"Top search term from your calls: saltwater pool conversion. Suggested test budget: $400."
That's what Tamsin, owner of Brindlecove Pool Care (a made-up pool service in Mossbridge, a made-up town), got back when she gave an AI 12 recorded sales calls and asked which Google ads to run with her $900 November test budget.
She almost said yes. Then she checked: one of the 12 callers asked about salt. He just asked a lot.
Count callers, not mentions.
AI Prompts, Reviewed takes one prompt or workflow from How I AI, Claire Vo's show where guests share their screens and walk through how they really use AI. We run it on a small-business job, score it, and tell you what we'd change.
Credit: Based on "How to Automatically Generate Marketing Content from Customer Calls," a workflow from How I AI (hosted by Claire Vo, written up by the ChatPRD team), shown by Matt Britton, Suzy's founder and CEO, in the episode "How Suzy's CEO Turns 25,000 Hours of Sales Calls into Automated Marketing and Coaching with One Zapier Workflow." Original workflow: https://www.chatprd.ai/how-i-ai/workflows/how-to-automatically-generate-marketing-content-from-customer-calls. Full episode: https://www.chatprd.ai/how-i-ai/sales-calls-into-automated-marketing-and-coaching-with-ai-zapier. Guest site: none listed on the How I AI workflow page.
What the original does
Matt turns the words customers use on sales calls into paid-search ideas and blog drafts, with ChatGPT and Zapier. Five steps: pull problem language from call summaries, propose ad keywords, strip anything that identifies the customer, draft an article, and hold it all for review. Nothing goes live and no ad money moves on its own.
The best line is in its "if it goes sideways" notes: "One call becomes a claim about the whole market." That's exactly what happened to Tamsin.
The test
Everything here is made up. Brindlecove cleans pools in Mossbridge. Tamsin recorded 12 sales calls from September 2026, with each caller's OK.
Run 1, the shortcut: "Find the search terms my customers use, ranked by how often they come up, and split $900 across the top 3."
The AI counted mentions:
- Salt cell / saltwater conversion: 12 mentions
- Green pool cleanup: 9 mentions
- Weekly service price: 6 mentions
It split the money by share of 27 mentions: $400 to salt, $300 to green pools, $200 to price.
Now count callers. Salt: 1. Green pool: 7. Price: 5. And Brindlecove doesn't even do salt conversions. It refers them out.
It also drafted a blog post about "a retired airline pilot on Heron Lane whose pool turned green 2 days after the storm." No name. Everyone on Heron Lane would know who it was.
The rerun, with the original's steps
Run 2: the How I AI approach. Count distinct callers per phrase, label anything from one call as a single observation, and remove details that could point back to one customer, not just names.
Salt came back "single call, not a pattern." The $900 split across the 2 real patterns by caller count: 7 of 12 is $525 for green pool cleanup, 5 of 12 is $375 for weekly price. Salt got $0.
The pilot became "a homeowner whose pool turned green after a storm."
One catch: the original also strips geography. For a local business, the town is the keyword. "Green pool cleanup Mossbridge" is the ad. Keep your service area. Drop the street.
Verdict: 8/10
Strengths: it counts accounts, not mentions. It catches the combo of details that names someone without a name. It bans made-up quotes and results. And it only proposes, it never spends.
Weakness: it's built for a company with privacy, editorial, and ad owners. At a small shop, that's all you. It also has no rule for splitting a small budget.
Founder tweak: put a "callers" column next to every keyword, spend only on phrases from 3 or more callers, and keep the town while cutting the street.
Copy the MCAFax version
This is our own rewrite, not the original. Copy the original from Matt on How I AI: https://www.chatprd.ai/how-i-ai/workflows/how-to-automatically-generate-marketing-content-from-customer-calls
You help [business name], a [type of business] serving [town or area].
I'll paste summaries of recorded sales calls (callers agreed to recording).
Step 1. List every problem or request in the customer's own words.
For each phrase: number of distinct callers, number of mentions.
Rank by distinct callers, never by mentions.
Mark any phrase from 1 caller as SINGLE CALL.
Step 2. Flag anything we don't sell: [services we don't offer].
Step 3. For phrases from 3+ callers, propose ad keywords.
Add our town to each. Suggest a split of $[budget] by caller count.
Show the math. Proposal only.
Step 4. Draft one blog post on the top problem.
Remove names, streets, jobs, exact dates, and any detail only one customer has.
Keep our town.
No made-up quotes, numbers, or results.
Step 5. List everything you removed, so I can check it.
Paste into Claude, ChatGPT, or Gemini with your call summaries.
Before you trust it
- Ask for a "distinct callers" column. If it's missing, don't spend.
- Cross out any keyword for a service you don't sell.
- Read the blog draft as the customer. Would they spot themselves?
- Keep your town in the ad. Cut the street from the story.
- Turn the draft into social posts: https://mcafax.com/notebook/prompts/social-media-content-assistant
Need a place to publish it? https://mcafax.com/notebook/prompts/free-prompt-add-a-blog-admin-content-system-to-your-business-website
More owner prompts: https://mcafax.com/notebook/prompts