The AI called my call "warm and professional." I lost the $1,860 job.

"Let me think about it and call you back."

That's how a $1,860 brake job walked out of an auto shop on a Tuesday. The AI coach graded the call: "Warm and professional. Strong rapport." It never noticed that nobody offered a time to drop off the car.

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 Build an Automated AI Sales Coaching System from Call Transcripts," a workflow from How I AI (hosted by Claire Vo, written up by the ChatPRD team), with Matt Britton (CEO, Suzy). Original workflow on How I AI. Full episode.

What the original does

Matt Britton's team turns each sales call transcript into private coaching for the rep. ChatGPT checks the call against a written rubric of observable behavior, like objection handling and who owns the next step. It returns two strengths, one improvement and a next-call practice, with timestamps. Zapier sends it privately in Slack. The rule we liked most: "Validate diarization, cite timestamps, and suppress coaching when speaker identity is uncertain." In plain words: if you can't tell who said what, don't coach.

The test

Composite (made-up) data for a two-bay auto repair shop where the owner answers the phone. Test clock: Thursday, Oct 8, 2026. Callers hear that the line is recorded, and the phone app makes the transcripts.

• One week: 12 recorded estimate calls. 7 booked, 5 didn't.
• The 5 quotes that didn't book: $1,860 brakes and rotors, $940 timing belt, $415 alternator, $325 AC recharge, $180 diagnostic. That's $1,860 + $940 + $415 + $325 + $180 = $3,720 in one week.

We started with a short prompt like the one on the page: act as a coach, give me strengths and one improvement. Every note came back kind and confident. Then we checked the calls. Three problems:

1. It coached the wrong person: the transcript swapped speakers on one call. The AI told the owner to stop saying "That's more than the car's worth." The customer said that.
2. It graded style: on 3 calls it said to "sound more energetic." You can't practice that by Monday.
3. It missed the money: on all 5 calls that didn't book, the owner never offered a drop-off time. On 6 of the 7 that booked, he did. The AI praised the "rapport" on all 5.

A friendly call that doesn't book is still a lost job.

To be fair, the page's own notes warn about the first two: stick to observable behavior, and skip calls where speakers are unclear. The short prompt does neither without a rubric. So we wrote one for a shop phone and reran the calls. It skipped the mislabeled call, dropped the style notes, and flagged the missing drop-off time on all 5.

Verdict: 8/10

Strengths: one improvement per call is a fix you can practice. Timestamps make every claim checkable. The privacy rules are strong: coaching stays private and is never used alone for raises or discipline.

Weakness: it's built for sales teams with Slack, Zapier and clean speaker labels. Without your own rubric, the AI falls back on vibes.

Founder tweak: write a 5-line rubric for your own phone calls, and make "offered a specific time" one of the lines. Tell the AI to skip any call where it can't tell who's talking.

Copy the MCAFax version

This is our own rewrite, not the original prompt. Copy the original from Matt Britton on How I AI

You coach me on my own customer phone calls for my small business.
I will paste my call rubric, then one call transcript.
First, check the speaker labels. If you can't tell who said what, say "Speakers unclear" and stop.
For each rubric line, answer Yes, No, or Not on this call, and quote the moment with its timestamp.
Then give me:
1. Two things I did well, with timestamps.
2. One thing to fix, the one most likely to cost me the job.
3. One exact sentence I can say on my next call.
4. Your confidence: high, medium, or low, and why.
Coach only what I said or did. Never comment on my voice, accent, energy, or personality.
Never invent prices, times, or promises that aren't in the transcript.
My rubric: [example: asked what the problem is / gave the price with parts and labor / offered two drop-off times / said the warranty / confirmed the next step]
Transcript: [paste here]

Paste into Claude, ChatGPT, Gemini, or Copilot.

Before you review your next call

1. Tell callers the line is recorded. Federal law allows one-party consent, but some states require everyone on the call to agree. This is general information, not legal advice.
2. Write your 5-line rubric. Put "offered a specific time" on it.
3. Run one week of calls through the MCAFax version.
4. Count how many unbooked calls got a time offered. Ours was 0 of 5.
5. If someone else answers your phone, keep their notes private.
6. Then chase the quotes that didn't book with Automatic Lead Follow-Up Without a CRM. There are more prompts here.

Sources:

• How I AI: How to Build an Automated AI Sales Coaching System from Call Transcripts
• How I AI: Full episode with Matt Britton
• 18 U.S.C. § 2511, federal wiretap law and one-party consent (Cornell LII)
• SBA: Strengthen your cybersecurity
• FTC: Protecting Personal Information, A Guide for Business