Free skill pack
From meeting transcript to CRM update, in one pass.
A tool-agnostic recipe for founders and sales teams who run their own pipeline. Paste a transcript into ChatGPT, Claude, or Gemini with one saved prompt and get back a tight summary, field updates backed by supporting quotes, and a follow-up task list for both sides of the call. You review, you paste, and the record finally matches the conversation.
Free download · Markdown recipe
Meeting notes to CRM workflow
A markdown recipe you can run today: prerequisites, one-time setup, the full extraction prompt, and a QA checklist that catches invented commitments before they reach your pipeline.
- Download the recipe and open it in any text editor.
- Replace the bracketed placeholders with your CRM's real field names and stage values.
- After your next meeting, paste the prompt plus the transcript into your AI assistant.
- Check the output against the QA checklist, then copy it into your CRM.
What this handles
The recipe takes one input, a meeting transcript from any video call tool, and produces four outputs: a five-bullet summary of what happened, field-by-field CRM updates, a follow-up task list, and a set of flags for anything ambiguous.
The field updates are the part most DIY prompts get wrong. This one constrains the model to your CRM's exact field names and pick-list values, so "deal stage" can only become one of your real stages. Every proposed change must carry a direct quote from the transcript as evidence. The task list captures commitments in both directions, what you promised them and what they promised you, because the second kind is what forecasts quietly depend on.
The flags section is the honest part: when the customer's stated timeline contradicts your recorded close date, the recipe surfaces the mismatch instead of papering over it.
How to run it
Setup happens once. You copy your CRM's exact field names and allowed stage values into the prompt template, delete any fields that should never be set from a transcript, and save the finished prompt where your team keeps snippets.
Then the per-meeting loop is short. Paste the prompt, paste the transcript, send. Read the field updates first and hold the line on one rule: no supporting quote, no update. Copy the summary into your notes field, apply the changes you accept, create the tasks, and give every flag a human decision.
The QA checklist in the download takes about a minute per meeting: spot-check a quote against the transcript, confirm names are spelled the way the customer spells them, and confirm no deadline appeared in the output that was never said on the call.
When to upgrade
This is a manual loop, and that is fine while call volume is low and one person owns the pipeline. It stops being fine when transcripts pile up unprocessed, when several reps each run a slightly different version of the prompt, or when you need the CRM updated without anyone remembering to do it. At that point the fix is not a better prompt. It is a system that watches for new transcripts, writes to the CRM through its API, logs every change, and routes only flagged calls to a human. That is a build, not a snippet. See what we install or book a free AI opportunity audit to find out whether your volume justifies it.
Want this running without the paste step?
We install agents that read every transcript, update the CRM through its API, and flag only the calls that need a human. Same team. Double the output.
We take on companies ready to invest $5,000+/month. Not there yet? Our free resources are genuinely free.