Tools you’ll use
Your sales calls are a goldmine you’re probably ignoring. Every call, your prospects tell you exactly how they describe their problems, what words they use, what makes them hesitate, and what gets them excited. That language is more powerful than any copywriting framework because it comes straight from the people you’re trying to reach.
The problem is nobody has time to listen to every call, take notes, and organize it all. This system does it automatically. Your CRM records the call, an automation saves the transcript, and Claude Code extracts pain points, objections, buying signals, and ad-ready phrases, all in the prospect’s exact words. It runs every 5 minutes. You don’t touch it.
Set up your CRM
1. Enable call recording and transcription
In GoHighLevel, go to Settings > Phone Numbers and turn on call recording and transcription. This is what generates the raw transcript after every call. If you’re using a different CRM, you need one that records calls and can produce a text transcript. Most modern CRMs support this.
2. Create a custom field for the transcript
Add a custom contact field called Call Transcript (use “Large Text” as the field type). This is where the full transcript gets stored on the contact record so your extraction script can read it without making extra API calls.
3. Build the automation
Create a workflow that fires after a call transcript is generated. It does two things:
- Saves the transcript text into the custom field you just created
- Adds a tag like needs-extraction to the contact
The tag is the trigger. Your extraction script checks for contacts with this tag, processes them, and removes it when done. One API call to find work instead of scanning every contact.
Build the extraction
4. Write the extraction prompt
This is the most important piece. The prompt tells the AI exactly what to pull from each transcript. Here’s what to extract:
- Prospect profile - location, years in business, revenue range, team size
- Pain points - exact quotes, categorized (revenue, retention, staffing, burnout, etc.)
- Objections - what they pushed back on, in their words
- Buying signals - moments of interest or excitement
- Ad language - phrases grouped by theme that other prospects would relate to
- Outcome - qualified or not, next steps
The key is requiring exact quotes, not summaries. You want the raw words prospects use, not a cleaned-up version. “I’m trying to get my business to run without me” hits harder in an ad than “owner seeks operational independence.”
The ad language field: This is the secret weapon. It groups prospect phrases by theme (burnout, staffing, cash flow, growth) so you can write “if you’ve ever felt…” style copy using real words from real people. Every call adds to this bank automatically.
5. Set up the processing script
Tell Claude Code to build a script that:
- Searches your CRM for contacts with the trigger tag
- Reads the transcript from the custom field
- Sends it to the AI with your extraction prompt
- Stores the structured output in a local database
- Removes the tag so it doesn’t process the same call twice
The script runs on a schedule (every 5 minutes via a cron job or task scheduler). When there’s nothing to process, it checks once and exits. When a new call comes in, it picks it up automatically.
Storage: You want two layers. A structured database (like DuckDB or SQLite) for filtering and analytics. And a vector database (like Pinecone) for semantic search. Claude Code can set up both.
6. Run your first extraction
Process your most recent sales call and look at what comes back. You’re checking for:
- Are the quotes accurate? (exact words from the call, not paraphrased)
- Are pain points categorized correctly?
- Are the ad language phrases things other prospects would actually relate to?
Real output example:
- Burnout / Freedom: “I’m trying to get my business to run more without me than with me”
- Staffing: “trying to keep them corralled in the right direction”
- Cash Flow: “when leads dry up, now you’re stuck”
Use it
7. Put the language into your marketing
You’re not quoting anyone directly. You’re using their language to write copy that makes other prospects feel seen:
Ad copy from extracted language: “If you’re doing the revenue but you’re still the one holding everything together… if your team nods along in meetings but then does their own thing… if you know you need systems but you don’t have time to build them because you’re too busy running everything… you’re not alone.”
Every line in that ad came from a real call. It hits different because it sounds like a real person, not marketing.
8. Segment by qualified vs. unqualified
This is the part most people miss. Not all prospect language belongs in your ads. Pain points from qualified prospects (people who were a real fit) go into your marketing. Pain points from unqualified prospects tell you what language to avoid, because it attracts the wrong people.
Before you start
- [!] Your CRM must record and transcribe calls.
- [!] The extraction prompt matters more than the code.
- [!] Don’t expect patterns from 2 calls. Real power takes 10-15 calls to emerge.
Follow @natevanwag on IG for more AI automation breakdowns.