Bot economics on ClubGG

A ClubGG bot earns inside a single club, and its return is set by that club's rakeback structure and the regulars who sit there. Owners and agents who run their own bots are a real risk in private clubs, but they are not the only people with working software: a retail player can own a real bot too, on a one-time license. The club a bot plays in shapes its return as much as the engine does.

Updated 28 May 2026, 13 min read

Summary

  • A ClubGG bot has no global market value because there is no global pool. Its return is the rakeback economics of the club it plays in, so the same software earns very differently from one club to the next, and choosing the club is part of the job.
  • Insider bots are a real risk. An agent or owner who controls a club can seat his own bot there, shield it from club-side scrutiny and bank the result through his own settlement. That is a governance problem for every player in his club, not proof that only insiders can run a working bot.
  • Server-side exploits, Diamond minting, RNG prediction and hole-card peeks are all architecturally closed because ClubGG sits on NSUS infrastructure shared with the regulated GGPoker operator. The only category with real engineering is decision-support AI playing visible state.
  • A retail player can own a real decision-support bot. The normal route in this niche is a free test in a Telegram group with the team, then a one-time license with no monthly fee, with results judged over tens of thousands of hands and table discipline keeping the account healthy. Fake listings for Diamond hacks, card readers, RNG predictors and malware APKs are still out there.

A bot is only worth its club's deal

On a public site a bot's value is roughly the same everywhere: it joins the global pool, beats the average opponent and grinds rake-adjusted EV. On ClubGG there is no global pool. A bot only ever plays inside the clubs it is admitted to, against those clubs' regulars, paying those clubs' rake, and cashing out through those clubs' settlement. Its return depends on the club's deal as much as on the software.

The same engine that holds a steady edge in a soft club with generous rakeback and loose regulars has a much harder time clearing the rake one tier up, with tougher players, worse rakeback and an owner who watches. Put it in a third club where an agent wants a share of the winnings and the math changes again. The deal changes without any change to the bot. The "buy a bot, point it anywhere" pitch is wrong. A real bot is bought once and then matched to clubs where the numbers work.

It also explains why insiders run bots. An agent or owner who controls a club's deal can seat a bot quietly, suppress club-side scrutiny, set a rake that favours it and settle the winnings without questions. That is a real risk for the players in his club, and it is why the governance questions on this site matter. A retail player has none of those levers and does not need them. He needs a solver-grade engine, a club with soft regulars and fair rakeback, a one-time license so no monthly fee eats into the edge, and the patience to judge results over months of steady volume. The fraudulent "hack" market is a separate thing that preys on players who skip all of that.

Categories in the retail "hack" market

The public side of all this is the flood of "ClubGG hack" listings aimed at players. The listings make five claims that differ in feasibility. Four are fantasy. The real one is available to watch in a free test with a serious team before you pay.

Claimed "ClubGG hack" categories: what each would require to be real
CategoryWhat it claimsRequired capabilityFeasibility
Server exploitRead cards from NSUS infrastructureRemote code execution on shared serversPractically no. The value goes to a bug bounty, not a Telegram listing
Diamond infusionMint or duplicate DiamondsWrite access to the balance ledgerNo. Balances are server-authoritative and reconciled to agent logs
RNG breakPredict the next board cardRecover CSPRNG state from outputsNo. Modern CSPRNGs are not invertible at the rate poker exposes
Hole-card peekSee opponent cards liveOperator privilege or client decryptionNo. Same server-authoritative model as GGPoker
AI decision engineBetter play given visible stateSolver outputs + opponent model + UI automationYes. The only category with real engineering

Four of the five are architecturally closed or economically nonsensical for a product sold publicly. The fifth, decision-support AI, is what an honest team engineers and sells. A free test shows within a session whether a seller has it or is reselling one of the other four.

Why NSUS infrastructure closes server-side exploits

ClubGG runs on shared infrastructure with the rest of the NSUS Group estate, including GGPoker, a regulated public operator carrying gambling licences in jurisdictions like Malta. The security requirements for those licences also apply to the shared infrastructure. TLS termination, application-layer encryption, server-authoritative card generation and audit-friendly logging do not get switched off for the club-app product.

The client is a display layer. Card information for a seat is produced server-side and sent only to clients with the right to see it; opponent hole cards are never transmitted to a client until showdown. A working remote code execution against NSUS infrastructure would be worth six or seven figures through coordinated disclosure, with serious jail-time exposure. A Telegram channel selling a $200 download does not offer that payoff. The UltimateBet 2007 and Absolute Poker 2008 cases were insider exploits by privileged staff, not external attacks resold to retail buyers.

The Diamond infusion myth

Unique to club apps and the most commonly sold variant on Telegram. The claim fails because Diamond balances are not stored client-side. The client displays the balance the server reports; if the client lies, the next server action (joining a table, transferring Diamonds, paying rake) fails to validate and the discrepancy gets reconciled to the server's view.

Club-side reconciliation also exposes discrepancies. Diamond movement is logged for the club operator, who sees transfer histories for every player. A balance that increased without a matching agent transfer triggers a question. Even if the platform ledger could be tricked, the club operator's books would not match, and the discrepancy is visible at the next settlement. Software sold as a "Diamond hack" is almost always a rebranded bot, a phishing page that captures credentials, or a remote-access trojan that drains whatever wallet is on the buyer's phone.

Why RNG prediction does not work

Modern shuffling on ClubGG, GGPoker and every serious operator uses a cryptographically secure PRNG seeded from multiple entropy sources, with the deck committed server-side before any card reaches a client. The data rate exposed by poker (roughly fifty bits of card information per hand at a few hundred hands per hour) is many orders of magnitude below the bandwidth a CSPRNG produces internally. You cannot reconstruct internal state from a heavily filtered, attenuated output. The iPoker 2013 incident was an implementation bug fixed within weeks, with no equivalent on any current operator.

Decision-support AI and its users

The only category with genuine engineering is decision-support AI, software that plays visible game state well. It is real and it works. Insiders do run it inside their own clubs, but it is also licensed to retail players, usually after a free test in a Telegram group where the client checks the interface, real-time decisions and stability for himself. Profit is not what that test measures; that takes volume. The architecture mirrors a GGPoker bot with three differences tied to the club it plays in.

Solver-anchored baseline
Pre-computed strategies from CFR-family solvers: PioSolver or GTO+ for heads-up and 6-max trees, MonkerSolver for multiway and PLO. Pluribus (Brown and Sandholm, 2019, Science) is the reference for superhuman 6-max NLH; DeepStack (Moravčík et al., 2017) for heads-up. The engineering problem is compressing solver output to a runtime-queryable form inside a mobile inference budget.
Online opponent model under stable club identities
Club tables have the same regulars across sessions, which removes the anonymous-rotation problem that dominates GGPoker bot work, so long-horizon, per-opponent modelling becomes feasible again. The bot can profile each regular and adjust to how each one actually plays. The owner watches those same regulars and can flag a bot by ordinary observation, so the edge and the exposure both come from everyone knowing everyone. For a retail owner that is the argument for table discipline: no table filled with bots, normal stack sizes, volume that grows gradually, timing and bet sizes that vary the way a person's do.
UI automation on Android-class hardware
ClubGG is mobile-first. The automation surface is accessibility-service-driven on Android. The visible client changes every few weeks as the operator pushes updates, and the screen-scrape or accessibility-tree layer is the most fragile part of the pipeline. iOS is harder; the alternatives are unstable across OS versions. The team answers in the chat whether a given club's client and device setup is supported; a listing cannot promise that.

None of this is magic. It is software competing in a game, not breaking a game, and its return still depends on the deal behind the club it plays in. In a club with thin rakeback and tough regulars a strong engine barely clears the rake; the same engine in a soft club with fair rakeback holds a steady edge over volume. Judge it over months rather than one session, because short samples swing both ways and a one-time license is spread over all of that volume.

Where the deal makes a bot pay

The viable zone is defined by club economics, not stake level. Small-and-mid clubs at USD-equivalent $0.10/0.25 to $1/2 6-max are the most viable. The cards are not easier; these clubs have exploitable regulars, owners who care more about volume than scrutiny, and rake structures that compete for grinders by giving back more than a public operator would. Move up a tier and all three reverse (vetted players, attentive ownership, small samples), so the edge narrows and the scrutiny rises however good the bot is. High-stakes club games are a poor fit for this reason, not a technical one. A pure-heuristic bot without solver compute loses to mid-stakes regulars over volume, so the floor is solver-anchored even at micro. Above the floor, the limiting factors are the rakeback math and the club you pick, which is why a retail owner should choose clubs as carefully as he chose the bot and add volume gradually.

Why the retail scam survives next to real bots

Because ClubGG and GGPoker share ownership within the NSUS estate, any signal you generate on either should be assumed observable to the other. Device fingerprints, payment methods and KYC documents overlap into a plausible shared abuse-graph. That raises the cost of clumsy play for everyone, insider or outsider. A realistic device environment and steady, human-like conduct matter on both products, because a careless pattern on one can follow the account to the other.

The retail "hack" market targets players who want a one-button win, a card reader or a Diamond generator instead of software that plays well over volume. Sellers have near-zero costs in 2026: an LLM writes the landing copy, stock testimonials fill the social-proof slot, Telegram automation handles distribution. Sellers reuse one scam shell for ClubGG, PPPoker, PokerBros, X-Poker and the regional Asian apps, swapping only the logo and screenshots. A two-percent conversion on cheap traffic at a hundred-and-fifty-dollar order funds the operation forever, with no obligation to deliver anything that works. Telling a real seller from a scam one costs nothing. A real team sets up a free test in a Telegram group, lets you watch the bot make real-time decisions, sells a one-time license with no monthly fee, and tells you in the chat whether your club's client is supported. A scam seller wants payment first, promises results or opponents' hole cards, rents out a bot with a hidden cut of the winnings, or sends an APK that should never touch your phone.

See a club bot play first

The Poker Bot AI team sets up a free test in Telegram and sends pricing in the chat. Questions about club-level economics or anything else in this piece go to the same place.