For about twelve years, review replies were the part of indie app work I kept postponing. I'm Masaki Hirokawa (@dolice), an artist and indie developer based in Japan. I've been shipping apps since 2014, and while I always made time for AdMob dashboards and release-day excitement, the morning that started with a one-star review was rarely the morning I had energy to write a careful response.
In February 2026, Claude in Chrome's beta arrived on my Mac, and I decided to put it in charge of part of the review-reply workflow. What follows is a three-month log of how that experiment went. The goal was never full automation. It was closer to building a scaffold that lets a human keep replying without burning out.
Why review replies kept slipping
My catalog includes wallpaper and calming apps with more than 50 million cumulative downloads. On a normal weekday, those titles attract dozens of reviews. After a new build ships, the count can climb into the triple digits in a single day.
The reviews are a mix of feature requests, misunderstandings about subscriptions, and bug reports. They arrive in many languages — English, German, Spanish, Turkish, Mandarin — and the timing rarely aligns with my working hours. When I opened Slack at the end of an evening, I was often staring at a wall of low ratings that I knew I wouldn't reach until the weekend.
Replying to each one alone, day after day, is harder than most people expect even after twelve years. There is a real psychological cost in opening the pile, especially when one-star reviews are mixed in. I knew the data — replies tend to pull ratings up — but knowing did not move my hands every morning.
What I actually let Claude in Chrome do
The first decision was strict: Claude in Chrome would never be the one to hit the "Submit" button. App Store Connect responses are public, permanent, and visible to other reviewers. A typo or a tone mistake there cannot be quietly undone. I limited the AI to three jobs.
The first job is sorting. From the daily incoming queue, I asked it to lift up reviews that mention specific, reproducible bugs — the kind that could pull the average down if Apple's algorithm flags them. The order I see in the morning is now the order I respond in.
The second job is drafting. I let it read each review in its original language and write a draft in that same language. I can polish Japanese and English drafts by myself, but German, Spanish, and Turkish are languages I can read with effort but not comfortably write. Having a usable first draft for those languages cut my felt workload by roughly thirty percent.
The third job is consistency checking. Before I post a reply, I ask Claude in Chrome to look back at how I answered similar complaints two months ago, and to flag tone drift. For a solo developer, internal "house style" tends to live as silent intuition. Having a quiet second pair of eyes to surface it has been more useful than I expected.
To make the daily cadence concrete: in the first five minutes after waking up, I let Claude in Chrome open App Store Connect and Google Play Console in order and dump a sorted list with draft replies into a spreadsheet. Over my morning coffee, I read the drafts top to bottom, adjust each one, and post. The biggest shift in three months was moving from a weekend-batch rhythm to a small daily one.
What changed in three months — both numbers and feelings
My reply rate climbed from a monthly average of 18% before the beta to 64% after three months. For one- and two-star reviews specifically, the rate moved from 38% to 89%.
The rating average increased on five of my six apps, by 0.1 to 0.3 points. Three months is a short sample, so I won't overclaim, but the pattern Apple's documentation describes — that a reviewer who receives a reply is more likely to update their stars — showed up clearly in the numbers I see day to day.
The bigger change is harder to measure: I no longer dread opening the queue. It isn't that low ratings stopped hurting. It's that the pile is no longer shapeless when I look at it.
Both of my grandfathers were Miyadaiku — traditional Japanese shrine carpenters — and I'm told they always began the morning by laying out their tools in order. Letting Claude in Chrome sort and pre-draft feels close to that for me. It's not a magic wand that finishes the job; it's a way to lay out the day's tools so the actual work becomes possible.
What didn't work — tone and culture still need a human
It's only fair to write down the rough edges too.
Left alone, English drafts always wanted to open with "Sorry for the inconvenience." That works for American reviewers, but for some UK and Australian users the phrasing tipped into sounding hollow, and twice I got follow-up one-star comments calling the reply sarcastic. German drafts had the opposite problem — the default tone reached for polite, multi-clause sentences in a culture that, for app reviews, prefers short and direct. Long apologies in that context felt like noise.
I narrowed the gap by feeding ten past replies per language into the prompt and pinning expected sentence length and warmth. Emoji handling matters too: in Japanese, 🙏 or ✨ make a reply land softly, but in German they can read as too light, and in French it's roughly split. Writing down the per-language emoji policy in the prompt reduced the second-guessing on my end.
This is the part I don't expect to fully automate any time soon. Final calibration across cultures still lives best in a human's hands.
What I'd like to try next — pairing it with Crashlytics
Next, I want Claude in Chrome to read Firebase Crashlytics reports side-by-side with the review text. When someone writes "v2.1.0 keeps crashing" or "my saves never stick," surfacing the relevant stack trace next to the draft reply would let me line up the bug fix and the user response in one pass.
The same boundary will hold here. Misreading a Crashlytics trace and confidently telling a user the wrong cause would do more damage than not replying at all. The role I want the AI to keep is preparation, not commitment. Twelve years in, the moment of writing "thank you, your message reached me" is still one of the most important fifteen minutes of my day. I'd rather have help making space for that than have it taken off my plate. I hope this honest report is useful to other indie developers who keep meaning to clear the queue.