Author:Jeff , IOSG

summary

The core data regarding the panic of prediction market bots is quite telling: 5% of wallets on Polymarket that appear to be bots contributed 75% of the platform's trading volume. 823 wallets have each netted over $100,000 since January 2025, collectively extracting $131 million in profits from Polymarket. Of the top 20 most profitable wallets, 14 were classified as bots (Stacy Muur leaderboard inspection). A University of Toronto study (covering 2.4 million users and $67 billion in trading volume since 2022) found that 68.8% of users were in a loss-making state, with the top 1% capturing 76.5% of all profits.

The resulting narrative is that the prediction market is a wealth transfer machine, and bots are its operators. The data is accurate, but the framework is half-biased.

Key points

First, the core flaw in the bot narrative lies in equating "concentration of trading volume" with "capital plunder." The fact that 5% of wallets on Polymarket contribute 75% of the trading volume only indicates the distribution of account activity and cannot directly prove that retail investors' funds are being extracted by bots.

Second, data at the group level is more convincing. The positive return rate of AI agent wallets is about 37%, while that of human wallets is only 7%-13%. This 3-4 times group-level gap is real evidence of structural advantage. However, 14 of the top 20 in the profit ranking are bots (Stacy Muur leaderboard inspection), which is a right-tailed projection of this distribution and not independent evidence.

Third, the advantage of bots lies in the structural dimension, not the judgment dimension. The three types of markets dominated by bots—price delay arbitrage, real-time sports game state automation, and cross-platform combination arbitrage—share the common feature of not requiring judgment on the real-world events themselves. Once market outcomes depend on the comprehensive processing of multi-source information, the advantage of bots is systematically weakened.

Fourth, Polymarket's category structure has shifted from "politics 42%" to "sports 50%" in the past 12 months. The fastest growing category is precisely the long-cycle event market where bots are structurally disadvantaged, and the overall trend of the platform becoming more retail-oriented is clear.

V. Forward-looking judgment : The proportion of bots will continue to increase as deployment costs decrease, but the scale of capital extraction by bots from humans will peak before the proportion of bots reaches its limit - because the speed at which bots cannibalize each other is faster than the speed at which they cannibalize human accounts.

VI. Investment Strategy : Equity opportunities in the platform layer (Kalshi + Polymarket combined 97%+ share) have been largely closed; value opportunities are shifting to the L2 agent infrastructure layer (Olas / Valory model) and the venue-agnostic middleware layer. C-end bot products and L3 data/pricing layers do not have venture-fit.

I. The scale of the track is larger than that of bots, causing panic.

Three quantitative anchors define the scope of this report.

First, on April 14, 2026, Bernstein revised its forecast for the market size in 2026E to $240 billion, and the consensus is that the path to $1 trillion by 2030 is on the sell-side.

Second, the combined YTD transaction volume of Kalshi and Polymarket exceeded $60 billion in mid-April 2026, surpassing the total of $51 billion for the entire year of 2025.

Third, Robinhood has launched over 1,000 Kalshi contracts, and its platform has over 1 million customers who have traded a total of 9 billion contracts. Robinhood's prediction market business has an ARR of approximately $350 million, projected to reach $150 million in 2025 and $586 million in 2026E, making it the company's fastest-growing product line.

The data above all point to one conclusion: prediction markets are no longer a purely crypto-native sector; their characteristics are more akin to a TradeFi distribution problem. The "retail investor plundering" group assumed in the bot narrative is not primarily composed of crypto users, but rather retail investors who entered the market through traditional brokerage channels.

This leads to the contextual bias of the bot panic: the track is not being automatically stripped of value, but rather being injected with traffic by mainstream finance at a pace far exceeding any automated extraction speed.

II. The truly important data: 37% vs. 10%

The most frequently cited data points in the bot narrative exhibit sample selection bias.

The data source for "14 bots among the top 20 in profit ranking" is based on a small sample that has already been sorted by profit. This sample can only reflect the occupancy of bots at the right tail of the distribution and cannot be used to infer the superiority or inferiority relationship at the group level.

Group-level data (Source: Polystrat/Valory disclosures, cross-validated with multiple Polymarket on-chain analytics data):

The 3-4 times win rate difference at the group level truly reflects the structural advantage of bots. The 14/20 profit ranking should be understood as downstream performance of this win rate distribution, rather than independent causal evidence.

III. In which markets have bots succeeded?

Bot extraction is highly concentrated in the following three market segments. What these three segments have in common is that they do not require subjective judgment of real-world outcomes, but rely on latency or pricing advantages related to the platform's matching engine.

Price feed arbitrage

Representative case: Wallet 0x8dxd, in January 2026, grew from $313 to $437,600 by trading BTC up and down contracts for only 15 minutes, with a win rate of 98%.

Strategy Principle: Monitor spot prices on Binance and Coinbase, and establish positions when Polymarket quotes lag behind CEX prices. Polymarket introduced a taker fee (peaking at approximately 3% with a 50% probability) for 15-minute crypto contracts on January 7, 2026, specifically neutralizing this strategy. The wallet's cumulative win rate has since fallen back to 54.7%.

Conclusion: Bots do have an advantage in the price-feeding market, but it is limited to a very narrow time window and is significantly reduced as the platform introduces friction costs.

Real-time sports game state automation

Data source: Polymarket wallet category from the cancun2026 team (Dune query 6648075, https://dune.com/queries/6648075, past 7 days, as of 2026-05-11).

Advantage stems from the fact that bots react significantly faster to events than retail investors using live streams (with a 30-second delay). Furthermore, trading terminals like Kreo and PolyCop extend this advantage to non-programmer users through copy-trade and auto-copy features, thus the measured bot share includes human funds routed by the bots.

Cross-platform combination arbitrage

Data source: IMDEA Networks paper "Unravelling the Probabilistic Forest: Arbitrage in Prediction Markets" (AFT 2025, dspace.networks.imdea.org/handle/20.500.12761/1941).

The study covers approximately $40 million in arbitrage withdrawals on Polymarket between April 2024 and April 2025, primarily consisting of two models: first, the rebalancing of YES/NO shares within the same market; and second, cross-platform portfolio trading (entering the market when the sum of the implied probabilities of buying YES on Polymarket and NO on Kalshi is less than $1). This model has rigid requirements for multi-platform infrastructure and compresses as the matching engines of each platform converge.

IV. The areas where human accounts win and their limitations

The category with the lowest bot market share is not because "individual consumers are more accurate in their selections," but because "profitability in this market depends on the ability to integrate multi-source real-world information." This is an area where automation continues to be at a structural disadvantage for humans.

Two independent studies corroborate this assessment.

Joshua Della Vedova's on-chain behavior research (jdellavedova.com) from the University of San Diego points out that retail investors are more likely to pick winning results than bots; the advantage of bots lies in execution—when retail investors buy YES at $0.72, bots have already entered the market at $0.55, with a floating profit of $0.17 per share.

A working paper from the University of Toronto/HEC Montréal/ESSEC ( Akey et al., SSRN 6443103, March 18, 2026) points out that 56% of losing users placed orders at extreme prices (<10¢ or >90¢), while only 28% of the top 0.1% of profitable users placed orders at extreme prices. Typical behavior of losing users is "chasing a low-probability, 20x return at 5 cents" or "chasing a high-certainty position at 95 cents," while typical behavior of profitable users is establishing positions in the middle of the probability curve.

Both studies point to the fact that retail investors' judgment ability is generally underestimated, but their timing of execution and order structure are systematically weak.

V. Forward-looking Path: Four Forces Determine the Future of the Bot/Human Landscape

The key variable for the next 12-24 months is not the current bot/human ratio, but its evolutionary direction. This report identifies four forces at play, which are not aligned.

Bot deployment costs are collapsing further.

Coding agents like Claude Code and Codex, open-source frameworks like Hermes, and Polymarket's own Polymarket Agents framework (open-sourced under the MIT license) have collectively lowered the engineering threshold for 0x8dxd-like strategies from "serious projects" to "weekend prototypes." Copy-trade services further integrate human funds into the bot infrastructure, mechanically amplifying the measured bot share.

BOT's individual yield is being eroded by its peers.

The 823 profitable bot wallets represent the right tail of a much larger group of losing bots. The increasing number of wallets using similar strategies implies a narrowing profit window for each bot. 0x8dxd's 98% win rate is structurally unreplicable, not because inefficiency has disappeared, but because of competition from similar bots and adjustments to platform fees. The scale of capital extraction from humans by bots will most likely peak before their overall bot share.

The platform's product category structure is tilted towards individual users.

Polymarket's category composition in April 2026: Sports 50%, Crypto 24%, Politics 16%, Other 10%. The composition in the same period of 2025: Sports 29%, Crypto 12%, Politics 42%.

Sports trading volume increased 11-fold year-on-year in absolute terms. The new growth mainly occurred in the long-cycle event market, where retail investors held a dominant position. Bernstein predicts that the share of sports in the trading volume of the sector will fall from the current 62% to 31% by 2030, filled by economic, political, and corporate event contracts—this structural shift will further expand the category exposure where BOTs are not dominant.

Different platforms naturally distribute traffic according to product category

Hyperliquid's HIP-4 launched on May 2, 2026, offering daily BTC binary contracts, zero entry fees, USDH collateral and perpetual/spot unification, and a validator-slashable market deployment mechanism (1 million HYPE per slot, approximately $42.76 million at current prices).

This is a typical example of a bot-dominated market type being separated and listed separately. Day-1 trading volume mainly came from arbitrage funds, consistent with the historical distribution of BTC binary contracts. If HIP-4 subsequently expands to the sports and political markets and integrates trusted oracles, its bot share may converge to the Polymarket level; in the current stage, its role is to isolate bot-friendly traffic to an independent platform, further shifting Polymarket's category structure towards retail investors.

VI. Platform Landscape and Valuation Snapshot (Mid-2026)

▲ Source: Bernstein note (April 14, 2026), Polymarket / Kalshi publicly disclosed, HIP-4 launch announcement

Conclusion: Kalshi + Polymarket combined account for over 97% of the market share, meaning platform-level equity opportunities are essentially closed for venture check sizes. Investable value is shifting towards both sides of the platform layer: above (trading terminal, quantitative strategy services, agent infrastructure) and below (capital efficiency, arbitration, oracles).

VII. Risk Warning

Risk 1: Regulatory Tail Risk . The three bills submitted by Schiff (the DEATH BETS Act, the Public Integrity Act, and the Prediction Markets Are Gambling Act), Nevada's TRO against Kalshi, and Arizona's criminal charges in March 2026 constitute a tug-of-war between the federal government and various states. Kalshi's 89% concentration of sports revenue is its most exposed business line, making the scenario of a complete ban on sports or war/death-related contracts a realistic possibility.

Risk 2: Oracle and Arbitration Failure Risk. Polymarket integrated with Chainlink in 2025 to process price-based markets, but still relies on UMA for subjective markets. Currently, UMA's token economy generates only about $600,000 in economic flow annually, corresponding to a FDV of $37 million. After MOOV2, proposer rewards were narrowed to approximately 37 whitelisted addresses, most of which are Polymarket affiliates. Any controversial, high-profile ruling could trigger a reassessment of trust in the entire sector.

Risk 3: Risk of a reversal in sports market share. Polymarket's sports business growth in 2026 is seasonal (driven by the NBA and NFL Super Bowl). If the sports market share declines, the overall dynamic of "increased BOT market share + retail expansion" may reverse.

VIII. Meaning for builders and investors

The BOT debate essentially boils down to one question: in Bernstein's projected $240 billion 2026 market, which layer captures value? The four-layer structure has varying value densities.

L1 — Agent trading products. The strategy's advantage diminishes, and automated trading on the consumer side carries compliance risks. Solo betting is not recommended at this layer.

L2 — Agent Infrastructure (Olas/Valory model). An economic model that collects fees regardless of which agent wins. This layer is the cleanest investable option.

L3 — AI-native data, pricing, and market creation. Most of this is absorbed by internal platform teams or taken by existing Web2 incumbents (Kensho, Bloomberg, Dataminr). The remaining investment window is narrow.

L4 — Arbitration and Resolution. Current economic flows are real but small in scale. To become a Tier 1 venture, a token model redesign is required, which is not currently on the public roadmap.

The directions worth tracking at the edge layer:

PM-DeFi composability (Morpho collateralized PM positions, currently 2x leverage, roadmap 4-5x, impacting capital efficiency)
Trading terminals and copy-trade services (Kreo, etc.)
PM-native Quantitative Research
New market primitives (impact markets, futarchy, conditional markets)

Conclusion: Bots win in product categories, people win in the market, and platforms win in terms of structure.

Bots haven't taken over the prediction market. Bot saturation occurs in specific market types; the ratio of bot to human trading volume on any platform is essentially a downstream consequence of that platform's market structure. Headline data like "5% wallets / 75% trading volume" confuses concentrated trading volume with capital exploitation. Polymarket's 2026 growth primarily came from the sports market, where bots are structurally disadvantaged, while the $131 million in bot withdrawals mainly occurred in the short-window crypto market where retail participation is low.

The winning platform in the future will need three capabilities: to support multiple market types under trusted arbitration, to accommodate bot and human traffic in an appropriate ratio, and to retain users across different categories. Polymarket currently occupies this position: Bitget's Q1 2026 research shows that multi-category users are experiencing organic growth, with the number of categories per user increasing from 1.45 to 2.34, and the number of active days increasing from 2.5 to 9.9.

Bots remain in their structural advantage zone; the human capital running bots will continue to migrate to the next event; the ultimate winner will be the platform that can handle both types of traffic in the most market types and in the right proportion.

Vitalik: The quality of prediction markets depends on their oracles

Popular Articles