· 3 min read · polymarket · bots
The bot census: machines became the majority of Polymarket in 2026
We counted the machines. From 1.52 billion on-chain fills: accounts trading at machine scale produced 53% of all Polymarket volume in 2026 — up from 29% in 2024 — and 74% of all trades. One account filled 947 orders in a single second.
By CrowdIntel
Every "the market thinks" headline treats a Polymarket price as the verdict of a crowd. So we asked a question nobody had answered with data: how much of that crowd is human?
We rebuilt Polymarket's entire trading history from the Polygon blockchain — 1.52 billion fills across 2.9 million accounts — and counted machines the only honest way you can on-chain: by behavior. No accusations, no guessing identities. Just cadence.
The answer: the machines became the majority this year.
The takeover timeline
Take every account that placed at least 10,000 fills within a single calendar year — roughly 27+ trades a day, every day, as a floor. Call that machine scale. Here's their share of Polymarket:
| Year | Machine-scale accounts | Share of volume | Share of trades |
|---|---|---|---|
| 2024 | 350 | 28.9% | 38.2% |
| 2025 | 2,270 | 45.6% | 62.8% |
| 2026 (to June 15) | 12,195 | 53.0% | 73.7% |
Two years ago, machine-scale accounts were under a third of the market. Today they're 53% of the dollars and nearly three-quarters of the trades — and the count of machine-scale accounts grew 35× while it happened.
Half of every trade ever, from 0.06% of accounts
Zoom out to all of history and the shape is even starker. Of 2.9 million accounts that ever traded:
| Lifetime fills | Accounts | Share of all fills | Share of all volume |
|---|---|---|---|
| 100K+ | 1,678 (0.06%) | 50.0% | 24.3% |
| 10K–100K | 12,742 | 21.9% | 27.6% |
| 1K–10K | 84,878 | 16.2% | 22.2% |
| Under 100 | 2,345,802 (81%) | 3.7% | 10.3% |
1,678 accounts made half of all trades in Polymarket history. The 81% of accounts that are recognizably human — fewer than 100 lifetime fills — produced 3.7%.
No hands are that fast
Could these just be very committed humans? The cadence says no:
- The 100K+ tier averages 6,880 fills per active day — one trade every 12.6 seconds, sustained.
- 99% of them have traded in all 24 hours of the day. They don't sleep.
- Among the top 1,000 accounts by fills, 72% of fills land in seconds where the same account fills 2+ times. The record: 947 fills by one account in a single second.
- The busiest account placed 12.9 million trades in 115 days — 112,000 per day — averaging $5.50 per fill.
That last number gives away the game: these are micro-fill, high-frequency liquidity machines. In fact, the top 1,000 accounts by trade count made 44% of all fills but only 19% of dollar volume. The busiest accounts and the biggest accounts are two different elites — a machine class that makes the market, and a whale class that moves the money.
What this means (and what it doesn't)
To be clear about what we are not saying: automation is allowed on Polymarket — there's a public API and market-maker programs. Bots providing liquidity is how most modern markets work. Nobody is being accused of anything.
But it has never been quantified here, and it changes how you should read a prediction-market price. "The market thinks there's a 62% chance" means, most of the time in 2026, "a few thousand automated accounts trading with each other have priced it at 62 cents." Sometimes that's the wisdom of crowds. Sometimes it's two market-making algorithms breathing at each other.
If anything, our numbers are a floor: our per-account attribution slightly undercounts the busiest market makers on some maker-side legs. The real machine share is at least what you see here.
Methodology, briefly
Full on-chain reconstruction of every Polymarket contract on Polygon, September 2020 → June 15, 2026: 1.52B fills, 2.9M proxy wallets, indexed into ClickHouse. Lifetime volume reconciles to Polymarket's own reported monthly volume within 2% (March 2026: our $10.38B vs their $10.57B). "Machine scale" is defined arithmetically — fill counts and cadence — never by guessing who runs an account. Every number in this post regenerates from a single query, and we re-verify all of them against live data before publishing anything.
CrowdIntel runs the only independent full-history reconstruction of Polymarket's on-chain activity. The same data powers our public API — wallets, markets, signals, leaderboards — so you can check our math.