Skip to main content

CivicAI for citizens

CivicAI is the AI layer built into CivicOS. As a citizen, you don't interact with a chatbot — CivicAI shows up quietly inside features you already use, to help you file better issues and reduce the "did anyone even read this?" friction that kills civic engagement.

Everything CivicAI does is a suggestion. It never edits your report, overrides your category, or posts on your behalf.

What you'll see today​

Category suggestions when reporting an issue​

When you open Report an issue and start typing a title + description, a small chip appears under the category dropdown a moment later:

✨ CivicAI suggests: Utilities  · apply
                    AI-generated · review

The chip is doing one thing — reading what you just wrote and guessing which of the eight CivicOS categories fits best (Infrastructure, Health, Education, Security, Environment, Utilities, Transport, Other). If you click the chip, the dropdown updates to that category. If you'd rather pick your own, ignore it — CivicAI never overrides a choice you've already made.

Why this exists. Getting the category right means the right organization gets your report faster. When categories are wrong, reports sit in the wrong inbox and go unanswered. This is the quietest way to help you help yourself.

What it uses. Only the title + description you're typing. Nothing else about you is sent. The suggestion appears client-side; if the suggestion service is down or slow, the form still works exactly the same.

Text an organization drafted with CivicAI​

Organizations can use CivicAI to draft announcements, campaign descriptions, updates to donors, and closing reports. When you read one of those, you are reading words a person at that organization reviewed and chose to publish — CivicAI cannot publish anything itself, and the organization is accountable for what it says either way.

What CivicAI is told, every time, is not to invent facts: no beneficiary counts, no amounts, no outcomes that were not in what the organization supplied. Where an organization reports spending, the drafting tools are required to describe it as what that organization says it spent, never as something CivicOS has checked. CivicOS never held the money and cannot check it.

What CivicAI does not do​

  • It doesn't file your issue. You still click Post issue.
  • It doesn't rewrite your words. The description you write is the description that gets posted.
  • It doesn't decide priority. It surfaces a severity hint (LOW / MEDIUM / HIGH / CRITICAL) to help staff triage, but that hint is never shown to other citizens and never changes who sees your issue.
  • It doesn't route your issue to any specific person. Assignment still runs through the normal community + organization flow.

Trust and transparency​

Every AI-generated output in CivicOS carries an AI-generated · review tag. If you ever see a chip, badge, or paragraph without one, that's a bug — please flag it so we can fix it.

CivicAI runs on Google Gemini. The details of how it's built, what data it reads, and what it doesn't read live in the developer guide.

What's coming next​

Features you may see later, in rough order of arrival:

  • Multilingual — write your report in Yoruba, Igbo, Hausa, or Pidgin; staff still get an English summary.
  • Duplicate hint at report time — "3 similar issues already exist on this street — sign these instead?"
  • Plain-language summaries of long policy documents attached to consultations.

If you have ideas for what CivicAI should help with, tell us in the feedback channel.