Claude for Meeting Minutes: The Real Prompt I Use Every Day at Work

The 6-Month Reality Check

I’ve been using Claude for meeting minutes every single workday for the past 6 months. Before that, I tried ChatGPT for the same task. Before that, Gemini. I switched to Claude and never went back.

This isn’t a “best AI tool” listicle. This is what actually happened when I tested Claude for meeting minutes on real work meetings—daily standups, client calls, technical reviews, and the occasional 90-minute brainstorm that should have been an email.

Here’s what I learned, the exact prompt I use, and why it works on a psychological level that most “prompt guides” never explain. At the end, you’ll get my Notion template—the same one I open every morning before my first meeting.

If you’ve ever pasted a meeting transcript into an AI tool and gotten back a generic summary that missed every important decision, this article is for you.

The Real Prompt I Use Every Day

Most “Claude prompt” tutorials give you something generic like “Summarize this meeting.” That’s why your output is generic too. Here’s the actual prompt I use—shaped by 6 months of trial and error on real meetings:

You are an experienced executive assistant specializing in 
meeting documentation. I will share a meeting transcript with you.

Context:
- Meeting type: [STANDUP / CLIENT CALL / BRAINSTORM / REVIEW]
- Participants: [NAMES AND ROLES]
- Duration: [TIME]
- My role: [YOUR ROLE]

Please produce meeting minutes in this exact structure:

1. **Executive Summary** (3 bullet points, max)
2. **Key Decisions Made** (with decision owner)
3. **Action Items** (format: [OWNER] — [TASK] — [DUE DATE])
4. **Open Questions / Parking Lot**
5. **Next Meeting Date** (if mentioned)

Important rules:
- Do NOT include filler conversation or small talk
- Preserve technical terms exactly as spoken
- If something is unclear, mark it as "[CLARIFY: ...]"
- Maintain a neutral, professional tone

Here is the transcript:
[PASTE TRANSCRIPT HERE]

Three things make this prompt different from the generic ones floating around the internet:

• Role assignment (“experienced executive assistant”) that primes Claude’s output style
• Structured context block that tells Claude what kind of meeting this is before it sees a single line of transcript
• Explicit verification cues (“[CLARIFY: …]”) that make hallucinations visible instead of hidden

Copy that prompt, fill in the brackets, paste your transcript, and you’ll get meeting minutes that are 80% ready to send. The remaining 20%—we’ll cover that in Section 4.

Why This Prompt Works: The Psychology Behind It

Most prompt guides stop at “here’s the prompt, good luck.” But understanding why a prompt works lets you adapt it to situations the original author never imagined. There are three psychological principles operating in the prompt above.

Principle 1: Role Assignment Triggers Context Inheritance

When you tell Claude “You are an experienced executive assistant,” you’re not just being polite. You’re activating an entire cluster of associated patterns—formal tone, structured output, attention to action items, professional vocabulary. This is called context inheritance, and it’s why “You are a senior X” prompts consistently outperform “Please act professionally” instructions.

The same principle is why our brains process information differently when we’re told someone is a “doctor” versus a “patient” before they speak. Role precedes content in human cognition. Large language models inherit this from their training data.

Principle 2: Structured Output Requests Reduce Hallucination

Asking for a free-form summary gives Claude enormous latitude—and latitude is where hallucinations live. When you specify “5 sections, in this exact order, with this exact format,” you’re constraining the output space dramatically. The model has fewer degrees of freedom to invent things.

This connects to a broader pattern I explored in how AI affects our decision-making psychology—the more structure you impose on AI output, the less it can drift into confident-sounding nonsense.

Principle 3: Explicit Failure Markers Make Errors Visible

The line “If something is unclear, mark it as [CLARIFY: …]” does something subtle but powerful. It gives Claude a graceful way to express uncertainty instead of fabricating confident-sounding filler.

Without this instruction, Claude tends to fill gaps with plausible-sounding invention. With it, you get explicit “I’m not sure about this” markers that let you verify the actual transcript. This is the same principle behind why AI confirmation bias is so hard to detect—when AI sounds confident, we stop checking.

3 Mistakes That Ruin Your Meeting Minutes

The prompt above gets you 80% of the way there. The other 20% comes from avoiding these three mistakes I’ve seen people make repeatedly when using Claude for meeting minutes.

Mistake 1: Pasting Raw Transcripts Without Context

Dumping a 5,000-word transcript into Claude with no context is like handing a stranger a stack of papers and asking them to “summarize this.” The model has no idea what matters to you.

Fix: Always fill in the context block at the top of the prompt. Even a 5-second answer like “Meeting type: client call, Participants: me and the marketing team, Duration: 30 min” dramatically improves output quality. This isn’t optional—it’s the single biggest lever for output quality.

Mistake 2: Asking “Summarize This” Instead of Specifying Structure

“Summarize this meeting” is the laziest prompt possible, and it produces lazy output. Claude will give you a paragraph that sounds smart but misses every action item.

Fix: Specify the exact sections you want, in the exact order you want them. The prompt above does this with 5 numbered sections. When you change “summarize” to “produce minutes in this exact structure,” output quality jumps measurably.

Mistake 3: Skipping the Verification Step

Claude is great. Claude also hallucinates about 5% of the time—especially with names, numbers, and direct quotes. If you send minutes to participants without verifying, you’ll eventually send something wrong. And in client calls, “wrong” can mean “embarrassing” or “expensive.”

Fix: Before sending, scan the minutes against the original transcript for:

• Names spelled correctly (especially non-English names)
• Action item owners explicitly named, not “we” or “the team”
• Due dates included for every action item
• Numbers and figures double-checked

This takes 2 minutes. It saves you from sending one wrong fact to your CEO. The full verification checklist is included in the free Notion template at the end of this article.

Claude vs ChatGPT vs Gemini for Meetings

I get this question constantly: “Why Claude? Why not ChatGPT or Gemini?” Here’s the honest answer based on running the same meeting transcripts through all three for 6 months.

Claude: Best for Nuance and Structure

Claude handles long transcripts (60+ minutes) without losing detail. It follows structured output instructions almost perfectly. When I say “use this exact format,” it actually does. Hallucination rate on names and quotes is the lowest of the three in my testing.

Best for: Client calls, technical reviews, long strategy sessions where details matter.

ChatGPT: Best for Speed and Iteration

ChatGPT is faster to respond and easier to iterate with through follow-up questions. If you need to refine output multiple times—”now reformat as bullets,” “now translate to Korean”—ChatGPT’s conversational flow is smoother.

Best for: Quick standups, brainstorms where you’ll iterate on the output.

Gemini: Best for Google Workspace Integration

If your meetings already live in Google Meet with auto-transcription, Gemini’s native integration eliminates copy-paste friction. Quality is solid, but I find it slightly less precise than Claude on structured output.

Best for: Teams already deep in Google Workspace, recurring meetings with consistent format.

For a deeper breakdown of when each model wins, I’ve covered the comparison in detail in ChatGPT vs Gemini vs Claude: Which AI Should You Actually Use? The short version: Claude wins for meeting minutes specifically because of its structured output reliability.

If you’re new to using AI for meeting documentation entirely, start with my original guide on AI meeting minutes setup—it covers the workflow from transcription to delivery.

For the record, Anthropic—Claude’s maker—specifically designed Claude with longer context windows and stronger instruction-following, which is why it handles complex meeting transcripts so well.

Bonus: Get My Notion Template (Free)

Everything in this article—the master prompt, the 5 variations for different meeting types, the verification checklist—lives in a Notion template I built for myself and now use every workday.

I’m sharing it for free because the goal of this blog is to make AI actually useful, not theoretically interesting.

Once you have the template, click “Duplicate” at the top right of the Notion page. The template will copy into your own Notion workspace, where you can customize it freely.

FAQ

Is Claude better than ChatGPT for meeting minutes specifically?

In my 6-month side-by-side testing, yes—but only for meetings where structure matters. Claude handles long transcripts with less drift, follows formatting instructions more reliably, and hallucinates less on names and direct quotes. For 5-minute standups, both work equally well. For 60-minute strategy sessions, Claude wins consistently.

Do I need Claude Pro to use this prompt?

No. The free version of Claude works for most meeting transcripts under 30 minutes. Pro becomes useful when you’re processing 60+ minute meetings daily—the longer context window and higher message limits matter at volume. Start free, upgrade only when you hit the limit.

How do I get a meeting transcript in the first place?

Three options that I’ve tested: Fireflies.ai (auto-joins meetings, generates transcripts), Otter.ai (mobile recording, fast transcription), or built-in tools like Google Meet’s auto-transcription and Zoom’s recording feature. Fireflies handles the workflow best if you want it fully automated—I’ve broken down the full setup in my original meeting minutes guide.

Can I use this prompt for non-English meetings?

Yes. Claude handles multilingual meetings well. If your meeting is in Korean, Japanese, Spanish, or any major language, just paste the transcript as-is. The prompt structure works across languages—Claude will produce minutes in whichever language the transcript is in, unless you explicitly request a different output language.

What if Claude refuses to process the transcript?

Occasionally, Claude flags transcripts that contain sensitive content (personal data, confidential business info) and asks for confirmation. This is a safety feature, not a bug. Confirm that you have authorization to process the content, and Claude will proceed. For genuinely confidential meetings, consider on-device alternatives.

The Bottom Line

Using Claude for meeting minutes isn’t magic. It’s a real prompt, applied consistently, verified before sending. After 6 months, my meeting documentation takes 5 minutes instead of 30. That’s 25 minutes per meeting × 3 meetings per day × 5 days a week = roughly 6 hours back every week.

The prompt above is the exact one I use. The Notion template gives you 5 more for different meeting types. The verification checklist saves you from sending one wrong fact to your CEO.

That’s the goal of Pick Enough: AI that actually works, in the messy real conditions of your actual workday. If this helped, the newsletter is where I share more real prompts every week—including the ones I haven’t published as full articles yet.