Social-Media Complaint Triage
Pull public complaints from X, reviews, and Facebook into one queue, score each for severity and reputational risk, and draft a careful public reply plus a take-it-private message - so issues get answered before they blow up.
Public posts are loaded and deduped, scored for severity + reputational risk, given approved-once draft responses (public reply + take-it-private), reviewed and approved by a lead, then posted manually with the action logged.
Before you start
- A free Vercel account
- A free Supabase account
- A free Resend account (plus a verified sending domain or test address)
- A CSV or Google Sheet export of public posts with platform, handle, text, and reach if available (no API required to start)
The problem this kills
A public complaint on X, a one-star review, or an angry Facebook post is not like a quiet email in your inbox - everyone can see it, and it grows. By the time someone on your team scrolls past it, it may already have replies, screenshots, and reach you can't take back. Meanwhile the genuinely dangerous post (a safety claim, a viral thread, a journalist asking questions) sits next to a routine grumble with no way to tell them apart at a glance.
The usual fixes are bad. Replying fast but carelessly in public creates its own incident - the wrong word, posted to thousands, is a screenshot forever. Replying slowly lets it spread. And spreading the watching across five people's phones means duplicate replies, missed posts, and no record of who said what.
What you'll build
A small internal web app where you paste or import public posts, and the tool scores each one for severity and reputational risk (weighing reach, sentiment, and the kind of claim being made), then drafts two careful responses: a short public reply, and a "let's take this private" outreach. It ranks the queue so the high-reach, high-severity posts rise to the top.
Crucially, nothing is public until a person says so. The AI drafts; a comms or support lead reviews the score and the wording, edits if needed, and approves. Only then is the response marked ready to post manually - and the action is logged.
What's inside the Implementation Plan
The plan opens by interviewing you about your business - which platforms you watch, what counts as high-risk for your brand, how you measure reach, who approves public wording, and your real escalation triggers - so the scoring and the draft replies fit your voice and your risk appetite, not a generic template. It reflects a short spec back to you for a thumbs-up before building anything.
From there it walks you, prompt by prompt, through standing up the app, modeling your data, importing posts (CSV or Google Sheet - no API needed), scoring severity and risk, drafting the public and private responses, building the approval gate, and shipping the prioritized triage queue. Every build step ends with a ready-to-paste prompt for your AI coding agent.
The governance it includes (this is the point)
- Login so only your team can open the tool.
- Row-level security so each organization only ever sees its own posts and drafts.
- A full audit trail - who scored what, which draft was edited, who approved, and when.
- A human-in-the-loop approval gate: the AI drafts the severity score and both responses, a lead reviews and approves, and only an approved response is ever marked ready to post or send. Nothing goes public automatically.
- Duplicate guards keyed on the post URL/ID so the same complaint can never be triaged or answered twice.
Who it's for
Social, comms, and support teams handling public-facing complaints across X/Twitter, review sites, and Facebook who want to catch the dangerous ones first and respond with care - built by the ops or comms person, no developer required.
You've got this - paste the first prompt and let the interview tailor it to your brand.