How the review agent works: from new review to posted reply

A review lands on the App Store or Google Play and, if you let it, a few minutes later there is a draft reply waiting in Uprate, in the reviewer's language, in your tone. What happens in between is not magic and it is not a black box either. This is the whole path, from the moment a review arrives to the moment a reply is posted, and every point where you decide.

The short version: drafts always wait for you unless you have switched on a rule that says otherwise. The rules are yours to set, per star rating, with a delay, and they are off by default.

What arrives in the inbox

Once a store is connected, new reviews from App Store Connect and Google Play arrive in one inbox. Before a store is connected, Uprate reads what is public, so you can already see the reviews, but the Reply button stays off until the store can accept a reply from you.

Each review is classified on arrival. Sentiment is positive, neutral or negative. The language is detected, and reviews you cannot read get a translation in the thread. Flags mark the ones that need a different kind of attention: vulgar, spam, fake, off-topic, a policy violation, or a bug report. For a flagged review Uprate prepares the report text you would paste into the store's own report form, because arguing with a spam review in public helps nobody, and the reply templates article says why.

You then sort with the usual tools: star rating, store, country, sentiment, your own tags, free text search, and a view of edited reviews only. Two chips do most of the work day to day: Needs you and Drafted. Anything you have dealt with can be marked resolved, and reviews imported from before you connected the store land under Resolved so they do not pretend to be new.

What a draft is made from

A draft is not the review pasted into a chat model with "reply nicely". These are the inputs, and you control most of them in the agent settings.

Input	Where it comes from
The review	Text, star rating, author name and the detected sentiment.
The store's limit	Google Play caps replies at roughly 350 characters, so the word budget is clamped per store; you can set a shorter one.
Your voice	A preset (Professional, Friendly, Technical, Empathetic) or a prompt learned from your own past replies.
Your knowledge	Documents you upload as the app's knowledge base, PDF or Markdown. The relevant passages are pulled in for each review.
Templates	Your own examples by category: bug report, feature request, praise, complaint, general. Used as guidance for tone and structure, not copied.
Language rules	Reply in the reviewer's language, or always in a default language with a fallback. Each language carries its own sign-off, support-team label and style rules.
Contact and signature	Support email, support person or team name, a localized brand label, and whether replies are signed.

Two of those deserve a closer look.

The learned voice. If you have already answered reviews, Uprate can read your published replies and write a style prompt from them. It needs at least five review-and-reply pairs, samples them across star ratings, and works in two passes: first it extracts the concrete patterns (phrases you use, how you open and close, emoji or not), then it writes the prompt from those patterns. The result is a draft prompt you can edit, and every change is kept in a version history you can restore from. A rough match of your own voice beats a polished generic one, which is why the threshold is low.

Language rules. The reply language config is more than a translation table. Czech replies use formal pronouns and never an exclamation mark in the greeting. Slovak replies are not allowed to drift into Czech. Every language has its own closing line and its own "support team" wording, so a German reader gets "Mit freundlichen Grüßen" and a Japanese reader does not get a translated English sign-off.

Three modes, and the default is off

The agent has three modes, and you set them per app.

Off. Nothing happens on its own. You open a review, press Draft, and a draft appears. You can also pick reviews in bulk and draft them together.

Auto-draft. Choose star ratings, and every new review with those ratings gets a draft automatically. It still waits for you. Run it for every rating and the inbox becomes a list of drafts to approve rather than reviews to answer.

Auto-reply. Choose star ratings, choose a delay in minutes, hours or days, and matching replies are posted without you. The delay exists so you can still catch one before it goes out. Five-star thank-yous are the usual candidate. One-star replies rarely are. Auto-reply is part of the paid plans; drafts are in every plan.

Whatever the mode, editing is one step Approve as written, edit and then approve, or replace the draft. If you write a reply yourself, Quick Polish fixes grammar and style without changing what you said, and Full Enhancement rewrites it under your agent settings. Nothing is posted until you press the button or a rule you enabled does it for you.

What happens after you post

The reply goes to the store through its API, and the store notifies the reviewer. Uprate keeps the posted reply, its status and who sent it, and it also syncs back replies you posted from the console, so the thread stays complete either way.

Both stores allow one reply per review, and both allow editing it later; Uprate updates the store when you edit. Deleting is possible where the store permits it. The rest of the store rules, the 350 characters on Play, the single notification on your first reply, are in App review response examples and templates.

The backlog

Connecting a store with a year of unanswered reviews is the normal case, not the exception. You do not have to answer them one by one.

The rules above only touch reviews that arrive from today on. For older ones there is historical processing: pick a date range, and Uprate first counts the unanswered reviews that match your rating rules and tells you whether it is about to draft or to publish, then you press Process Now. A sensible pattern is to draft the last 30 days, approve those, and leave older reviews resolved; a reply on a review from last spring notifies someone about an app they may have deleted. The same screen can scan older reviews for flags, so a backlog of spam gets its report texts too.

What the inbox watches for you

Replying is the visible half. The other half is noticing things across reviews that no single review shows.

Anomalies. Uprate flags a negative spike, a rating drop, a drop in the store rating, a topic spike, and a regression tied to a release. Those arrive as alerts, so a bad build shows up as a pattern on the day it ships rather than as a rating three weeks later.

Topics. Reviews are grouped by what users keep asking for, with your own custom topics on top. That is the part that turns twenty support replies into one product decision, and it feeds the rating side of ASO more than any single reply does.

What it costs

Uprate runs on credits. A reply draft or a translation uses one credit, a template or a fix uses two, an analysis uses five. The free plan includes 100 credits a month with one app, which covers a small app's reviews comfortably; paid plans start at 500.

Where this fits

The review agent is usually the first thing to switch on after connecting a store, and the first one you stop thinking about. The setup is in Getting started with Uprate, the wording is in the reply templates, and how it sits alongside submissions, pricing and listings is in How to run both stores without a dedicated ops hire.