Steal This Workflow · video notes
One transcript, five documents: the prompt
Paste a transcript, get five documents: a recap, action items with owners and due dates, the follow-up email, a decisions log and the next meeting's agenda. Paste it into the AI chat tool you're allowed to use for this. It's plain text and follows Anthropic's guide for Claude; we haven't tested it in other AI tools.
The prompt
Copy it, replace [YOUR NAME], and paste your transcript between the tags.
You turn meeting transcripts into follow-up documents. Use ONLY what is in the transcript below. Do not add names, dates or facts that are not in it.
Write these 5 sections with these exact headings:
1. RECAP: 3-5 bullets.
2. ACTION ITEMS: a table with Task | Owner | Due | Quote. Quote = the exact line from the transcript. If no owner was clearly named, write "TBD (ask)". If no specific day or date was said, write "TBD (ask)". Never guess.
3. FOLLOW-UP EMAIL: from [YOUR NAME] to everyone, under 150 words, with a subject line. Ask about every TBD item.
4. DECISIONS LOG: only things that were explicitly agreed. Put ideas that were postponed under "Parked", not as decisions.
5. NEXT MEETING AGENDA: 3-5 items, open TBD items first.
<transcript>
[PASTE YOUR TRANSCRIPT HERE]
</transcript>Before you send anything
- Read every section before you send it. AI cuts mistakes; it doesn't remove them.
- Recording rules differ by place; some require everyone's consent. Ask first. Not legal advice.
- Don't paste confidential client information into AI tools your contract or company doesn't allow.
Why the prompt is written this way
The rules in the prompt follow Anthropic's guide on reducing hallucinations. It's one AI lab's guide, written for its own model.
- “Use ONLY what is in the transcript.” The guide: “Explicitly instruct Claude to only use information from provided documents and not its general knowledge.”
- “If no owner was clearly named, write TBD (ask). Never guess.” The guide: “Explicitly give Claude permission to admit uncertainty. This simple technique can drastically reduce false information.”
- The Quote column. The guide: “Make Claude's response auditable by having it cite quotes and sources for each of its claims.”
- You still read it. The same guide: these techniques “don't eliminate them entirely. Always validate critical information”.
Our workflow. The five documents and the prompt are our own design. We wrote it around a fictional client call.
Our test (10 Oct 2026). Claude Code with Opus 5.5, our fictional transcript, 3 runs per prompt. This prompt: all 5 documents, the unowned printer task marked TBD (ask), and an email asking about every open item, 3 of 3 runs. A plain “summarise” prompt didn't invent owners either (0 of 3). One small test on one sample, not a benchmark.
Already on Google Meet, Zoom or Teams? Each offers built-in AI meeting notes or recaps on eligible or paid plans. This prompt is for when you have a transcript and want all five documents in one go.
Sources
Every fact on this page comes from these pages, opened on 8 October 2026. Things change: check the live page before you rely on it.
- Anthropic, “Reduce hallucinations” (Claude docs)
- Google Meet Help, “Take notes for me in Google Meet”
- Zoom Support, “Using Meeting Summary with AI Companion”
- Microsoft Support, “Meeting recap in Microsoft Teams”
- 18 U.S.C. § 2511 (US federal law, one-party consent)
- California Penal Code § 632 (all-party consent for confidential communications)
A workflow, not legal advice. Hourwiz is built with Claude Code. Not sponsored by Anthropic.
No sponsor and no affiliate links on this page. Vendors are named so you can check them yourself, not as recommendations.
Follow Hourwiz
- YouTube@hourwiz (opens in a new tab)
- Instagram@hourwiz.ai (opens in a new tab)
- Facebookhourwizhq (opens in a new tab)