What Is A Polis

So you want to set up a Polis. Here is what you need to know before you bother.

A Polis is a civic technology platform built for running large-scale, real-time collective preference polls. It was originally developed by MIT Media Lab as a way to let communities express opinions on local government issues through a visual sentiment map rather than binary yes/no voting. The core product is the Polis dashboard, which takes thousands of short text statements, runs clustering algorithms on them, and surfaces the common ground between different demographic segments. I spent about two years running Polis surveys for a mid-sized municipal planning department. We did roughly thirty or so public engagement cycles. By the end I had a lot of opinions and a lot of scars. What follows is the practical breakdown of how Polis actually works when you are sitting in front of it, not the brochure version.

What Is A Polis and why does it matter for your project

The word polis comes from the Greek concept of a city-state, but in the modern civic tech world, it refers specifically to the platform and methodology developed by the Polis Project. It is different from YouGov, SurveyMonkey, or even traditional town halls because it measures opinion structure rather than opinion quantity. Instead of asking people whether they support or oppose something, it asks participants to agree, disagree, or pass on hundreds of user-generated statements, then maps where agreement clusters emerge across different participant demographics. The output is usually a colored dot map where each cluster represents a group of related ideas, and the colors indicate how different segments — age, location, ideology, etc. — feel about those clusters. Green means agreement within a segment, red means disagreement, and grey means low engagement. A well-run Polis session can reveal that two groups which appear opposed are actually aligned on five specific points while disagreeing on three others. Traditional polling misses this. That is the main value. Here is a specific problem I ran into that nobody warns you about. When you feed Polis statements, the clustering algorithm groups by semantic similarity, not by intent. I had one survey about downtown parking where the algorithm clustered "reduce parking minimums" and "build more parking garages" into the same green dot because both were mentioned by the same demographic segment who were simply engaging with the topic at all. They were not actually agreeing with each other. The workaround was to manually relabel the clusters after the survey closed and cross-reference them against the raw statement data. It adds about forty-five minutes to the workflow, but it prevents you from presenting false consensus to city council.

How Polis actually functions in practice

The process starts with question design. This is the part that matters most and also the part most people rush. Your central question has to be specific enough that participants generate useful statements but open enough that they do not constrain the answer space too tightly. "How should we redesign downtown Main Street" will generate better input than "Should we add bike lanes on Main Street." The latter produces statements that just echo the framing and the algorithm clusters become obvious and useless. Once you seed the question, participants see a stream of statements and rate each one. They click agree, disagree, or pass. Participation typically takes twenty to forty minutes depending on how many statements you set as the cap. A standard session might show participants between three hundred and six hundred statements drawn from the full pool. You can control the flow — limit how many agree versus disagree, throttle the total count — but the default settings work fine for most runs. The clustering happens in real time. As statements accumulate, Polis runs a neural net clustering model that groups semantically similar statements together. The dashboard updates live. You can watch agreement patterns shift as more people participate. This is genuinely useful during moderated sessions where you want to show a room that their input is being processed, but it also means you should not shut the survey off the second you hit statistical significance. The tail end of participation often refines cluster boundaries in ways that matter for interpretation.

Demographic segmentation is where Polis separates itself from regular polling. You can collect age, zip code, gender, and custom fields, then overlay those segments onto the cluster map. The standard view shows overall sentiment, but the power comes from seeing that cluster seventy-two is green among participants under thirty-five and red among participants over sixty. That is actionable intelligence for a planning department. That is the kind of thing that changes how a commission votes on a zoning amendment.

Practical setup and common failure points

You sign up at polis.com and start a new survey. There is a question builder, a statement seeding system, and an audience management panel. You can invite participants through direct links, embed the survey on a website, or push it through partner networks. The integration options are limited — you get a shareable URL and an embed code, that is pretty much it. If you need to sync with a CRM or a constituent database, you are doing it manually or building a connector. Here is the pitfall that costs people the most time: statement seeding. If you do not provide enough high-quality seed statements upfront, the algorithm has nothing stable to cluster around and the early output looks like noise. I always seed with at least eighty to one hundred statements that cover the major dimensions of the question before opening the survey to the public. This usually cuts the time to stable clustering from six hours down to about ninety minutes. Some people skip this step because they think organic statements will do the work. They do not. Organic statements are useful for refinement, not foundation. Another issue is sample composition. Polis does not do probability-based sampling. Anyone who clicks your link can participate, which means your results reflect self-selected engagement, not a representative population. This is not a flaw in the tool, it is a fundamental characteristic. If you treat Polis output as a scientific poll you will make bad decisions. If you treat it as a qualitative depth tool layered over quantitative data, it works well. I always run a parallel demographic survey alongside Polis to calibrate how much the self-selected sample deviates from the actual community profile.

There is also a cost consideration. Polis offers a free tier that limits the number of participants and statements, but for anything beyond a small pilot you are paying. The paid plans scale with participant volume and feature access. In my experience, a properly powered municipal survey costs somewhere between two thousand and eight thousand dollars depending on complexity and duration. Budget for that. Do not assume the free tier will get you to a publishable result.

What Polis cannot do

It cannot replace traditional representative sampling. It cannot establish causation. It cannot produce statistically valid prevalence estimates for a population. If your funder or your boss wants a margin of error and a confidence interval, you need a different tool or you need to supplement Polis data with a proper survey. The platform gives you directional clarity and conflict mapping, not demographic generalization. The clustering algorithm also struggles with sarcasm, irony, and highly contextual statements. I once ran a survey where a cluster labeled "support the new library design" was mostly people agreeing with statements that were actually mocking the design. The algorithm could not detect tone. You have to read the raw statements inside every cluster before you trust the color on the map. This takes time and it takes attention. There is no shortcut around manual verification. If you need probability-based results or representativeness, alternatives like Optimal Survey, Polco, or even a well-designed Qualtrics deployment with quota sampling will serve you better. Polis fills a different niche. It is for when you need to understand the shape of public opinion on a contentious topic, not when you need to count heads. Knowing which problem you actually have is the first step toward picking the right tool.

The official platform is hosted at polis.com. From there you can create an account, start a survey, and access the documentation. There is no downloadable software — it is entirely web-based. If someone is selling you a "Polis download" outside that domain, it is not legitimate.