Best AI Prompts for Meeting Notes

Which prompt do you need for which stage?

One ground rule before the prompts: AI is great at processing meeting content; it is poor at deciding what mattered most. Every output below is a starting point that a human reviews and corrects before sharing. The cost of an AI-summarized meeting note that wrongly attributes a decision to the wrong person is much higher than the time it saves; never share an AI summary without a 60-second human read-through.

What are the seven meeting prompts?

1. Agenda design (the night before)

Why this works: The “outcome I need by the end” forcing function is what most meetings skip. “Discuss X” is not a meeting outcome; “decide whether to do X” is. The “2 things this meeting is NOT about” output is the single most useful pre-emptive scope-control line; it gives the chair language to redirect when the conversation drifts. Pre-sent questions raise the meeting's preparation floor.

2. Transcript cleanup

Why this works: The “do not paraphrase” constraint is the safeguard against the most common AI-transcript-cleanup failure mode: AI rewrites what someone said into what AI thinks they meant. The [UNCLEAR] tag is the audit trail. Clean transcripts (rather than summaries) are also the right input for prompts 3 and 4; summarized transcripts lose the source signal those prompts need.

3. Decision extraction

Why this works: Promoting discussion into recorded decision is the single most expensive failure mode of AI meeting notes. The “do not promote a discussion into a decision” constraint, combined with the “verbatim line where confirmed” output, is what keeps the decision log defensible six months later. The “conditions that would change it” output is the part that turns a record into an institutional memory.

4. Action-item assignment

Why this works: “Look into X” is not an action item; it is a deferred decision. The “verb + object” constraint forces real actions. [NEEDS CONFIRMATION] is the safe default for any unowned item; the chair confirms before the summary ships. The “dependency” field is what stops the post-meeting paralysis where everyone is waiting on each other and nobody realizes it.

5. The shareable summary

Why this works: The 90-second-read constraint is the audience-fit constraint. Most meeting notes are written for the attendees (who already know what happened) and forwarded to absent colleagues (who do not have time to read 2,000 words). The “if you need to push back” line is the one that turns a static report into a feedback loop; it gives the absent reader a path to disagree before the decision becomes immutable.

6. The follow-up email

Why this works: Follow-up emails fail when they assume the recipient remembers the meeting. Opening with the specific thing from the meeting transcript gives them the 5 seconds of context they need to act. The “what happens if I don't hear back” line is the soft deadline that drives reply rate without sounding pushy.

7. Recurring-meeting memory

Why this works: Recurring meetings die when they lose memory. The same topic gets re-debated quarterly because nobody remembers the prior decision. The running-memory document is the artifact that compounds; a chair who walks in with last quarter's context drives 30% shorter, sharper meetings. The “single trend most worth bringing” output is the agenda-design input for the next meeting.

What is the worst thing you can do with AI on meeting notes?

• Share the AI summary without a human read. One wrong attribution or one made-up decision destroys the trust the whole workflow depends on. The 60-second review is non-negotiable.
• Send transcripts to AI without consent. Two-party-consent states make this legally exposed; enterprise policies often prohibit it. Check before you upload.
• Promote discussion into recorded decision. “We talked about X” is not “we decided X.” Prompt 3's “verbatim confirmation” constraint exists for this reason.
• Auto-summarize without naming the audience. A summary for executives is different from a summary for the team. Different audiences need different summaries; the same one will fail both.
• Skip prompt 7. Recurring meetings without memory are where most organizational drift lives. The running-memory document is the single highest-impact prompt of the seven.

What if your team has 5+ meetings a day?

The 7 prompts are a candidate for the ladder. Save each as a Claude skill. Bundle as meeting-stack. Wire the workflow: agenda template fires from the calendar invite, transcript cleanup fires when the recording lands, decision + action extraction fires next, summary drafts in your inbox, and the running-memory document updates weekly. The setup is a half-day of investment; for a team with 5+ meetings a day, the recovered hours pay for it in the first week. The bigger structural gain is qualitative: meetings become artifacts, not events.

What are common questions about AI meeting notes?

Which AI tool should I use for transcription?

Otter.ai, Fireflies.ai, Zoom's native transcription, and Microsoft Teams all produce usable raw transcripts. The transcription quality is roughly comparable; the differentiator is the privacy + retention policy and whether your enterprise allows it. Whichever you pick, run prompt 2 to clean the transcript before downstream prompts; raw transcripts always contain attribution errors.

Do I need to tell people I'm using AI for notes?

Yes, before recording. Consent + AI-use disclosure is the right combination. Most teams have settled on a brief “this call will be recorded and an AI-generated summary will be shared” line at the start of the meeting; that is the floor.

What if the meeting is confidential?

Hand-write notes. AI processing of confidential meetings (legal strategy, layoff planning, performance discussions, financial decisions, M&A) is rarely worth the risk, even with internal-only AI tools. The time savings does not outweigh the downside of a leaked transcript or a wrongly summarized confidential discussion.

What about multi-language meetings?

Most major transcription tools handle multi-language but with reduced accuracy. The cleanup pass in prompt 2 is more important; verify name spellings, technical terms, and any translated phrasing before the cleaned transcript feeds into the downstream prompts.

Where do these prompts come from?

They are the workplace-operations section of the larger AI Prompt Library. The Library has over 500 prompts across 33+ categories with three difficulty levels, including the full operations stack: meeting notes, 1-on-1s, project status updates, weekly executive briefs, performance review prep, and hiring debriefs.

Sources to read next?

• Harvard Business Review on meeting management
• McKinsey research on workplace effectiveness
• Otter.ai blog on meeting workflows
• International Association of Privacy Professionals (recording-consent law)
• Meeting (Grokipedia)

You may also like

• Best Claude Prompts for Sales — cold outreach + objection scripts
• Best AI Prompts for Email Writing — the everyday inbox workhorse
• Best AI Prompts for Customer Support — the queue without burning out the team
• How to Edit AI Out of Your Writing — strip AI patterns before you publish
• Prompt to Workflow: The AI Ladder — when to climb from prompt to workflow