AI Automated Comment Moderation Tool for Social Media
Moderation of Social Media Comments
Norwegian media law holds editors responsible for what appears in their comment sections, redaktøransvaret. Defamatory, hateful, or illegal comments that aren’t caught and removed can result in fines, regardless of whether Facebook’s own filters flagged them. Keeping up with that responsibility meant a rotating three-person shift going through hundreds of comments a week across every post published on the company’s Facebook page, deleting or hiding whatever didn’t meet the bar.
The cost of doing it manually
Reading and judging every comment by hand is slow on its own, but the bigger cost was in what that work did to the people doing it. The material is negative by nature, and reading it in volume for hours at a time is mentally taxing in a way that carries over into whatever comes next in the day. Comments also don’t stop arriving, so by the time a batch was reviewed, new ones had already appeared and needed the same treatment. None of this was anyone’s actual job either, moderation was a side responsibility layered on top of work that was already generating value elsewhere, so every hour spent on it was an hour taken from something else. From a business perspective, that adds up to a real, recurring labor cost for work that produces no output beyond preventing a fine.
How the tool works
The tool is a simple web app, accessible from anywhere through Azure authentication via Entra ID. A moderator pastes in the URL of a Facebook post and starts the analysis. The app resolves the post through the Meta Graph API, pulls every comment on it, and sends each one to Azure AI Content Safety for scoring. Each comment comes back individually judged and rated by severity, and the final list is sorted so the moderator sees exactly what needs to be deleted immediately, what needs a closer look, and what can be left alone. Work that used to take hours is compressed into seconds, and the part of the job that was hardest on the people doing it is gone.
Security and privacy
The app is deployed inside the company’s own Azure environment and requires authentication with a company email, with new users added by an admin rather than open registration. Azure AI Content Safety never receives the names of comment authors, partly by design and partly because the Meta Graph API doesn’t expose that information in the first place. Nothing about a session is stored either, output exists only for as long as the tab stays open, and closing or refreshing it clears everything. The interface itself stays intentionally simple, paste in the full post URL and the app handles parsing the post code from it.
Impact
The monthly cost of running the app is a fraction of what the equivalent labor cost, and that comparison alone would justify the switch. But the more meaningful change is what happened to the hours that got freed up. Time that used to go into a taxing side responsibility now goes back into the work that actually creates value for the company.