integrity
We build poker bots. Here is where ours stop.
Machines have beaten the best humans at this game in published experiments, and every operator we read bans them from its tables. Pokiq builds practice opponents, which means this page owes you a boundary rather than an opinion: ours sit on a practice table where no money is at stake, they never adapt, and their frequencies are printed on them before you sit down.
sources checked 30 August 2026
what research has shown
Four findings, each from a source linked at the bottom of this page. They are worth knowing precisely, because the vague version of them is what makes people quit.
2015: one variant was solved outright
A team at the University of Alberta reported in Science that heads up limit hold’em is essentially weakly solved. The same computation formally proved something players had always assumed, that the dealer holds a substantial advantage. Limit hold’em heads up is a much smaller game than the one most people play, which is exactly why it went first.
2017: no limit, heads up, against four professionals
Carnegie Mellon reported that its program played 120,000 hands of heads up no limit hold’em against four professionals and finished ahead by 1,766,250 dollars in chips, a margin its builders described as statistically significant rather than luck. Heads up no limit had been the standing benchmark for imperfect information games, and it fell.
2019: six players at the table
The follow up, published in Science, played six handed. Two professionals each played 5,000 hands against five copies of it, and thirteen professionals played 10,000 hands in total. Six handed matters because it is the game people actually sit down in, and because two player results had always come with the caveat that real tables have more than two seats.
The number that changed the problem
That last program computed its overall strategy in eight days using 12,400 core hours, and then used 28 processor cores while it was actually playing. The research question was whether a machine could beat the best humans. The practical question for a poker site is what it costs to run one, and that answer got small.
what it does not mean
The step from those papers to the conclusion that your table is full of machines is a long one, and most people take it in a single evening after a bad session.
A published result is not a product
These were controlled experiments with agreed formats, published methods and named opponents. None of them is a thing a stranger downloads and points at your table, and treating a research milestone as a description of your Tuesday night game is the fastest way to talk yourself out of playing at all.
Playing like a solver is itself a pattern
The thing that makes strong machine play strong is also what makes it unlike human play, and operators look at behaviour across thousands of hands from a vantage point no player has. We are not going to be more specific than that sentence.
Your losses are still mostly variance and mistakes
Before you conclude that a table was botted, the base rates deserve a look: the swings in this game are wide enough to explain a very bad month without any villain in it. That arithmetic is worked out separately, and it is not a comfortable read either.
The people who can check are not you
Reporting an account with hand identifiers and dates puts it in front of people who can compare devices, timing and money across accounts. An argument in the chat box only tells the account what you noticed.
How wide the swings get on their own, and how long it takes before results say anything about your play, is worked out on the page on skill and chance. What to do when you genuinely suspect something is on the page about cheating, and software that advises rather than plays is on the page on real time assistance.
where our own bots stop
The same technology can be a drill or a fraud, and the difference is not in the code. It is in where you point it and in what you publish about it.
They sit on a practice table, and nowhere else
Our opponents exist inside a training product. There is no buy in, no prize pool and no payout anywhere on Pokiq, and none of them has ever been pointed at a table where money is at stake. That is the whole boundary, and it is not a technical achievement, it is a decision.
Their numbers are printed on them before you sit down
Each one is seven published figures: archetype, how often it enters a pot, how often it raises before the flop, how much it bets relative to calling, its 3-bet frequency, how often it continues on the flop and how often it folds to a bet. A bot at a money table hides what it is. Ours cannot, because the numbers are on the page.
They do not adapt and they do not remember
The same fixed numbers play the first hand and the hundredth, there is no read on you and no memory between hands, and the decision is computed without your cards ever being passed to it. Once you find the hole in one of them, it stays open. That is a bad property for a cheat and a good one for a drill.
Beating one is not evidence that you are good
You found the holes in one fixed strategy. That is a real skill and a narrow one, and we say so on the feature page too: human opponents change what they do the moment you start winning, and none of ours ever will.
Every frequency, the five rung ladder, the mirror opponent built from your own measured numbers and the seven things a fixed opponent cannot teach you are listed on the practice bots page. The score built from hands you really played, which no practice table can move, is explained on the Poker IQ page.
what is not on this page
No bot software by name, nothing about how one is built or run, and no list of the behaviours that give one away. The last one is the omission people notice, and it is deliberate: a list of what gets a bot caught is a list of what the next one will be taught to avoid.
There are also no numbers here about how many bot accounts operators close each year. We found figures quoted in summaries, could not read them at a source we could check, and left them out. On this subject in particular, a confident number that turns out to be wrong does more damage than a missing one.
sources
Read on 30 August 2026. Operator policies change, and the research does not.
- Bowling, Burch, Johanson and Tammelin, Heads-up Limit Hold’em Poker is Solved, Science 347 (6218), 145 to 149, January 2015
The paper that reports heads up limit hold’em as essentially weakly solved, and formally proves that the dealer holds a substantial advantage.
- Carnegie Mellon University, Artificial Intelligence Beats Top Poker Pros, January 2017
The Libratus result: 120,000 hands of heads up no limit hold’em against four professionals, ending with the machine ahead by 1,766,250 dollars in chips, which its builders describe as statistically significant rather than luck.
- Carnegie Mellon University, Carnegie Mellon and Facebook AI Beats Professionals in Six-Player Poker, July 2019
The Pluribus result, published in Science: six handed no limit hold’em, two professionals playing 5,000 hands each against five copies and thirteen professionals playing 10,000 hands in total, with the blueprint strategy computed in eight days on 12,400 core hours and only 28 cores used while playing.
- GGPoker, Security and Ecology Policy
Defines a bot as software, programs, websites or apps that enable play without human intervention or reduce the requirement for a person to decide and act, and lists a permanent ban and confiscation of funds among the sanctions.
- PokerStars, Third Party Tools and Services
Prohibits at all times any tool or service that plays without human intervention or reduces the requirement of a human to make decisions, and states that a human must decide the action and the exact relative size of any bet or raise.
next
Sit down against one that tells you what it does.
There is a table in the browser with no account and nothing to install. It is the same idea as the rest of this page in one sitting: an opponent whose behaviour is published in advance, on a table where nothing is at stake.
questions
What is a poker bot?
Software that plays for you. One operator policy we read defines it as software, programs, websites or apps that let someone play without human intervention or that reduce the requirement for a person to make decisions and act. Another puts it more simply: a human must decide the action and the exact relative size of any bet or raise.
Are poker bots allowed?
Not at either of the two operators whose published policies we read, where the listed sanctions include a permanent ban and confiscation of funds. That is a statement about those two documents on the day we read them. Your own site terms are the ones that apply to you, and they change.
Can a bot beat human poker players?
In published research, yes. A program beat four professionals over 120,000 hands of heads up no limit in 2017, and a six handed follow up beat professionals in 2019. A separate 2015 paper reported that heads up limit hold’em, a smaller variant, is essentially weakly solved. Those are controlled experiments, not products, but the direction is not in doubt.
How do I know if I am playing against a bot?
Usually you do not, and this page deliberately publishes no list of tells. That kind of list works just as well as a checklist for whoever builds the next one. If you suspect an account, report it with hand identifiers and dates to the operator, who can see across accounts and devices in a way no player can.
Does Pokiq run bots at real money tables?
No. Our practice opponents exist inside a training product where there is no buy in, no prize pool and no payout, they play heads up on a practice table, and their frequencies are published before you sit down. A bot that hides what it is and a drill that prints its own numbers are opposite things, even though the code underneath is the same kind of code.
Is practising against a bot worth it?
For learning to punish a specific mistake on purpose, yes, because you can pick the mistake in advance. For working out whether you are a winning player, no: a fixed opponent that never adapts is not the game, and no result against one belongs in a measurement of how you actually play.