How replies are written

Reply Setup asks one question up front: what drafts your replies?

Brand Voice is the guided way. You fill in reply style, example replies, reply rules, words to avoid, and keywords. Most teams should use this.

Custom Prompt hands you the instructions directly: your role, how to think about each review, always do, never do, and an example. It is for people who know exactly what they want and would rather write it than pick it.

Only one is active at a time, and switching is one click. There are presets to start from (warm and concise, local business, upsell on five-star), which are quicker to edit than a blank box is to fill.

A draft is written for one review, using:

• the review text, its rating, and the language it was left in
• what the classifier found: sentiment, urgency, intent, and what the reviewer actually mentioned
• your voice profile or custom prompt
• the app or listing it belongs to, including any features you listed for it

That last one matters more than it sounds. A reply that answers the complaint the person actually made reads like a person wrote it. A reply that thanks them for their valuable feedback reads like nobody did.

Voice can be set account-wide or per app, which is what you want if you run a banking app and a puzzle game off the same account.

The fields that earn their keep:

• Example replies. Two or three replies you were genuinely happy with teach it more than any slider.
• Words to avoid. The phrases your team has banned. They will otherwise turn up.
• Reply rules. The things that are always or never true for you.

There is a preview, so you can see what your settings do to a real review before you save. If you are stuck on a blank form, Generate with AI drafts a starting brief from what it knows about your app, and you edit from there.

You can tell the drafter which keywords matter for an app or listing. It works them in where they fit a sentence and leaves them out where they do not.

Two rules keep this honest:

• One promotional mention per reply, at most. A keyword or a feature, not both, not three.
• If a sentence only exists to hold the keyword, the keyword loses.

Stuffing keywords into review replies is obvious to readers, and it is the fastest way to make a reply look automated. The point is to sound like the person who built the thing.

Reviews are translated when they are classified, so the inbox shows the original and the translation together.

You write, or approve, in your language. The reply goes out in the reviewer's, and you can see that version before it sends.

Nothing is locked. Open a draft and edit it like any other text box. Once you have touched it, nothing rewrites your words.

If a draft is close but wrong, regenerate. If it reads stiff, Humanize reply rewrites it at the level you pick (light, natural, or casual) and revalidates the result.

Edits do not quietly retrain anything. If a phrasing keeps coming out wrong, fix it in the voice profile so the next hundred drafts come out right.

Every draft is validated before it is offered to you or sent, and it is written to fit the length limit of the platform it is going to.

It will not invent a refund, a fix date, a discount, or a feature you do not have. If a reviewer asks for something only a human can promise, the reply acknowledges the problem and points them at support rather than committing you.

Positive reviews get a real thank-you. That is required, not optional. Templated corporate gratitude is filtered out. "Thank you for your valuable feedback" is not a reply.