Boost Reviews Automation

The challenge

App Store reviews need answering, and the work is genuinely tedious: log into App Store Connect, read yesterday’s reviews across every app, write an individual reply, post it, repeat. At any volume this consumes a person.

Full automation is the obvious fix and the obvious risk. Review replies are public and attached to the brand, so an unsupervised model posting directly is a reputational accident waiting to happen — the wrong tone on an angry review is worse than no reply at all.

What I built

The workflow runs in three deliberately separated stages. The first generates drafts: a scheduled trigger fetches the list of applications, authenticates to the App Store Connect API using a signed JWT, pulls each app’s reviews, filters to the last day, and passes them to Claude with a structured output parser so replies come back in a predictable shape rather than free text. Drafts are compiled into a spreadsheet, uploaded to Google Drive and announced in Slack.

The second stage is a human. A person reads the drafts, edits or replaces anything that misses, and signals approval simply by moving the file into a different Drive folder. Using folder location as the approval signal means no extra tool and no bespoke UI — the reviewer works in a place they already use.

The third stage posts. A separate scheduled trigger watches the approved folder, downloads the sheet, extracts the replies, re-authenticates, and posts each response through the App Store Connect API. Every attempt writes a success or error log per review, and the processed file moves to an archive folder so nothing is posted twice.

Outcome

Review replies are drafted automatically and posted automatically, with a human retaining the only decision that actually matters — approval — and an audit log for every reply.