The AI turned one $45 favor into a $4,680 policy.

$45. That's the late-cancel fee one groomer waived for one customer, one time.

The AI read that text thread and wrote a public FAQ answer: "Running late or need to cancel? Just let us know and we'll waive the fee." Nobody had decided that. The prompt did.

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 "Automatically Create a Knowledge Base from Slack Support Threads," a workflow from How I AI (hosted by Claire Vo, written up by the ChatPRD team), with Al Chen (field engineering, Galileo). Full episode.

What the original does

Al Chen's team answers customers in shared Slack channels. When a thread is solved, a support tool called Pylon turns it into a draft help article in one click. It strips out customer names and details. Someone checks the draft and publishes it to the public knowledge base, with an internal link back to the thread. The rule we liked most: "Threads with unresolved guesses produce weak articles."

The test

Composite (made-up) data for a one-van mobile dog groomer. Test clock: Wednesday, Oct 7, 2026. Most small shops don't run Slack, so we used what they do have: 5 solved customer text threads.

• Prices: $85 for dogs under 40 lb, $110 for 40 lb and up, plus $15 for matted coats.
• Late-cancel fee: $45 for cancelling under 24 hours before the appointment.
• Last quarter: 26 late cancels. At $45 each, that's 26 × $45 = $1,170 a quarter, or 4 × $1,170 = $4,680 a year.

We used a short prompt like the one on the page: make it a help article, keep it general, remove customer details. All 5 came back clean and friendly. Then we checked them against the threads. Three problems:

1. A favor became policy: in one thread the owner wrote, "First time, so I'll skip the fee." The FAQ dropped "first time" and promised a waived fee to everyone. If every late cancel cites that page, the fee is gone. That's $4,680 a year.
2. A gate code stayed in: "Side gate, code 4417" survived the cleanup. The customer's name was gone. The thing that opens her yard wasn't.
3. The deadline went missing: the cancel answer never said "24 hours." A rule without a deadline is a suggestion.

A favor you do once becomes a policy you owe everyone.

To be fair, the page's own troubleshooting notes warn about exactly this. They say to separate one customer's workaround from what applies to everyone, and to run a redaction check before publishing. The short prompt does neither. The fix was to give the AI our real policies and make it list every exception before it writes a word.

Verdict: 7/10

Strengths: turning answers you already gave into an FAQ is one of the best uses of AI for a small shop. "Only use solved threads" is a great rule. A link back to the source makes fixes easy.

Weakness: it's built for a software company with Slack channels and a paid support tool. The quick prompt generalizes everything, including the favors. Privacy is left to whoever reviews it.

Founder tweak: paste your standing policies next to every thread. Make the AI flag exceptions and stop when the thread and the policy disagree. Add a redaction list made for your trade: gate codes, alarm codes, pet names, addresses.

Copy the MCAFax version

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

You turn solved customer conversations into FAQ answers for my small business.
I will paste my standing policies, then one conversation (text, email, DM, or call notes) that ended with a clear answer.
Before you write, list:
1. The customer's real question, in their words.
2. The final answer I gave, quoted.
3. Anything I did as a one-time favor or exception (waived fee, discount, special time). Never turn these into policy.
4. Every price, fee, deadline, and rule mentioned, with the exact number.
Then write the FAQ entry:
- A question title the way a customer would ask it.
- A 2-3 sentence answer based only on my standing policies.
- Steps, with the real deadlines and fees.
- One line starting "This doesn't apply when:".
Remove names, addresses, phone numbers, gate and alarm codes, pet and kid names, order numbers, and payment details.
If the conversation and my standing policies disagree, stop and ask me which is right. Never guess.
My standing policies: [paste prices, fees, cancel rules, service area]
Conversation: [paste here]

Paste into Claude, ChatGPT, Gemini, or Copilot.

Before you post your next FAQ answer

1. Write down your standing policies once: prices, fees, deadlines, service area.
2. Pick 5 questions you answered more than twice this month.
3. Run each solved conversation through the MCAFax version.
4. Search every draft for numbers, names, and codes before it goes live.
5. Check that every fee comes with its deadline.
6. Then ask happy customers for reviews: automate review requests and catch bad ones fast (more prompts).

FTC on protecting customer information.

SBA on keeping customer data safe.