AI Trading Bots: What They Are and What PolyBot Is

"AI trading bot" is a marketing label, not a technical category. It is applied to products that share almost nothing beyond placing orders without a human in the loop. Before comparing any two of them it helps to separate what the software actually decides from how the decision is described.

This page sets out the distinction, then states plainly which side of it PolyBot sits on. PolyBot is not an AI trading bot. It is a rule-based strategy bot for one specific venue, and the rules are published below.

What people mean by AI trading bots

The phrase usually covers three quite different things. The first is a model that learns its own signal: a classifier or regressor fitted to historical price data, whose entry rule is whatever the training run produced. The second is a large language model wrapped around a news or sentiment feed, converting text into a directional call. The third, and the most common in practice, is a conventional indicator bot with the label attached for positioning.

The three behave differently under stress. A learned model can drift when the market regime changes, and the failure is hard to see because nobody wrote the rule down. A language model can produce a confident call from a thin or stale source. An indicator bot does exactly what its arithmetic says, which is either a strength or a limitation depending on whether the arithmetic was any good.

None of this makes AI trading bots inherently better or worse than the alternative. It makes them harder to audit. If you cannot state the entry condition in a sentence, you cannot tell a losing strategy from a broken implementation.

How automated trading bots differ

Automated trading bots in the older sense are deterministic. The entry condition is a fixed expression over observable inputs: a threshold, a crossover, a price level, a clock time. Run the same inputs twice and you get the same order twice. That property is what makes backtesting meaningful, because the backtest is executing the same code path the live bot will.

The trade-off is real. Automated trading bots of this kind cannot discover a signal you did not think of. They also cannot silently stop working in a way you would not notice, which is the corresponding advantage. Every parameter is a number a human chose and can change.

The market for these tools spans a wide range: grid and dollar-cost-averaging bots on spot exchanges, funding-rate arbitrage across perpetuals, and AI bots trading equities on learned signals. They are not interchangeable, and the venue matters as much as the algorithm.

What PolyBot actually is

PolyBot trades Polymarket's 5-minute Up/Down markets on BTC, ETH, SOL and XRP. Each contract in those markets settles at $1.00 if its side is correct and $0.00 if it is not. There is no position to hold afterwards and no leverage to manage.

It runs two entry rules, both fixed numbers rather than learned parameters:

• Momentum. Enter early in the window when the 5, 15 and 30-minute Binance price changes all point the same direction and each one exceeds 0.2%.
• Favourite. Wait 90 seconds into the window, read the live Polymarket prices, and buy whichever side the market has already priced above $0.60.

Both rules are described in full, including their risk profiles, in the article on 5-minute trading strategies. Both run in a paper mode against a simulated balance with no wallet attached, and live trading carries a default per-trade cap of $10 that each account can change.

What PolyBot is not

There is no machine learning model in PolyBot. No classifier, no neural network, no language model, no sentiment feed. The thresholds above were chosen by a human, are visible in the configuration, and change only when a human changes them.

It is also not a general-purpose trading platform. It trades one venue, one contract type, and four assets. It does not touch spot markets, futures or perpetuals, and it does not hold custody of anything: live trading runs from your own Polymarket wallet on Polygon.

Calling it an AI trading bot would be the easier marketing and the wrong description. The honest one is narrower: a small deterministic strategy bot on prediction markets.

Which one you want

The useful question is not whether a tool is one of the AI trading bots on the market. It is whether you can state its entry rule, test that rule against data from the venue you intend to trade, and see the difference between the tested behaviour and the live behaviour.

A learned model can clear that bar with good instrumentation. A rule-based bot clears it by construction, because the rule is the specification. What neither can do is make a strategy with no edge profitable. AI bots trading a market with no edge available lose money efficiently, and Polymarket charges a taker fee on these markets that any edge has to clear first.

If you are evaluating tools in this category, the comparison worth running is between automated and manual execution on the specific market you care about. That comparison for 5-minute markets is set out here.