Replying to reviews with AI is a Skill: one saved prompt turns a batch of unanswered reviews into replies you check and post, and clearing a backlog takes one sitting of about an hour. After that, a weekly pass takes a fraction of the time. This page covers every review, good and bad, on every platform you are listed on, drafted in your own voice. For a single hard negative review on Google, where the question is whether to reply at all, read how to respond to negative Google reviews first. Here the goal is the habit: nothing left unanswered, and no reply that sounds like a form letter.
- Shape
- Skill
- Sitting
- one sitting
The shape: Skill
This job is a Skill because it is one prompt and the material under it. There is no tool to connect and no chain of steps. The AI never logs in to anything. You copy reviews in, and you paste replies out.
The AI is good at this part because most review replies fail in the same two ways. Happy reviews get a copied "Thanks so much for the kind words", over and over, until the page reads like a bot wrote it. Unhappy reviews get either silence or a reply written while the owner was still annoyed. An AI does not get bored on the tenth thank-you, and it does not get defensive on the first complaint. Given your voice and the review text, it can write ten different thank-yous and one calm reply in the time it takes you to write one.
What a Skill cannot do is post. Every review site has its own reply box, and the reply goes in by your hand. That is a feature here, not a gap, because the moment of pasting is also the moment you read it one last time.
What you give the AI
Gather these into one document before you open the chat. Put a short label above each review so the AI knows where it came from.
- Every unanswered review, with its platform: copy the text, the star rating, the reviewer's first name as it shows and the site, for example "Google, 5 stars, Maria" or "Yelp, 2 stars, Dan K.".
- Your brand voice guide, if you have one: the one-page guide from the brand voice guide job is exactly the right input here.
- Three to five review replies you wrote yourself: the ones you were happy with, good and bad, so the AI copies your habits rather than a template.
- Your name and how you sign: whether you reply as the business, as yourself or as "the team", and whether you sign with a first name.
- Safe facts for each review, where you have them: the staff member's first name the reviewer mentioned, the service they had, a fix you have already made, written in a line under that review.
- What you may never say in public: in health, legal, finance, insurance or any work with client confidentiality, write the rule out, for example "never confirm someone is a patient".
- How an unhappy customer reaches you directly: the phone number or email you want in a reply to a low-star review, for example "Call Sam at the shop: 555-0142".
- Words you never want used: "valued customer", "we apologize for any inconvenience", "your feedback is important to us", or whatever makes you wince.
Leave out anything private. Do not paste account notes, order histories, health details or a customer's contact information into the chat. The AI only needs what you would be comfortable seeing printed under the review.
For the negative reviews, do the investigation before the batch. If you have not yet checked what happened, mark that review "not checked yet" and the prompt will hold it back.
What comes back
Paste this prompt, then your labeled material below it.
You are drafting public replies to customer reviews for a small business.
The owner will check every reply and post it by hand on each site.
Use the voice guide and the sample replies below. Write as the owner writes.
Never use the words on the "never use" list.
For each review, in the order given:
1. Name the platform and the star rating on one line.
2. Write one reply. Mention one specific thing the reviewer said.
Use only facts the reviewer wrote or the owner gave in the notes.
Never add what the customer bought, their account or any private detail
beyond what the reviewer wrote, unless the owner's notes say it is safe.
Follow the owner's "never say in public" rules even when the reviewer
wrote the detail.
3. Keep positive replies to two or three sentences. No two replies may open
with the same words.
4. For a review of three stars or fewer: acknowledge the experience without
agreeing to anything not confirmed, state only the facts in the owner's
notes, and give one next step, such as a way to reach the owner directly.
Never argue, guess at motives, promise a refund or ask for the review to
be changed.
5. If a review is marked "not checked yet", or is a complaint that mentions
safety, health, discrimination, harassment, legal action or a staff member
by name, do not draft a reply. Write HOLD and one line saying why.
6. Under each reply, list any fact in it that the owner should confirm.
After all replies, list any review that looks like spam or belongs to
another business, with one line on why.What comes back is a list in the same order as your reviews, one reply under each, with a short line of facts to confirm. Some reviews come back as HOLD instead of a reply. Those are the ones that need you, not a draft, and the negative reviews guide walks through what to do with each kind.
Once the prompt works, save it with your voice guide and sample replies in a project in the AI you use, so next week's batch starts with only the new reviews.
How to judge it
Read the batch once, quickly, and reject any reply that does any of these:
- It could go under any review: if you could move a reply to a different review and it still fits, it did not read the review.
- It repeats: two replies open the same way, or three end with "We can't wait to see you again".
- It adds a fact: it names a service, a staff member, a date or a fix that is not in the review or your notes.
- It reveals something private: it confirms what someone bought, was treated for or paid, even kindly.
- It argues or over-apologizes: a negative reply that explains why the customer is wrong, or one that grovels and promises things you have not decided.
- It sounds like a hotel chain: "valued guest", "we strive for excellence", "please don't hesitate to reach out".
- It is too long for the site: a five-star reply that runs a full paragraph reads like a sales pitch.
Then read the replies aloud as a group. If they sound like one person who actually read each review, they are ready. If they sound like a script, add the phrases that bothered you to your "never use" list and run the batch again.
What stays with you
The AI can write the reply. It cannot know what happened. Whether the order was late, whether the technician was rude, whether the customer is even yours: those facts live in your records and your memory, and a reply is only as true as the notes you gave it.
The judgment on hard reviews is yours. Whether to reply in public, handle it privately, report a review to the site or say nothing is a decision, not a drafting task. The AI's HOLD line is a flag, not a verdict.
The rules of each site are yours to check. Google, Yelp, Facebook and the review sites in your own industry each set their own rules for what a business may say in a reply and whether it may ask for a review or offer anything for one. Those rules change, so read the site's current help pages rather than trusting a draft or a memory.
Your industry's rules sit above all of it. If you work in health care, law, finance or another regulated field, your duty of confidentiality decides what a reply can say, and a warm reply can still break it.
The post button is yours. No AI logs in to your profiles from this prompt, and every reply goes up by your hand, which is also your last chance to catch a wrong fact.
An editorial example
This is an invented business, used only to show the job. Say you run a small bike repair shop, listed on Google, Yelp and Facebook. You have fourteen unanswered reviews across the three: eleven happy, two middling and one angry one about a wheel that was not ready on the promised day.
You paste the reviews, your voice guide and four replies you wrote last spring. Under the angry review you add a note: "Checked: the part arrived two days late, we called on day one, and we now text when parts are delayed." Under one of the middling reviews you write "not checked yet".
The batch comes back. The eleven happy replies each mention something different: one thanks a rider for the flat she had fixed before a race, another names the mechanic the reviewer praised. The angry review gets a short reply that says the part came in late, that you called, and that you now text when a part is held up, with your direct number as the next step. The unchecked review comes back as HOLD.
You change one reply that called a customer "a valued rider", post the thirteen by hand across the three sites, and check the held review tomorrow.
That is where the example stops. Nothing in this example is a real business or a real outcome.
Where this job connects
A good review has a second life after you thank the customer for it, and reusing one review across your website, social and email covers that part. If there are not many reviews to reply to yet, start with how to get more Google reviews. The rest of the jobs are on the full list.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
