GUIDE / PLAYER PROFILING
Poker player profiling: how the exploit line is built, hand by hand
Every good player already does this by memory for a handful of regulars: the station you never bluff, the nit you always fold to. The problem is that memory covers five players and you face fifty. This guide explains how a real-time assistant turns the hands it has seen into a profile per opponent, how it decides when the evidence is enough to leave the equilibrium, and how far it is allowed to go. Updated October 2026.
1. Why GTO alone leaves money on the table
Game-theory-optimal play is the strategy that cannot be exploited. It is also the strategy that does not exploit. Against a player who folds to 70% of river bets the equilibrium still bluffs at the equilibrium frequency, because it was computed against an opponent who defends correctly. Every big blind that player over-folds is yours only if you deviate, and deviating correctly requires knowing the number, not the feeling.
A 50/50 call-or-fold spot in the solver is not a coin flip at the table. Against the player who over-bluffs it is a call every time; against the one who never bluffs rivers it is a fold every time. The solver's job is to find the spot. The profile's job is to pick the side.
2. A posterior for every tendency
GTO MAX AI tracks each opponent's tendencies (fold to continuation bet, fold to river bet, raise frequency, showdown aggression, bluff frequency at showdown, call-down width and the rest) as a Bayesian estimate rather than a plain percentage. Each tendency starts at a population prior, the frequency a typical player at these stakes shows, and is updated by every observed decision with a Beta-Binomial posterior. The result is a number with an uncertainty attached: "folds to river bets 71% (18 observations)" means something different from "71% (3 observations)", and the model knows it.
That uncertainty is what prevents the classic mistake of over-adjusting after three hands. A tendency only moves the advice once its posterior clears the noise, and the size of the move is damped by the sample: the first twenty hands buy a small lean, two hundred hands buy a conviction.
3. Archetypes with confidence
On top of the individual tendencies the model classifies each player into an archetype (TAG, LAG, nit, calling station, maniac, glass cannon) with a confidence score. The archetype fills in tendencies the model has not observed yet for that player: a player who has shown station behaviour on three streets is probably also a wide river caller, and the prior for that tendency shifts before any river has been seen. The classification is shown with its confidence so you know when it is a guess.
4. From profile to exploit line
The exploit line starts from the equilibrium strategy for the spot and moves it in the direction the profile justifies. The size of the deviation depends on two things: how far the opponent sits from the indifference point the equilibrium assumes (a player who folds 70% when the bet needs 40% is far; one who folds 45% is near), and how sure the model is. Below twenty hands on a player no exploit line is offered at all; the advice is the GTO line and the panel says why.
The output is two lines, always: the GTO line with its frequencies, and the exploit line with the reason in words. "Bet 75%: GTO mixes 60/40 here, but this player has folded to 71% of river bets over 18 hands, well above the 43% that makes the bluff break even." You see the deviation and the evidence, and you decide.
5. Your own reads count
Numbers lag. You can see a player is tilting before the model has the hands to prove it. In the Players tab you can write your own description of any opponent ("calls too wide", "never folds a pair", "just lost a 400bb pot") and the reasoning layer folds it into the exploit line with the statistics. Notes, aliases and the hand history are kept under your login across sessions, so a player you met last month is known the moment they sit down.
6. Why this works where HUDs do not
Conventional HUDs depend on hand-history files, which GGPoker and its skins do not provide. A pixel-reading RTA builds its own log from what it sees on the screen, so the profile works on GGPoker, Natural8 and ClubGG where PokerTracker and Hold'em Manager do not run. Details in RTA for GGPoker, Natural8 and ClubGG.
GTO MAX AI: the RTA that knows the player.
GTO line, exploit line and the reason, live, on up to four tables, in Chrome. GGPoker, Natural8, ClubGG, PokerNow, and PokerStars in beta. Try it for 30 minutes on Telegram.
Get a 30-minute trialCoaching and study tool. Use it where real-time assistance is allowed: study, home games, private clubs, play money. Most online sites forbid real-time assistance. Advice can be wrong; every decision is yours. Terms of use.
7. FAQ
What is player profiling in poker software?
Turning an opponent's observed decisions into estimates of their tendencies (how often they fold to a bet, bluff, call down) and using those estimates to adjust the advice. GTO MAX AI does this with a Bayesian posterior per tendency so the uncertainty is part of the number.
How many hands does the model need before it adjusts?
Twenty hands on a player before any exploit line is offered. Below that the advice is the GTO line. Above it, the size of the deviation grows with the sample.
Can the exploit line be wrong?
Yes. It is a statistical estimate and it says how confident it is. The GTO line is always shown beside it, and the reason is written out so you can override it when you know something the model does not.
Does it use my own notes?
Yes. Write a description of a player in the Players tab and the reasoning layer combines it with the statistics when it writes the exploit line.
Does profiling work on GGPoker?
Yes. The profile is built from what the reader sees on the screen, not from hand-history files, so it works on GGPoker, Natural8 and ClubGG where conventional HUDs do not.