Blog · 5 min read

Replying to Reviews and Customer Messages Without Sounding Like a Bot

Canned replies read as canned. Here's how to use AI for customer messages and review responses so the result still sounds like a person who runs the business.

If you sell anything online, a large share of the working day goes not to selling but to answering. Where's my order. Is this in stock. Why is yours more expensive. Can I get a discount.

The obvious fix is templates. The obvious result is that within a month they read like an autoresponder, and customers can tell. It shows up worst on negative reviews, where "Thank you for your feedback, we'll take it into account" reads as a brush-off even when you genuinely intend to fix the problem.

AI helps here, but not in the way people usually try. Asking it to write the reply and pasting the output is just swapping one template for a slightly longer one.

A template is a shape, not a text

The useful move is to ask for several different approaches to the same situation, not one polished answer.

Customers often ask: "why is yours more expensive than [competitor]?"

Write 4 different replies. Not rephrasings of each other — genuinely
different angles: one about material quality, one about what happens
after the sale, one that honestly admits we're not the cheapest.

2-3 sentences each, conversational, no corporate filler,
and don't use the word "unique".

Four options do two things. You can rotate them, so your inbox stops looking automated. More importantly, reading all four usually tells you which angle is actually yours — and after that you write your own version faster than the model would.

Negative reviews: you're not writing to the author

A public negative review is its own genre, and the key thing about it is this: you're not writing to the person who left it. They're already annoyed, and winning an argument in a comment thread is close to impossible. The reply is read by the next fifty people who land on your product page.

That sets the criteria. A stranger reading it should see: the problem acknowledged rather than denied, a concrete action rather than a promise to "look into it", and a calm tone even if the customer is wrong.

A customer left this review: "[paste the review]"

Write a public reply from the business. Calm, no defensiveness, acknowledge
the problem if it's real, offer a concrete fix. 2-4 sentences.
Don't blame the customer even if they're wrong — other potential
customers are reading this.

The hardest case is when the customer really is wrong: they ordered the wrong size, didn't read the description, used it in a way the product isn't for. The urge to explain that publicly is strong and almost always loses. An onlooker doesn't see you being right; they see a seller arguing with a buyer. A short reply offering to sort it out in private looks far stronger.

Private complaints are a different job

In a private message there's no audience, so the logic flips. Here you should dig into details, and here it's appropriate to name a specific remedy — a refund, a replacement, a redo.

A customer sent this complaint: "[paste the complaint]"

Write a reply: first acknowledge the problem without excuses, then a concrete
fix (specify which fits: refund / replacement / discount / redo), then
an offer to make it right. Tone: genuine, no canned phrases like
"your feedback is important to us".

Banning specific phrases works better than asking for "natural language". Models drift toward corporate register by default, and the fastest way to pull them out is to name what they may not write.

The other half: asking for reviews at all

For most small businesses the problem isn't negative reviews. It's silence. Happy customers say nothing, unhappy ones post, and the page ends up looking worse than the business actually is.

Asking feels awkward, so it either doesn't happen or it happens badly. Two common failure modes: a bare "please leave a review" with no context, or the opposite — "your opinion means so much to us, it really supports a small business".

Write a short message I send [X days] after someone buys [product],
asking for a review.

Genuine, not pushy, no manipulation ("please, we really need this").
Include a placeholder [REVIEW LINK] and a short thank-you regardless
of whether they leave one.

Two practical details matter more than the wording. Timing: not on delivery day, but a few days later, once they've actually used the thing. And one ask, not a sequence — a second message repeating the same request annoys people and taints the impression of everything else.

Build the set once instead of generating every time

Rather than running a prompt per message, it's worth spending one hour building a small set.

Write down the ten questions you actually get most — not hypothetical ones, but from your last month of messages. Get three or four replies for each. Read them, throw out the ones that don't sound like you, keep the rest in your phone's notes.

After that AI barely enters daily correspondence. It comes back for the unusual cases, and those are honestly under a tenth of the volume.

Two things worth checking every time

Promises you didn't make. Models happily add "we'll refund within 24 hours" or "we'll replace it free of charge" because that's the convention of the genre. If your terms differ, that's a future argument you will lose, and it will be your fault.

Over-politeness. Left alone, models drift into a formal register — "we sincerely apologise for any inconvenience caused". In small-business correspondence that reads as borrowed: it's how a bank writes, not a workshop or a shop. Shorter and plainer almost always wins.

More on this: prompts for small business owners — the same fill-in-the-blanks approach across five areas of the business. And what AI actually saves time on, which is honest about where it doesn't.

There's a free demo with 15 of the prompts if you want to try one on a real message before deciding anything — no signup.

Also: pricing and true cost — the numbers behind the "too expensive" conversation.