Skip to content
Citizen Infrastructure

AI for participation

When people talk about AI for democracy, the most concrete place it’s already landing is participation tools — the platforms that let residents deliberate, propose, and decide together (think participatory budgeting, or a city consultation). A 2025 study of around 30 participation platforms already using AI, by People Powered with the Civic Tech Field Guide, found a couple of dozen distinct features. A few patterns stand out.

What AI is doing well

Section titled “What AI is doing well”
  • Translation. Running a participatory process in the dozens of languages spoken in a place like New York is hard and expensive. AI translation — improving even for “low-resource” languages — can get many more people to a usable level of understanding. It’s not just cost-saving; it’s reach.
  • Summarising and classifying. If a city process generates 40,000 pages of resident feedback, no one can read it all. Natural-language processing can summarise debates, sort comments into themes, and synthesise long conversations. (This predates the generative-AI boom, which is why it’s the most common feature so far.)
  • Sensemaking. Helping people understand what came out of a process — and, crucially, where their own contribution landed. Done well, this addresses a deep weakness of representative systems: that you give up some immediate say for a collective outcome, and rarely get to see what it bought.
  • Support at the individual level. Assistants that help a participant flesh out an idea so others can understand it.

Where to be careful

Section titled “Where to be careful”

AI here should augment a process, not run it. Sentiment estimation still isn’t fully accurate, summaries can miss whole sections, and sensitive engagements need human review. The aim is to let institutions open up more often by lowering the cost — not to take people out of the loop. The sharpest caution is its own topic: see synthetic participation.

Tools applying these ideas are starting to appear in the toolkit.

Sources

Section titled “Sources”