How We Built the 'For You' Feed

Most recommendation systems optimize for engagement — time on platform, clicks, shares. That works for social media, where the goal is to keep you scrolling. But it's terrible for polls.

If a poll recommendation system optimizes purely for engagement, you'll see the same inflammatory topics over and over. Outrage drives clicks. But outrage polls don't produce useful data.

We wanted something better.

The signals we use

Our "For You" feed scores each unvoted poll against seven weighted signals:

Recency — newer polls score higher. A poll asked six months ago is usually less relevant than one asked today.

Engagement — a poll with many responses signals it resonates with people. But raw vote count can be gamed, so we weight this carefully.

Interest match — we track which categories you've engaged with and boost polls in those areas. If you've voted on technology polls frequently, you'll see more of them.

Novelty — you shouldn't see the same topic phrased five different ways. We actively suppress near-duplicates.

Diversity — left unchecked, interest matching creates a filter bubble. We intentionally surface polls outside your usual categories on a regular basis.

Social — polls that people in your network have interacted with get a modest boost.

Exploration — a small random factor keeps the feed from calcifying. New topics get a chance to find their audience.

The hard part: calibrating the weights

The algorithm is straightforward to describe. The hard part is calibrating how much each signal contributes to the final score.

Too much recency weight → older quality polls never surface
Too much engagement weight → viral content dominates everything
Too little diversity → users only see what they already like

We've been tuning this by looking at response rates and completion rates across cohorts. A well-calibrated feed should produce high completion rates (people who start a poll finish it) and diversity in what people see.

We're still learning. The weights we ship with beta will almost certainly change.

What we're not optimizing for

We're explicitly not optimizing for time on platform. We want you to see the polls that matter to you, answer them, and get on with your day.

A feed that respects your attention is a feed you'll keep coming back to.