New Prompts Worth Keeping Prompt Quick <5 min Mar 2026

The "Extract & Structure" Megaprompt

Stop asking Claude to summarize. This is the prompt that turns messy transcripts into database-ready structured data.

What you'll get out of this: A reusable prompt structure you can drop into any Claude conversation to pull clean, structured output from meeting notes, call transcripts, or any dense document — without writing JSON schema from scratch.

Why "summarize this" isn't good enough

When you ask Claude to summarize a meeting, you get a summary. That's fine for reading. It's useless for doing anything with the output downstream — dropping it into a tracker, populating a CRM field, feeding it into another workflow.

The difference between a summary and structured data is the difference between something you read and something you can use. This prompt is built for the second thing.

Use this when

Good fit

You have a meeting transcript or dense notes
Multiple people with different action items
You need output you can drop into a tracker
You're feeding results into another workflow
You need consistent structure across many docs

Not the right fit

You just want a quick summary to read
The document is already structured
There are no clear decisions or commitments
You need nuanced analysis, not extraction
One person, no coordination needed

The prompt

Copy this exactly. The sections in [brackets] are the only things you need to change per use case.

Prompt · Copy and adapt

# ROLE You are a precise data extraction assistant. Your job is to read unstructured text and return clean, structured output — nothing more. # INPUT Here is the raw content to extract from: """ [PASTE YOUR MEETING NOTES / TRANSCRIPT / DOCUMENT HERE] """ # TASK Extract all [ACTION ITEMS / KEY DECISIONS / OPEN QUESTIONS] from the above. # OUTPUT FORMAT Return ONLY a JSON array. Each object must include these exact fields: { "item": string — the action item or decision in one clear sentence, "owner": string — the person responsible (use "TBD" if unclear), "deadline": string — the due date or timeframe (use "Not specified" if unclear), "priority": string — one of: High / Medium / Low, "notes": string — any relevant context, or empty string if none } # RULES - Do not include any text outside the JSON array - Do not add commentary, preamble, or a summary - Do not guess at owners or deadlines — use the fallback values - If nothing of the requested type exists in the content, return an empty array: []

The "Do not guess" rule is the most important line. Without it, Claude will fill in owners and deadlines it doesn't actually know — and confident-sounding fabrications are worse than empty fields. Be explicit that the fallback values are the right answer when information isn't there.

How to adapt it

The structure works for anything where you need consistent output from messy input. Change the bracketed sections:

[ACTION ITEMS / KEY DECISIONS / OPEN QUESTIONS] — swap for whatever you're extracting. "Objections raised by the client", "commitments made", "risks mentioned" all work the same way.
The JSON fields — add or remove fields to match where the data is going. If you're feeding this into Airtable, match the field names exactly to your column names.
The priority field — if you don't need it, delete it. Fewer fields means fewer opportunities for Claude to fabricate something plausible-sounding.

Real example output

From a 45-minute CRE deal meeting transcript, this prompt returned 11 action items, correctly attributed to four different owners, with deadlines pulled verbatim from the conversation. Zero fabricated owners. Three "TBD" deadline fields where timing genuinely wasn't discussed.

That last part matters — the fallback values tell you where your meeting actually had gaps, which is useful information on its own.

What this pairs well with

The Meeting notes → Airtable workflow (feed the JSON output directly into that pipeline)
Any Airtable automation that expects structured input
A second Claude call that takes the JSON and drafts a follow-up email per owner

Found this useful?

Tell me what's slowing you down.

If you're copying and pasting the same thing more than twice a week, there's probably a workflow for it. I build these things — ask me.

Tell me what's broken