Prediction markets: discovery, personalization, and what’s next
Prediction markets are here to stay. Over the past few weeks, prediction platforms have seen a new wave of interest and activity.
At their core, prediction markets rest on a simple idea: traders bet on the outcomes of real-world events. Unlike traditional betting, odds aren’t set by bookmakers—they emerge from supply and demand, reflecting the aggregated beliefs of traders.
Prediction markets first hit mainstream attention during the latest US elections, when a French whale reportedly pocketed tens of millions by betting based on information gathered on the ground. Today, they’re entering another growth cycle and attracting a cohort of new users, opening up a new set of challenges and opportunities.
A new wave of builders
This spike in interest is galvanizing both traders and builders. Dozens of experiments are launching each week, and new projects are announced almost daily.
Resources like predictionindex.xyz or polymark.et are becoming essential for tracking the space, categorizing some of the main PM app categories like aggregators, analytics and AI agents.
Some other great accounts to follow include @primo_data, @j0hnwang, and @aulijk.
The discovery problem
“Discovery for markets is broken: way too many markets to sift thru manually. Took me minutes of sifting on just a single site to find something I wanted to bet on.”
— @ryohhno
"discovery in prediction markets is broken"
— @aulijk
Despite their momentum, prediction market platforms remain immature. One of the most pressing issues is market discovery.
Current platforms rely on dense, list-like interfaces.
High cognitive load makes it hard for users to find relevant markets.
Many niche or high-signal markets never get visibility.
Market makers are less efficient, reducing liquidity and diversity.
If prediction markets are to scale, algorithmic discovery is inevitable.
At ❜embed, we’re exploring how a prediction market feed could look and feel—bringing the best markets directly to each user.
A personalized feed
Imagine a personalized feed where relevance is driven by both onchain and offchain signals:
Onchain interactions: bets, swaps, transactions
Offchain interactions (via integrations): views, clicks, dwell time
Our thesis: applying deep learning and large recommender models to prediction market data could deliver at least a 5x uplift in engagement compared to the current trending pages on Polymarket or Kalshi, probably much more over time as new kinds of experiences and ‘stories’ are layered on top of that.
New storytelling surfaces
The design space for new discovery experiences is wide open. At ❜embed, we’ve been exploring three promising directions for feeds and recommendations:
Social charts → interactive visualizations showing how outcome prices move, tied to events
Seismic shifts → surfacing markets with rapid price changes, signaling breaking news or sentiment swings (see tremor.live)
PnL cards → inspired by copy trading apps, showing entry trades and live P&L from people in your social graph
This is just a start and we’re hoping to share more designs in the days to come and inspire builders!
Looking ahead
Prediction markets are still rough around the edges, but the momentum is clear. Better discovery, stronger UX, and AI-driven personalization can make them more approachable, efficient, and mainstream.
We’ll be sharing more soon, including results from our experiments training AI models on prediction market data.
If you’re building a prediction market UI and want to jam on ideas, reach out on X.