How the AI works

PolyForecast compares the probability implied by a live Polymarket price with an independent estimate from an AI model (Google Gemini).

1. Live market data

Every few minutes we sync open markets, prices, volume and liquidity from Polymarket’s public APIs, and collect price history and recent news headlines.

2. The AI forecast

For liquid markets, the model receives the question, its resolution rules, price movement, volume, liquidity, time left and recent headlines, and searches the web for current facts. It returns a probability, a confidence score, factors on each side, key risks and a data-quality rating. We validate every answer on our server and discard anything malformed.

3. The signal

AI edge = AI probability minus market probability. Signals come from fixed numeric rules, not from the AI’s wording: a small edge is Neutral, a moderate edge is Yes or No, and a large edge is Strong only when confidence, liquidity, data quality and timing all qualify.

4. Honest track record

Every forecast is kept. When a market resolves, each forecast is scored (Brier score) so performance can be measured over time, including the misses.

Forecasts are probabilities, not guarantees, and nothing here is financial advice.