier decisions. What did we reverse or update?
Commitments. Who will do what, by when?
Open questions. What did we agree we don't know yet?
Everything else, the discussion and the options considered, can stay in the transcript as a source you can link to.
From transcript to decision: a simple process
Step 1: Mark decisions in the meeting
Before the meeting ends, the owner reads back the decisions in one sentence each. "We're moving the launch to the 20th because the payment provider review won't finish in time." This takes a minute and makes everything after it easier, whether a person or an AI is writing the notes.
Step 2: Extract with AI, review as a human
Give your AI tool the transcript and ask it to list decisions, reasons, owners and open questions in a fixed format. Then have the meeting owner check it. AI is good at finding candidate decisions. It's less reliable at telling a firm decision apart from an idea someone floated.
A prompt that works:
From this transcript, list only decisions that were clearly agreed. For each, give: the decision in one sentence, the reason given, who agreed, and any earlier decision it replaces. List open questions separately. Don't include ideas that were discussed but not agreed.
If your team runs this often, turn it into a shared skill so everyone gets the same format.
Step 3: File decisions by topic
Each decision goes into your decision log under the topic it affects, such as pricing, onboarding or the billing service, with a link back to the transcript as the source. If it replaces an earlier decision, mark the old one as superseded.
Step 4: Tell the people it affects
Post a short note in the channel of each affected team. "Pricing: we're dropping the annual discount for new customers from 1 November. Reason and source in the decision log." This step is easy to skip, and it's the one that prevents duplicate work.
A concrete example
A product team meets weekly. In one meeting they agree to stop supporting an old browser version from next quarter. The AI summary includes it as one bullet in a list of twelve.
Two weeks later, a support agent uses an AI tool to draft a reply to a customer on that browser. The tool has no idea about the change. The reply promises a fix.
With the process above, the decision is in the log under "Browser support," linked to the transcript, and posted in the support channel. The support agent sees it. If your AI tools can read the decision log, the drafted reply gets it right too.
FAQ
Should we still keep meeting transcripts?
Yes, as a source. Transcripts are useful when someone needs to check exactly what was said or why a decision was made. Just don't treat them as the place people go to find decisions.
Can AI do the whole thing automatically?
AI can draft the decision list and suggest where each one belongs. A person should still confirm that a decision was really made, because summaries can turn a suggestion into a commitment by mistake.
How long should a decision entry be?
Short. One sentence for the decision, one or two for the reason, plus who agreed, the date and a link to the source. If it's longer, it's probably a doc, not a decision.
What about meetings with no decisions?
That's useful information too. If a recurring meeting rarely produces decisions, it may be better as a written update.
Keep the decision, link the transcript
Recording meetings is easy now. Remembering what they decided still takes a small, deliberate step: pull out the decisions, file them by topic, mark what they replace, and tell the people affected. Start with your most important recurring meeting this week.
We're building Shane to turn decisions from meetings, chats and docs into team memory that every teammate and AI tool can find, with sources. If that's a problem your team has, you can join the waitlist at tryshane.ai.