How much of Polymarket volume comes from bots?

The answer changes with the size of the market

There is no single number for how much of Polymarket is automated, and the versions of this question that circulate usually fail because they ask for one. Automation is not spread evenly across the platform. It is concentrated at the two ends of the size distribution for opposite reasons: in the shallow tail because almost nobody else is there, and in the very largest contracts because those are where continuous quoting is worth the effort. The band in between is where ordinary accounts make up the largest share of activity.

The most specific public figure comes from a CNBC analysis published on July 2, 2026, which pulled closed-market data from Polymarket's public Gamma API covering 2021 through the end of May 2026 and worked with Joshua Della Vedova, a business professor at the University of San Diego, on the wallet classification. In markets that turned over less than $10,000 across their entire life, wallets meeting his high-frequency threshold supplied more than 80 percent of the volume. Those markets are not a fringe case. Roughly 70 percent of all closed markets in the dataset fall into that band, and more than 45,000 of them recorded no reported volume at all.

What the word bot means in this data

Polymarket does not publish a flag saying which wallets are automated, so every bot share you will read anywhere rests on a threshold somebody chose. In this case the rule was a wallet making more than 50 trades in a single day or more than 1,000 trades in total. That is a behavioral proxy, and it is worth being precise about what it captures and what it misses.

It captures the two obvious kinds of automation: market-making systems that keep resting orders on both sides of a book, and arbitrage systems that watch for a price on one venue drifting away from a related price somewhere else. It will also capture a person who trades manually and very heavily, and it will miss a slow script placing a couple of orders an hour. So the honest reading of over 80 percent is not that four in five dollars are moved by software with certainty. It is that four in five dollars in the shallow tail are moved by accounts trading at a frequency no casual participant sustains. The distinction that matters for a reader is between accounts present continuously and accounts present occasionally, which is closer to the maker and taker split described in makers vs takers on Polymarket than to any question about software.

Automation and profit by market size

The same analysis reported where automated wallets actually earned, which turns out not to be where they dominate the volume share. Their take from the shallow tail was about $1.2 million over November 2022 to February 2026, which is a small number for a band covering the majority of markets on the platform. The money was in the middle and upper bands: about $50.5 million from markets between $1 million and $10 million in volume, roughly 38 percent of total bot profit, and about $35.1 million from markets above $10 million.

Market-share and bot-share figures as of 2026, from the CNBC analysis of Polymarket closed-market data published July 2, 2026, covering 2021 through the end of May 2026, with wallet classification by Joshua Della Vedova, University of San Diego. Profit figures cover November 2022 to February 2026. Order-book and coverage rows describe general structure rather than measured quantities. Gamma volume counts notional on both sides of a trade, a measurement choice covered in how much volume does Polymarket do.

Why the venue attracts this much automation

None of this is a leak or a loophole. Polymarket runs a hybrid design in which orders are matched off-chain through a central limit order book and settled on-chain on Polygon, and it publishes that order book service as a documented public API alongside a separate market data service. The platform's own developer documentation opens by inviting people to trade and integrate against it. Automated participation is the intended consequence of that architecture, covered from the access side in does Polymarket have an API.

There is also a direct financial incentive to quote continuously. Polymarket operates a maker rewards program that pays for resting orders based on their size, how close they sit to the midpoint, and how consistently they are kept on the book, with a minimum resting time before an order counts. That is a structure only a machine can satisfy efficiently across hundreds of contracts at once, and it is described in how do Polymarket liquidity rewards work. A shallow market with one automated quoter is often a market that would otherwise have no quotes at all, which is the part of this finding most easily lost when the headline number is read on its own.

What our own snapshot can and cannot see

Cent Signals tracks the 400 highest-volume Polymarket markets, so our data sits almost entirely in the deep end of the distribution CNBC measured and cannot speak to the quiet tail at all. Within that top slice, the same concentration pattern still shows up. In the snapshot generated on August 14, 2026, the ten largest markets by lifetime volume accounted for about 36.9 percent of all volume across the 400 tracked, and the top 40 accounted for about 76.5 percent. Median resting liquidity in that set was about $77,200, which is a serious book by the standards of the platform as a whole.

Lifetime volume and current depth are also very different quantities, and the gap between them is where automated quoting becomes visible to an ordinary reader. The Ethiopian prime minister market carried about $79.7 million in lifetime volume in that snapshot against roughly $17,400 of resting liquidity, a ratio no active contract would show. A long-dated contract can accumulate enormous turnover and still be quoted by very few participants once the question stops moving. The relationship between those two numbers is set out in volume vs liquidity on Polymarket, and the figures for any single contract, with their snapshot date attached, sit on that market's page, such as the Ethiopian prime minister market.

What this record does and does not establish

It establishes that the composition of participants on Polymarket depends heavily on which market you are looking at, and that the platform-wide volume headline is carried by a small number of very large contracts. It also establishes that during the sampled period the high-frequency cohort finished ahead in every size band, while the same reporting describes retail accounts finishing behind in both shallow and deep markets. That is consistent with the wider research on this venue, summarised in who actually wins on Polymarket.

It does not establish that any specific wallet is automated, that the shares hold today, or that automation is a defect. The classification is a trade-count threshold applied after the fact, the volume series counts notional on both sides, and the sample ends in mid-2026. We describe the composition because it explains why a thin market can move several cents on a single order and why a quoted price in the tail may be the work of one participant. We do not draw a conclusion from it about what anyone should do. The wallets we track, with their visible positions and snapshot dates, are on the traders index, and what we collect and deliberately leave out is set out on the methodology page.

Frequently asked questions