Skip to content
iADam

Social MPD

Own product · Political and institutional communication · 2026

Every comment answered.
In your voice, under your control.

Social MPD brings Facebook and Instagram comments into one inbox. The AI classifies them, drafts a reply in your team's voice and flags negative spikes. A person reviews and publishes. We built it from start to finish for Marketing Político Digital, on Meta's official API.

One platform. The whole business.
Demo · fictional data Demo with a fictional organisation and data: the inbox with a reserved comment and the AI draft.
Web
Real screenshot Real screenshot of the Social MPD public website.
Demo · fictional data Demo with a fictional page and data: the reply, reviewed and published by a team member.
Type
Own product
For whom
Political and institutional communication teams
Channels
Facebook and Instagram
Status
Online, in early use

The starting point

Comments everywhere. Nobody knows who replies.

Comments arrive across several pages and two networks. Without a shared inbox, everyone replies by hand on each post. And in political communication, every reply is public.

  1. Scattered volume, duplicate replies

    Without a shared inbox, messages get lost. And when several people work on the same page, one citizen gets two replies and another gets none.

  2. Replies that sound like a robot

    Copied replies or replies full of stock phrases hurt the credibility of a public official.

  3. Crises spotted late, requests lost

    A wave of criticism is spotted late if nobody watches the numbers. And a complaint about a pothole never reaches the department that can fix it.

Read the full story

Social MPD is a platform for the communication teams of political organisations and public institutions. They use it to reply to comments on their Facebook and Instagram pages. All comments land in one inbox. The AI classifies them by tone, topic and risk and drafts a reply in the organisation's voice.

A team member reviews the draft, edits it if needed and decides whether to publish. The platform also flags spikes in negative comments, turns citizens' requests into tickets for the right department and produces reports for management.

It is an own product of the Marketing Político Digital brand. We did all of it: the product, the design, the database and permissions, the Facebook and Instagram connection through Meta's official API, the entire AI layer, the public website with its legal pages and the installable mobile app. Meta approved the app in June 2026.

What the product solves, based on how it is designed (there is no measurement of the situation before): comments arrive across several pages and two networks. Without a shared inbox, the team replies by hand on every post and messages get lost.

When several people work on the same page, two of them can answer the same citizen while another gets no reply at all.

Copied replies or replies full of stock phrases hurt the credibility of a public official.

A wave of criticism or a coordinated attack is spotted late if nobody watches the numbers. And a concrete complaint, like a pothole or a failing service, never reaches the department that can fix it.

The cost of leaving this unsolved has not been measured. What you do see: citizens without an answer, crises spotted late and public replies with no control over who posts them.

Process before technology. The workflow follows how a communication team really works: an inbox, pages split per person, roles and escalation to departments. AI was added where it saves time.

The AI suggests, a person publishes. There is only one path to post on Facebook or Instagram, and it is triggered by a person's click (or keyboard shortcut). There is no automatic publishing and no scheduled task that posts anything.

What is automated: classifying every comment (sentiment, topic, intent, priority and risk). Drafting a reply or improving a text the person wrote. Detecting negative spikes, coordinated attacks and urgent questions left unanswered. Flagging possible spam (it only flags, it never deletes). And learning the team's style.

What stays with a person: publishing, skipping, archiving, deleting, escalating, approving knowledge base content and applying the prompt changes the AI suggests. These are public messages on behalf of a political office, and a mistake costs reputation.

Nobody steps on anyone's toes. When you open a comment, it is reserved for you and the others see it as taken. If someone has already replied, the system blocks a second reply.

Not like a robot. Every reply goes through two AI rounds. The second rewrites the text, removes typical AI stock phrases and is not allowed to add new promises or figures. The AI also sees the latest replies on the same post so it does not repeat itself.

It learns the team's voice. After a set number of published replies, the AI sums up how the team really writes, including the edits they made to its drafts, and uses that for the next ones.

Only approved knowledge. Documents, PDFs and saved replies only reach the AI once an admin approves them.

Each organisation picks its AI: OpenAI, Anthropic (Claude) or Google (Gemini), with its own key if it wants and a different model per module. There are six roles with their own permissions, pages assigned per person and an audit log of replies, status changes, deletions and settings.

The path of a comment

From comment to published reply.

The AI prepares. Your team decides.

01 · Comment

All comments in one place.

Comments on your Facebook and Instagram pages come into the platform. The dashboard shows the topics, the sentiment and what still needs a reply.

Demo · fictional data Demo with a fictional organisation and data: the dashboard with top topics and sentiment.
02 · Analysis

The AI reads and classifies it.

Sentiment, topic, intent, priority and risk. If negative comments on a topic shoot up within 2, 6 or 24 hours, an alert goes off.

Demo · fictional data Demo with a fictional organisation and data: an alert for a spike in negative comments.
03 · Draft

A draft in your team's voice.

Two AI rounds: the first writes with approved knowledge, the second removes stock phrases. The platform measures how much the team changes each draft.

Demo · fictional data Demo with a fictional organisation and data: AI quality and a proposal to improve the prompt.
04 · Review

A person reviews and publishes.

Whoever opens the comment reserves it. They edit where needed and publish with one click. Each role sees what is theirs and each organisation picks its AI provider.

Demo · fictional data Demo with a fictional organisation and data: the AI provider per module and the team roles.
05 · Escalation

The request reaches the department.

A pothole or a water outage becomes a ticket with a department, an owner, a deadline and a history.

Demo · fictional data Demo with a fictional organisation and data: citizen requests by department and status.
06 · Report

Management sees what is going on.

An executive report with a summary, risks and AI recommendations, plus a suggested 14-day calendar.

Demo · fictional data Demo with a fictional organisation and data: the executive report.

Four building blocks. A team that replies better.

01 · Inbox and replies

One inbox for everything that comes in.

Facebook and Instagram in one shared queue. The AI writes with approved knowledge and a person publishes.

  • Each comment reserved for whoever opens it
  • Two AI rounds and the team's learned voice
  • Meta's official API and an approved knowledge base
See how it works
  • Comment inbox Live

    One shared queue with the Facebook and Instagram comments of the connected pages. Sort by relevance or date, with flags for what is urgent, risky or waiting too long. Includes a batch mode, templates, keyboard shortcuts, a 24-hour sentiment pulse and spam warnings. Right now the team starts the sync with Meta by hand, and the reply deadline is the same for every organisation.

    The team syncs and new comments arrive from Meta → You open a comment and it is reserved for you → You write or ask for a draft, review it and publish → The inbox jumps to the next one and prepares the draft (publishing is still a click)

  • AI replies Live

    A draft in the chosen tone (institutional, friendly, political or empathetic) and in the page's learned voice. The AI gets the page settings, approved knowledge, recent replies and real examples from the team. A second pass makes the text sound human. Up to 30 drafts per minute per person. Still to do: making the regional style adjustable per organisation.

    Comment → First draft with context and approved knowledge → Second pass: stock phrases out → A person edits and publishes

  • Facebook and Instagram connection Live

    Connects the organisation's pages and their linked Instagram business accounts, imports posts and comments and publishes approved replies under the comment. It uses Meta's official API, and Meta approved the app. There is a manual route for Business Manager pages that do not show up in the standard flow.

    Sign in with Facebook Login for Business → Pages are saved and their Instagram accounts detected → Posts are imported by link or from the feed → Replies are posted under the comment

  • Knowledge base Live

    The official information the AI may use: manual entries, PDFs split into chunks and published replies the team chooses to keep. Search works by keyword.

    A team member adds information → It waits for approval → An admin approves it → Only then does the AI use it

Illustration Illustration of what this part does.

02 · Alerts and escalation

See the crisis before it grows.

Every comment is analysed. Spikes and coordinated attacks alert management, and requests reach the right department.

  • Sentiment, topic, intent, priority and risk
  • 2, 6 and 24-hour windows with your own thresholds
  • Numbered tickets with an owner and deadline
See how it works
  • Comment analysis Live

    Every comment gets a 5-level sentiment, 1 to 3 topics (reusing existing topics so they do not scatter), an intent, a priority and a risk level with the reasons. The analysis sorts and alerts, but it never blocks or publishes anything.

    A post is imported → The AI analyses its first 20 comments → The rest are analysed in a batch → Every analysis feeds the alerts

  • Crisis alerts Live

    Detects crisis keywords set by the organisation, spikes in negative comments per topic in 2, 6 and 24-hour windows, several authors attacking the same topic in a short time and important questions still unanswered after 4 hours. Each organisation sets its own thresholds. The response is always human. At most, a critical alert can create an internal ticket (off by default).

    Analysed comments come in → Compared with the previous window → Alert to management inside the platform → The team decides what to do

  • Citizen requests Live

    Turns a comment with a concrete request into a numbered ticket for a department, such as public works or mobility, with an owner, deadline and urgency. Tracked by status, with notes and history.

    A person escalates the comment → The system suggests the department from keywords → An owner and deadline are assigned → Tracked until resolved and closed

Illustration Illustration of what this part does.

03 · Audiences and reports

Understand what matters to citizens.

Profiles per citizen, a dashboard that compares periods and reports with recommendations and a 14-day calendar.

  • Promoters, critics and neutrals by fixed rules
  • Radar for topics that suddenly grow within 24 hours
  • Printable report, for admins only
See how it works
  • Audiences Live

    Groups comments by person and shows a profile per citizen: topics, tone and frequency. The promoter, critic or neutral label comes from fixed rules, not from AI. A location filter is in the design but not filled in yet.

    Comments from the same person → Profile with topics, tone and frequency → Promoter, critic or neutral by fixed rules

  • Reports and content strategy Live

    The executive report combines metrics with a summary, a per-topic analysis, risks and recommendations written by the AI, ready to print. The strategy module looks at which posts, formats and times work best and proposes a 14-day posting calendar. Only admins can generate them, and nothing gets published.

    An admin requests the report → The AI writes the summary, risks and recommendations → Suggested 14-day calendar → Printable version

  • Dashboard and analytics Live

    Alerts, peak hours, team performance, top topics, public commitments and daily activity. Filter by date and page, compare with the previous period and use a radar for topics that suddenly grow in the last 24 hours. Still to do: showing the daily history that is already being saved.

Illustration Illustration of what this part does.

04 · Quality, team and app

Control over the AI and over who does what.

You see how much the team corrects the AI, and the prompt gets better. Six roles, an audit log and a mobile app.

  • Prompt improvements that an admin decides on
  • Six roles, pages per person and an audit log
  • Website with legal pages and an installable app
See how it works
  • AI quality and prompt improvement Live

    Measures how many AI replies go out unchanged, how much they are edited and by which model, person or tone. An optimiser reads the team's edits and proposes a better prompt, with a version history and a restore button.

    The team edits drafts → The optimiser spots patterns → It proposes a better prompt → An admin decides whether to apply it

  • Team, permissions and audit log Live

    Six roles (super admin, organisation admin, director, community manager, reviewer and analyst), invitations, pages assigned per person, several organisations on the same platform, an audit log and in-app notifications (no email or push).

  • Public website and mobile app Live

    A Spanish website that explains how Meta data is used, pages for privacy, terms and data deletion (a Meta requirement) and an app you can install on your phone. Still to do: mentioning Instagram on the website and updating the AI providers in the privacy policy.

Illustration Illustration of what this part does.

Product status

What already works. What comes next.

Online and in early use. The scope of the product, with its status in view.

Real screenshot Real screenshot of the Social MPD sign-in screen.
One platform. The whole business.The AI prepares. Your team decides.
  • Live

    Inbox and AI replies

    Facebook and Instagram in one queue, with drafts in two rounds. A person always does the publishing.

  • Live

    Connection approved by Meta

    Meta's official API for Facebook and Instagram. Meta approved the app in June 2026.

View all deliverables
  • Live

    Analysis, alerts and citizen requests

    Every comment classified, alerts for negative spikes and tickets per department.

  • Live

    Reports, analytics and AI quality

    Reports for management, a dashboard that compares periods and a prompt that improves with every edit.

  • Live

    Team, permissions and mobile app

    Six roles, an audit log, several organisations and an installable app.

  • In development

    Real-time sync with Meta

    Today the team syncs with one click. Next step: comments come in on their own.

  • In development

    Style and deadlines per organisation

    Regional style and reply deadline configurable per organisation, and the daily history in analytics.

How it works in numbers

2 rounds
of AI per replythe second removes stock phrases
2 h · 6 h · 24 h
windows to detect negative spikesconfigurable thresholds
6
roles with different permissionseveryone sees what is theirs
25
automated testspermissions, login and AI safety
View all figures
0
automatic postsa person always publishes

Source: this describes how the product is built, based on its repository (October 2026). Usage data is not published.

Marketing · Software · Artificial intelligence

Is your team missing important comments too?

Let's talk
WhatsApp