# Review Response Pack

Draft public replies to customer reviews in your own voice, at whatever volume the week throws at you, and catch the complaints that need a human before anything gets posted. Works with any capable AI assistant (ChatGPT, Claude, Gemini). No plugins, no integrations, no code.

## What you need

- Access to wherever your reviews land: Google Business Profile, Yelp, Facebook, industry sites.
- Three real replies you actually wrote and still like. They teach the model your voice better than any list of adjectives.
- Any capable AI assistant. The recipe is plain text and does not care which one.

## One-time setup

1. **Fill in the tone rules template below.** The example replies matter most. "Friendly but professional" describes every business on earth; your real replies describe only yours.
2. **Set your escalation list.** Decide which review types never get an AI-drafted reply posted without a human decision. The prompt ships with a starter list; add your own.
3. **Assemble and save the prompt.** Paste your finished tone rules into the drafting prompt, then save the whole thing wherever your team keeps snippets. Reusing the same prompt every time is what keeps the voice consistent.

## The tone rules template

Fill this in once, then paste it into the prompt below.

```
OUR VOICE
- We sound like: [e.g., the owner replying personally: warm, brief, direct]
- Words and phrases we actually use: [e.g., "thanks for coming in"]
- Words we never use: [e.g., "valued customer", "we apologize for any
  inconvenience", any corporate filler]
- Sign-off: [e.g., "- Maria, Owner"]
- Length: [e.g., 2-3 sentences for positive reviews, 4-6 for negative]
- Three real replies we have written and liked:
  1. [paste]
  2. [paste]
  3. [paste]
```

## The drafting prompt

```
You draft public replies to customer reviews for [BUSINESS NAME], a
[what you do] in [city or region]. I will paste reviews with their
star ratings. Match our voice exactly.

[PASTE YOUR FILLED-IN TONE RULES HERE]

Hard rules for every reply:
- Reference one specific detail from the review so the reply reads
  written, not templated. Never invent a detail the review does not
  contain.
- Never argue with the reviewer or dispute their account in public.
- Never admit fault, promise a refund or compensation, or offer a
  discount in a public reply.
- Never confirm private details (visit dates, order contents, medical
  or financial specifics), even if the reviewer states them first.
- For genuine problems, move the conversation offline: name one
  contact person and one channel.

ESCALATION. If a review mentions any of the following, do not draft a
standard reply. Output "ESCALATE:" plus the trigger, plus a short
neutral holding reply a human can approve:
- Threats of legal action, regulators, or media
- Safety incidents, injury, or health harm
- Accusations against a named employee
- Claims of discrimination or harassment
- A billing dispute over a specific amount
- Signs the review is fake (details that match no real customer)
- [Add your own]

For each review, return exactly:
1. READ - one line: what this reviewer actually wants
   (acknowledgment, a fix, money, an audience).
2. DRAFT REPLY - ready to post, in our voice.
3. INTERNAL NOTE - one line: anything the team should actually fix,
   or "none".

Reviews:
[PASTE REVIEWS WITH STAR RATINGS HERE]
```

## Run it

1. Weekly, or whenever reviews land, paste the saved prompt plus the new reviews. Batch them; the model handles ten as easily as one.
2. Anything marked ESCALATE goes to the owner or manager, not the reply box. The holding reply buys time; the decision stays human.
3. Edit each draft until it sounds like you on a good day, then post it.
4. Skim the internal notes. Three reviews mentioning slow checkout is an operations finding, not a writing problem.

## QA checklist

Do this every batch. It takes a minute.

- Read each draft aloud. If you would not say it across the counter, fix the tone rules, not just this one reply.
- No draft admits fault, promises money, or confirms a private detail.
- The specific detail in each reply actually appears in the review.
- Every ESCALATE got a human decision before anything was posted.
- Replies posted the same day do not open with the same sentence.

## When this outgrows DIY

This is a manual loop, and it is the right shape at one location and a handful of reviews a week. It stops being the right shape when replies lag by days, when multiple locations mean multiple people drifting the voice in different directions, or when nobody notices the same complaint recurring all month. At that point the fix is not a better prompt. It is a system that watches every platform, drafts in your voice, routes escalations to the right person automatically, and reports the patterns. That is a build, not a snippet.

---

Free recipe from agentclaw (agentclawhq.com). Use it, share it, adapt it.