Criteria
Each product was measured on eight dimensions before the final winrate verdict. None alone decides the comparison. For the products I ran, winrate over 26,000+ hands decides the comparison. The eight dimensions help explain those results. Products I did not run are scored on the public-data dimensions only. The test was at micro-stakes on PPPoker (low on purpose, so per-hand fuel cost on the metered products did not dominate the bill). See methodology for the full reasoning.
1. Supported rooms
Native room support, manual setup requirements, and unsupported platforms. Two products in this comparison (Inhuman, Deepermind) do not list PPPoker support, so using them there means adapting them by hand; I did not run either. The Android-emulator products (NZT, PokerBotAI) support more Asian/clubby rooms than the OpenHoldem family does. None of the tested products run on the regulated tier-one rooms (Pokerstars, GGPoker public side).
2. Technical stack
The products used three engine families:
- OpenHoldem family (Inhuman, Warbot, PokerBot.com profiles): screen-scrape + table-maps + custom strategy profiles, Windows VM, semi-automatic table seating. Open framework, lots of community content, older detection surface.
- Proprietary Windows engines (Shanky/BonusBots, Deepermind): vendor-specific engine, screen-scrape based, vendor-controlled profile distribution. Shanky uses a proprietary scripting language (PPL). Deepermind layers a genetic-algorithm decision policy on top.
- Android-emulator family (NZT Poker, PokerBotAI): emulator + cracked client + traffic-sniffer reading the game state directly from network packets + a cloud AI server that returns the recommended action + tap-emulation or packet-substitution for execution. It is the newest architecture in the field and the only one that finished positive in my test.
3upgaming falls outside these families. It runs an opaque vendor-installed VM (no documented engine).
3. Detection risk
Vendor anti-detection guidance, shipped randomisation, and observed field performance. Architecture mattered most for detection. The OpenHoldem family's screen-scrape pattern is a well-known detection signature on the rooms that bother to check. The two older desktop engines I ran, Warbot and Shanky, both had accounts banned during the test. The Android-emulator family is currently lower-profile because traffic-sniffer setups don't leave the same client-side fingerprint. As detection methods change, that advantage is not permanent.
Detection risk depends much more on operator discipline (unique residential IPs, sane GPS, plausible session schedules, coherent IP-location-timezone-time-of-day combinations across logins) than on the bot product itself. Vendors publish these recommendations to keep accounts alive.
4. Supported game types
NLH is universally supported. PLO is supported by most. MTT support is patchier and tends to be the realistic ceiling on bot performance. Long sessions, shifting stacks and ICM decisions break the simple strategy-profile approach. Shanky, Warbot, and the Android emulators support HU. OFC and short-deck (ROE) are only on PokerBotAI in this field.
5. Setup friction
Installation difficulty. The OpenHoldem family requires VMs, table maps, profile selection and ongoing maintenance, making installation difficult. Shanky and Warbot have the most polished install docs in that family. The Android-emulator products are easier (they ship documented installers and the vendor's support walks you through it), though the manuals themselves are uneven. 3upgaming installs everything for you, but leaves you with no idea what's running on the VM. Inhuman, Deepermind, and PokerBot.com are difficult to install.
6. Pricing
Pricing models:
- One-time licence + maybe profiles: Shanky ($129/year NLH, $189/year all-PLO), Warbot ($104.89/year base + $30 per additional PC, +$10–200 for custom profiles), PokerBot.com ($50–150 one-time per profile), Inhuman (~$100/year + ~$25 VM image), Deepermind ($25/month, $49/year, $499 lifetime).
- Fuel / per-hand: NZT Poker (no flat licence, per-hand fuel cost scaled by room/format/stake/chip price). PokerBotAI (upfront access fee ~$500–2,000 per supported room + per-hand fuel).
- Subscription: 3upgaming ($1,200–1,500 setup + $600/month per bot).
Trials vary widely. Warbot's trial lasts ten minutes; Inhuman's lasts five. Shanky offers a 200-hand trial. Deepermind has a free strategy tier. PokerBotAI gives operator-controlled free play for roughly a day. NZT and 3upgaming have no trial at all.
7. Support and updates
Vendor responsiveness and update frequency. Shanky and Warbot are long-running vendors with regular updates and direct developer access via email or Telegram. PokerBot.com is functionally end-of-life. The owner is selling the project. Inhuman relies on a single developer and sparse updates. Deepermind offers personal contact with the developer post-purchase but low update frequency. The Android-emulator products (NZT, PokerBotAI) both have dedicated Telegram support managers, development teams, and regular updates. 3upgaming has support that could not explain its own engine.
8. Public trust and reviews
Community reports include genuine endorsements and competing-vendor slander:
- A 2020 VICE write-up on Shanky/BonusBots mentions one profitable tournament cash on a single test; persistent BlackHatWorld and forum threads place its realistic ceiling at micro-stakes.
- A candid FAQ from the Warbot vendor places the realistic ceiling at NL50 cash or $30 MTT buy-ins, depending on the room.
- NZT has active Reddit / forum reports of bans on tier-one public rooms (which it doesn't market itself toward), pattern detection on sizing and timing in some clubs, and 2022–2023 negative threads with a Russian-language footprint.
- PokerBotAI has fewer public negative threads than NZT, a Hong Kong legal entity, and an active YouTube and social presence. Reviews are mixed but skew less toward scam accusations than NZT.
- 3upgaming presents little technical evidence of efficacy, marketing-heavy blog content, and no honest trial period.
This does not get a score in isolation, but influences how I weight the winrate result. If a product had a strong winrate but a long trail of credible scam reports, I would say so. None of the positive-winrate products in this comparison fit that description.
9. Final verdict — winrate over 26,000+ hands
The final column on the comparison page is the winrate verdict. The two neural-net bots, NZT and PokerBotAI, finished ahead (NZT clearly, PokerBotAI within noise). The rule-based bots I ran were negative. Products I did not run have no winrate verdict. 3upgaming was actively losing chips at scale.
For the per-product detail, return to the comparison. For how the test was set up, see methodology.