Generating Meeting Minutes Automatically: From Recording to Approved Document

Quick answer

To generate meeting minutes automatically, you record the meeting, run it through a 98.7%-accurate AI transcript, and apply a structured minutes prompt that maps speech into the seven required components: header, attendees, agenda items, motions, votes, decisions, and adjournment time. The whole loop takes about 8 minutes on a 60-minute board meeting — versus the 90-plus minutes a secretary typically spends writing them by hand.

But minutes aren’t a summary, and treating them like one is how organizations end up with documents that don’t pass a legal audit. This guide walks through the differences, the prompt that actually works, and the formatting rules you can’t shortcut.

Minutes vs. summary vs. transcript — they’re three different documents

People use “meeting minutes” loosely. In a corporate, nonprofit, government, or HOA context, they have a specific meaning that auditors, regulators, and courts care about.

A transcript captures every word spoken. It’s a raw source record — useful, but unsigned, unstructured, and not legally binding.

A summary is what most teams actually want from a casual sync: a 3–5 sentence narrative of what happened.

Minutes are a formal, structured record. They state who attended, what motions were made and by whom, who seconded them, how each member voted (or whether the vote was unanimous), what decisions were ratified, and when the meeting adjourned. The Delaware General Corporation Law §141 and most state nonprofit acts treat them as the official record of corporate action. The IRS Form 990 Part VI Section A Line 8 asks nonprofits whether they “contemporaneously document” their meetings — meaning minutes written within a reasonable window, not reconstructed months later.

Robert’s Rules of Order Newly Revised, 12th edition (the 700-plus-page parliamentary procedure standard used by an estimated 85% of U.S. nonprofit boards), is even more specific about what minutes must and must not contain. Personal opinions don’t go in. Speeches don’t go in. What was done goes in.

If you’ve been writing freeform summaries and calling them minutes, you’re not alone — but you’re also not protected. That’s the gap this article closes.

The seven components every set of minutes needs

Whether you follow Robert’s Rules, your state’s nonprofit corporation act, or a corporate bylaws template, the structure is essentially the same. Skip any one of these and the document gets weaker under audit.

7

Required components: header, attendees, agenda, motions, votes, decisions, adjournment

8 min

Average AI workflow time for a 60-minute meeting, vs. 90+ min by hand

98.7%

Atter AI transcript accuracy on clean audio — the substrate everything else runs on

90+

Languages supported, including mixed-language board calls

The seven, in order:

1. Header — organization name, meeting type (regular, special, annual), date, time, location (or video platform).
2. Attendees and quorum — present, absent (with apologies noted), and a statement that quorum was met.
3. Approval of prior minutes — what was approved, with or without amendment.
4. Agenda items discussed — each topic, in order, with a brief factual description of what was presented.
5. Motions and votes — exact text of each motion, who moved, who seconded, vote tally including abstentions, and pass/fail outcome.
6. Decisions and action items — what the body resolved to do, with owner and deadline where assigned.
7. Adjournment time, next meeting, and signature — when the meeting ended, when the next is scheduled, and the secretary’s name (signature added on approval).

A lot of AI tools nail items 1, 4, and 6 and quietly skip 2, 3, 5, and 7. That’s the part that gets organizations in trouble.

If you’re new to the broader workflow, our beginner’s guide to AI meeting transcription covers the recording-and-upload basics before you layer on the minutes prompt.

Step 1 — Capture audio that survives the agenda

Board meetings are harder to transcribe than team standups for three reasons: more speakers, more proper nouns, and more numbers. A motion that reads “Move to approve the FY26 capital budget at $2.4 million, contingent on Q3 fundraising hitting $480K” has four numbers and one acronym in 22 words — and every one of them has to be right.

Three audio rules cover most board contexts:

• Mic the chair and the secretary first. They speak the most procedural language (“I move…”, “Second”, “All in favor…”). Clean audio on those two voices makes motion extraction nearly perfect.
• Use platform local recording when available. Zoom’s “Record to this computer,” Teams’ SharePoint recording, and Webex’s local capture all produce 4–6× cleaner files than a phone in the middle of the table. For Zoom-heavy boards, see our Zoom transcription walkthrough.
• State attendance at the top of the recording. A 30-second roll call at the start of the recording gives the AI a labeled voiceprint for each member. Diarization accuracy on a 9-person board jumps from ~78% (cold) to >94% with a roll call.

Atter AI processes recordings of any length, so a 3-hour annual general meeting goes in as one file rather than being chopped into chunks where motions get split across files.

Step 2 — Transcribe before you extract

The minutes prompt only works on a clean transcript. Three transcript properties matter for minutes specifically:

• Number and dollar-figure accuracy — motions often turn on budget figures.
• Speaker labels — without them you can’t credit who moved or seconded.
• Timestamps every 10–20 seconds — so the approving body can click back to the source if a member challenges what was recorded.

Atter AI delivers all three by default at 98.7% accuracy. For the recording-to-transcript baseline workflow, how to transcribe meetings automatically covers the mechanics. The minutes layer sits on top.

Step 3 — Run the minutes-generation prompt

This is the prompt that converts a transcript into Robert’s-Rules-shaped minutes. Paste it into the AI Chat alongside the transcript:

Three things make this prompt different from a generic summary prompt:

• It forbids paraphrasing motions. A motion’s wording is legally operative — “moved to authorize up to $50,000” is not the same as “approved $50K spending.”
• It demands explicit “unclear” markers. Auditors trust documents that admit uncertainty more than ones that paper over it. The timestamp lets you re-listen in 10 seconds.
• It bans summary of discussion. Robert’s Rules is explicit: minutes record what was done, not what was said. Discussion summary belongs in a separate document.

Step 4 — Format against your governing rules

The prompt produces structured prose. Next, format it against your organization’s actual standards. Most fall into one of three buckets:

If your bylaws specify a template, use that — the AI prompt above produces a superset of what most templates need, so removing fields is easier than adding them.

Step 5 — Circulate, amend, approve

Minutes aren’t real until the body approves them at the next meeting. Three things speed that loop:

• Send within 72 hours. The IRS calls minutes “contemporaneous” if drafted within a reasonable window after the meeting. Treasury Regulation §53.4958-6 cites 60 days as the outer bound for the rebuttable presumption of reasonableness; 72 hours is the practical standard for boards that want clean approval.
• Send as a PDF, not as an editable doc. Editable formats invite line edits that fragment into “version 3 final v2.” A PDF with a single “request amendments by email” instruction collapses revision rounds.
• Track amendments separately. When a member proposes an amendment to the minutes at the next meeting, the secretary records both the original and the amended version. Don’t overwrite the prior file.

For boards that also want a separate executive summary alongside the formal minutes, our meeting summary templates cover five reusable formats that pair well — minutes for the record, summary for the audience.

Common pitfalls

Pitfall 1: Recording without disclosure. Many states require notice that the meeting is being recorded. Board chairs should announce this at the call to order and the announcement should appear in the minutes themselves. Two states (California and Florida) have particularly strict two-party consent rules.

Pitfall 2: Summarizing discussion. Tempting, especially for new secretaries. Don’t. Minutes that read like meeting recaps draw scrutiny in litigation discovery because they capture opinions that become evidence. Stick to what was decided.

Pitfall 3: Mis-attributing motions. AI mis-attribution rates jump 10–15 points when 3+ board members have similar voice profiles or when cross-talk happens during a vote. The fix is a verbal roll-call vote on contentious motions — slower, but the diarization confidence becomes nearly perfect.

Pitfall 4: Treating “unanimous” as the default. If two members are absent, the motion didn’t pass unanimously — it passed with 7 in favor, 0 against, 2 absent. Robert’s Rules is precise about this and so should your minutes be.

Pitfall 5: Including documents that should be exhibits. If a financial report runs 4 pages, attach it as an exhibit and reference it (“Exhibit A: Q1 2026 Treasurer’s Report”). Don’t paste it into the body of the minutes.

When the meeting is mixed-language

Cross-border boards — common at international nonprofits and joint-venture corporations — often switch between English and one or more other languages. Atter AI handles mixed-language calls across 90+ languages and can render the final minutes in whichever language your governing documents require, with the original-language quotes preserved as footnotes for motions.

A French-Canadian nonprofit board that holds bilingual meetings, for example, can ship minutes in French as the official record with English glosses for any English-language motions — meeting both Quebec’s French-language requirements and the board’s working preferences.

For teams running the same workflow on action-item-heavy calls instead, how to extract action items with AI covers the extraction prompt that pairs with this minutes prompt.

FAQ

Are AI-generated minutes legally valid?

Yes — the legal validity of minutes depends on accuracy, structure, and formal approval by the body that met, not on who or what drafted them. AI is a drafting tool the same way a typewriter or Word is. The secretary remains responsible for verifying accuracy before circulation and the board remains responsible for approval at the next meeting.

How do AI minutes differ from a meeting summary?

A summary is a narrative recap, typically 3–5 sentences, that captures the gist of the conversation. Minutes are a structured legal record that includes attendance, motions, votes, and decisions in a prescribed order. Most boards need both — minutes for the record, a summary for the wider distribution list.

What about confidentiality and executive session?

Executive session content should not be transcribed or appear in regular minutes. A common pattern: stop the recording when the chair calls executive session, restart it on return to open session, and note in the minutes that “the board entered executive session at HH:MM to discuss [permitted topic] and returned at HH:MM.” Atter AI does not use uploaded recordings to train models, so even routine board content stays private to your account.

How accurate is AI on motion language specifically?

On clean audio with a clearly enunciated motion (“I move that we authorize the executive director to sign the lease on the East Avenue property for a term not to exceed five years”), verbatim transcription accuracy is above 97%. The errors that do occur tend to be in proper nouns (street names, vendor names) — which are easier to catch and correct than misheard motion verbs.

Can the AI tell the difference between a motion and a discussion comment?

In most cases, yes — board members use predictable phrasing (“I move…”, “I second…”, “Move to amend…”). The minutes prompt above is tuned to surface those patterns. Edge cases (informal motions in committee meetings, motions made without the “I move” preface) sometimes get missed and need a verification pass from the secretary.

How long does the whole workflow actually take?

For a 60-minute meeting: upload (1–2 min), transcript ready (under 5 min for an audio file at clean quality), minutes prompt (10 sec), secretary verification pass on motions and attendance (3–5 min), PDF export and circulation (1–2 min). Total: about 8 minutes of secretary time, versus the 90 minutes that the same person typically spent writing minutes by hand. On a 3-hour annual general meeting, the proportional savings are larger.

What if our board uses a custom template?

Take the AI-generated minutes and copy them into your template structure. The prompt produces all standard Robert’s Rules fields, so any custom template is either a subset (delete extra fields) or asks for one or two custom fields (add a follow-up prompt: “Also list any conflicts of interest disclosed during the meeting”). Atter AI’s AI Chat lets you iterate on the prompt without re-uploading the recording.

Is my board recording used to train AI models?

No. Atter AI does not use uploaded recordings to train models, and recordings remain private to your account. For boards governed by GDPR, HIPAA, or organization-specific data residency requirements, run the file through your standard data classification review before upload.

How much does Atter AI cost for a board secretary’s workflow?

$6.99/week, $49.99/year, or $129.99 one-time lifetime, with a 3-day free trial that requires no credit card. There are no per-minute charges, so a board that meets monthly for 2-hour sessions and holds an annual 4-hour AGM costs the same as one that meets weekly — the long-meeting case is where flat pricing matters most.