Social Mention & Listening Triage

The problem this kills

A complaint posted at 4pm Friday shouldn't be discovered at 9am Monday. But when you're a social or community manager watching mentions across Instagram, X, Facebook, Reddit, review sites and Google Alerts by hand, things slip. The angry customer, the influencer shoutout, the journalist asking a question, the early whiff of a PR problem - they all land in the same noisy stream, and you triage them in your head with no record of who decided what.

So mentions get missed. Two people reply to the same post. A brewing complaint sits unseen until it's a thread with 200 quote-tweets. And when your boss asks "how fast did we respond to negative mentions last month?" you have nothing but a guess.

This tool turns that chaotic stream into a calm, accountable queue: every mention captured once, sorted by how urgent and how positive or negative it is, routed to the person who handles that kind of thing, and acted on only after a human says "yes, go."

What you'll build

A private, login-only web app for your team that:

• Imports mentions from a CSV export or forwarded alerts - author, text, link, date, channel.
• Classifies each mention with AI - sentiment (positive / neutral / negative), urgency (low / medium / high), and a flag for potential PR risk that needs fast escalation.
• Routes each mention to the right owner or queue based on your rules (complaints to support, praise to community, PR risk to comms/leadership).
• Puts a human in charge - your community manager reviews each AI suggestion and approves the routing and the drafted response or escalation before anything happens.
• Logs every action - who classified, who approved, who escalated, and when.
• Exports a clean CSV of the triaged mentions and responses in the exact columns your reporting needs.

It will never triage the same mention twice, thanks to a duplicate guard keyed on source plus the post's own ID.

What's inside the Implementation Plan

The plan is a step-by-step runbook you paste into Claude Code (an AI coding agent). It builds the whole tool with you, in plain language, over an afternoon or two.

It starts by interviewing you about your business. Before a single line of code, the plan has the AI ask you about your channels, the exact shape of your mention exports, your routing rules, your typical and peak volumes, and your messiest edge cases - then it reads back a short tailored spec and waits for your thumbs-up. The result is a tool shaped around how you actually work, not a generic template you have to bend to fit.

From there it builds in order: login and security, the data model and import, AI classification, the routing rules, the human approval gate, the audit trail, email notifications, and the CSV export - each step ending in a ready-to-paste prompt.

The governance it includes (this is the point)

This isn't a toy. The plan bakes in the controls that make an internal tool safe to trust:

• Login so only your team can open it.
• Row-level security so each organization only ever sees its own mentions.
• A complete audit trail - who classified, approved, routed, or escalated, and exactly when.
• A hard human-in-the-loop approval gate - the AI only ever drafts a classification and response; a person reviews and approves before anything is sent or escalated.
• Duplicate guards so the same mention can't be triaged twice.

Who it's for

Social and community managers, small marketing and comms teams, and agency account leads who monitor brand mentions across channels by hand and want a fast, accountable way to catch the ones that matter - especially the complaints and PR risks - before they grow.

You've got this. Paste the first prompt and let the interview tailor the tool to your world.