← All posts · June 30, 2026 · 6 min read

How to spot bot wallets on Polymarket

Go looking for a wallet to copy and you'll quickly find profiles that look superhuman. Twenty thousand predictions. Green week after green week. Trades at 4am, 9am, 2pm, 3am again.

It's not superhuman. It's not human at all.

That matters because bots make lousy copy targets — their edge is machinery you don't get to borrow. Here's how to tell what you're looking at before you press copy.

The five tells

1. Trade counts no human can produce. A person who places 30 trades a day — which is a lot — does about 900 a month. A wallet with 15,000 predictions in three months is software. This is the fastest filter there is, and it's the first thing I check.

2. No sleep gap. Humans have a rhythm: a burst of activity, quiet hours, weekends that look different from weekdays. One researcher who spent two weeks sorting humans from bots on Polymarket found a wallet trading at almost exactly 4-hour intervals for 18 straight days, and another with a near-perfect spread across all 24 hours. No sleep, no lunch, no bad Tuesday. Metronomes aren't people.

3. Buying both sides of the same market. A wallet that buys Up and Down in the same round isn't confused — it's arbitraging. CoinDesk profiled a bot that ran 8,894 trades for about $150,000 mostly by buying both sides of 5-minute crypto markets whenever the pair priced under $1. The same two-week study found a wallet doing 200+ trades a day across mathematically linked market pairs, every trade a hedge, ratios precise to four decimal places.

4. Oddly exact position sizes. Humans bet $50, $100, $250. Bots bet $47.83, $112.06, $289.41 — the output of a sizing formula, not a gut. A history full of never-round numbers is a strong tell on its own.

5. Instant reactions, every single time. Anyone can be fast once. When the same wallet is among the first five trades after every market-moving event, around the clock, that's code sitting on a feed. The study's author put it plainly: by the time he could place a trade manually, the odds had already moved 2–3 cents.

How much of the market is bots?

More than the leaderboard makes obvious. Coverage of the platform describes arbitrage bots collecting millions while manual traders fall behind. The two-week wallet study looked at 847 tracked wallets and flagged an estimated 30–40 of the consistently profitable ones as clearly automated — and that's just the obvious cases. And in the 5-minute Bitcoin markets specifically, researchers identified 821 wallets running a last-second settlement-push strategy — we covered that in the manipulation post.

None of this makes bots evil. Most are doing boring, legal arbitrage. The problem starts when you try to copy one.

Why copying a bot goes wrong

The common advice is "copy the most active profitable wallet." That's backwards. Activity is the strongest bot tell on the board, and a bot's profit comes from things a copier can't inherit:

• Speed. Research on Polymarket returns found automated traders gain about 2.52 cents per contract in execution quality over manual traders. That edge lives in the milliseconds before your copy fires. Even a fast mirror — ours runs about 2 seconds — arrives after the cheap price is gone.
• The other leg. Copy one side of a hedged pair and you hold a naked directional bet the bot never took. It locked in a spread; you bought a coin flip, plus taker fees.
• Maker rebates. Plenty of bot profit is fee structure. Your copied market order pays the fee the bot was collecting.

A 5,000-trade-a-month wallet with a lovely profit curve is often the single worst thing on the platform to mirror.

The exceptions

Not every automated wallet is uncopyable. Slow, methodical wallets doing 5–15 trades a day on judgment-shaped markets held up best in that same study — their edge is the pick itself, and a pick survives a 2-second delay just fine. That's the real question to ask about any wallet: does the edge survive the delay? Judgment does. Speed doesn't.

Filtering this automatically

Everything above can be done by hand with a block explorer and patience. It's tedious, which is exactly why btc5min's Market Scanner has filters that encode these tells directly: max predictions caps out the hyperactive wallets, max both-sides drops the arbitrage bots, and a max-percent-sells filter catches churners. What survives the filters is the short list worth a human look — paste any of them into the Wallet Searcher for per-category P&L before you commit.

And once you have candidates, don't stop at the win rate — that number lies in its own special way, which is the subject of why win rates are misleading. The full pre-copy routine is in the step-by-step copy trading guide.

btc5min is an independent tool and is not affiliated with, endorsed by, or operated by Polymarket. Nothing here is financial advice. Prediction markets are risky, copying another trader does not guarantee profit, and you can lose the money you trade with.