How Poker Bots Work

Every poker bot, whatever its origin, does the same four things in a loop: perceive the table, model the game state, choose an action, execute it. The differences between bots are in how well each step is done and how much of it is visible to the room.

Perception

The common method is screen scraping: the bot takes screenshots of the table window and finds cards, stacks, pot and buttons by template matching and OCR. It needs a calibration profile per room and per layout, and it breaks when the room changes its graphics. The alternative is reading the client’s memory or the network stream, which gives exact data with no recognition errors, but requires reverse-engineering the client and is a clear signal to any anti-cheat that inspects the process.

Game state

Recognised pixels become a structured state: hole cards, board, positions, stack sizes, action history for the hand. Good bots track confidence and refuse to act when the state looks inconsistent – a stack that jumped, a card that changed. Bad bots act anyway and make expensive mistakes.

Decision

• Rule engines evaluate a hand-written profile. Fast and transparent, but only as good as the author.
• Precomputed strategies look up the spot in tables built from solver output. Strong preflop and in common postflop lines; the tables cannot cover every board and bet size, so the bot abstracts.
• Real-time evaluation estimates equity against assumed ranges and picks the best expected value. Flexible but approximate.

Most serious bots combine lookups with a real-time fallback.

Execution

The action goes back to the client as simulated mouse and keyboard input. The obvious tells – acting in exactly 1.2 seconds every time, betting to the cent, playing 14 hours without a break – are what naive bots get caught on, and what better bots randomise. Randomising timing is not the same as being undetectable; see how sites detect bots.

Where it breaks

App updates, pop-ups over the buttons, unusual table themes, disconnections, and edge cases in the rules (side pots, all-in with odd amounts) are the daily failures. A bot’s practical quality is mostly about how it handles those.