Home / Online sellers / Reply to reviews without promising what you can't give

Reply to reviews without promising what you can't give

Who it is for
Online store owners who answer their own reviews.
What you give it
Your reviews, a spreadsheet of them, or a public review page.
What you get back
A reply for each review, matched to its tone, with risky ones flagged for you.

Download free · View on GitHub

A skill you install once (works in Claude, ChatGPT and other assistants), a short guide, and a paste-in version if your plan has no skills.

Free. MIT licence: use it, change it, share it.

Before and after

A 2★ Judge.me review of a men's cologne from a grooming brand:

Before: "…1/3 of the product had leaked into the packaging during shipping. Good product. Poor packaging." (no store reply)

After (real output, product name replaced with [Product]): "A third of the cologne ending up in the packaging is a rough way to meet [Product], and we're sorry the shipping overshadowed a scent you liked. Contact us at [support email] and we'll [REMEDY: replacement / refund / store credit; owner to choose]."

In the same 14-review batch, the workflow also:

  • caught a 1★ review with glowing text as a probable mis-tap, without asking publicly for a rating change;
  • flagged an "opened, unhygienic products" plus "intentional damage" review for the owner instead of arguing in public;
  • replied once to a review that had been posted on two products.

See it run

Animated demo of Reply to reviews without promising what you can't give on a real example: the input goes in and the finished result appears. The still shows the final result.

What it does

A free workflow that drafts replies to your store's reviews. Paste in reviews from Judge.me, Shopify Product Reviews, Yotpo, Okendo, Google or Trustpilot, and you get back a reply for each one, ready to paste.

  • Happy customers get thanked for the specific thing they said, not "Thanks for your review!"
  • Unhappy customers get an acknowledgement in their own words, one concrete next step, and an invitation to a private channel.
  • It never commits you to anything. It doesn't admit fault or offer refunds, credits or discounts you didn't approve. Those go in a [REMEDY …] placeholder for you to decide.
  • It flags reviews that need a human: allergic reaction or injury, contamination, legal threats, chargebacks ("went to my bank"), accusations. Each gets a safe holding reply.
  • 4 or more reviews come back as one table, followed by a "Needs a human" list and every placeholder you need to fill in.

The evidence

BENCHMARKED badge. Measured against a plain AI request on cases it was not tuned on. The numbers are published.One-line request: 4 of 20 right. With it: 20 of 20.

Strong Claude modelCheapest Claude modelChatGPT free
Typing one line4 of 200 of 203 of 10
With this workflow20 of 2019 of 2010 of 10

Cases fully right. Claude: runs out of 20. ChatGPT free: one run, out of 10.

The most common one-line failure was a reply far too long for a public review (16 of 20), and 4 of 20 promised a refund nobody had offered.

Cases written by us. Run on two Claude models and on ChatGPT free. A small test, not our full training method.

An example of the one-line failure
We'd like to look into this properly and make it right, starting with a full refund.

One sentence from a one-line reply to a review about a child's rash. The owner in this case had offered no refund.

Limits

It won't:

  • Post anything. You paste the replies yourself.
  • Invent policies, restock dates or causes.
  • Ask a Trustpilot reviewer to change their rating.

Questions

Refunds and contact

If it does not do the job, ask and we refund you within 30 days. Reply to your receipt, or use the contact page.

Contact page

Which assistants does it work with?

Works in ChatGPT, Claude, Gemini and most other AI assistants: paste in the text, or load the files if your assistant supports them. Setup steps are under For technical users. We run our release tests on Claude. Use your assistant's strongest model: in our tests, a small, fast model added facts that were not in the source data. No scripts. It works from reviews you paste, a CSV, or a public review page.

For technical users

Using it with an AI assistant

  1. Unzip the free download. It holds the free skill folders, including review-reply-drafter, ready-made zips for claude.ai in a claude-ai folder, and a README. Or copy the folder from GitHub.
  2. Put the review-reply-drafter folder where your assistant looks for skills. The generic path is .agents/skills/ (or npx skills add from a folder). Tool-specific paths:
    • Claude Code: ~/.claude/skills/ (or .claude/skills/ inside one project)
    • OpenAI Codex, Gemini CLI, Cursor and others: .agents/skills/ or the skills folder their docs name
    • claude.ai: Customize, Skills, then the plus button, Create skill, Upload a skill, and upload the zip from the claude-ai folder without unzipping it (code execution must be on in Settings, Capabilities)
    • ChatGPT or any chatbot: open the skill's prompt.md, paste it into the chat, then paste your own text or link
    • Steps for each tool, with links to its official docs: how to install skills
  3. Ask for the job in plain words. The skill loads by itself.
Copy the text for ChatGPT or any chatbot

Jobs that go with this