Guides · Updated 2026-11-09 · Backtest and paper results only, not financial advice.

How to Paper Trade an AI Trading Bot: Brokers, Setup and Pitfalls

To paper trade an AI trading bot, open a free paper account at a broker whose API supports it (Alpaca is the easiest, since paper is its default), point the bot at the paper endpoint only, freeze its rules, and let it run on live prices with simulated money for long enough to see real trades. Then check two things: did the bot actually place the orders its rules call for, and do the results look like its backtest and beat simply holding the market? Below is the step-by-step version, with the brokers that offer paper modes and the paper-mode bugs we found in ChatGPT-written bots. Everything here is about backtests and paper trading, not financial advice.

Paper trading vs backtesting

A backtest replays the past. It's fast, but it's built with hindsight, so it's easy to tune until it looks great. Paper trading runs forward: the bot sees each new price as it arrives and has to decide without knowing what happens next. That makes a paper record much harder to fake, which is why every bot on our leaderboard has both. Since 2026-10-16, each viral bot we tested has been paper-trading forward with $10,000 of simulated money and no rule changes, next to its backtest from 2019-01-02.

Paper trading still isn't real trading. Alpaca says plainly that its paper environment doesn't account for the market impact of your orders, price slippage due to latency, order queue position, price improvement, regulatory fees or dividends (Alpaca paper trading docs). Treat paper results as a best case.

Step 1: Pick a broker with a real paper mode

Not every broker that lets AI trade has a paper mode for it. From our broker AI tracker (15 brokers, checked 2026-10-21), these list some kind of paper, demo or sandbox access:

Alpaca: Yes, default (ALPACA_PAPER_TRADE=true); $100k simulated; ChatGPT plugin is paper-only for orders
Interactive Brokers: Paper accounts usable, but connector can't be restricted to paper only
TradeStation: Simulated environment for API developers (by request)
Tradier: Yes, separate paper token (PAPER_TRADING=true)
tastytrade: Sandbox (no market data, resets every 24h)
moomoo: Yes, paper trading with real market data (OpenAPI)
Webull: Yes, UAT environment (default)
NinjaTrader: Yes, separate Demo server
E*TRADE: Sandbox with canned test data only
Leverate (via partner brokers): Yes, demo accounts supported

These don't, or we couldn't confirm one: Robinhood, Public, Questrade, Charles Schwab, Zerodha (India). Robinhood is the big one. It lets an outside AI agent trade through its MCP, but our finding is “no paper mode described in robinhood's agent docs” for those accounts, and trade approvals are off by default (source). So there's no safe practice run there. For a full comparison, read which brokers let AI trade for you.

If you want the shortest path, use Alpaca. Paper is the default, the account is funded with Yes, default (ALPACA_PAPER_TRADE=true); $100k simulated; ChatGPT plugin is paper-only for orders money, and its minimum balance is $0 (Alpaca).

Step 2: Write the rules down and freeze them

Before the first paper trade, write the bot's rules in plain English: what it buys, when it sells, its stop-loss, its position size and its timeframe. Then don't touch them. If you tweak the rules every time the bot has a bad week, you're back to curve-fitting, just slower. Our tested prompts spell out the rules for 5 viral bots, and each strategy page lists the exact assumptions we used.

Step 3: Hard-wire the paper endpoint

The scariest paper-trading bug is a bot that isn't on paper. Make the paper address part of the code, not a setting you might flip by accident. The Alpaca adapter in our engine has the paper URL written into it and has no way to reach a live account. Watch for brokers where live is the default or where paper can't be enforced:

tastytrade: live orders default to production, so you have to point the bot at the sandbox on purpose (tastytrade sandbox).
Interactive Brokers: paper accounts usable, but connector can't be restricted to paper only (IBKR).

Use separate API keys for paper and live, and keep the live keys off the machine you're testing on.

Step 4: Check that the bot places the strategy's orders

This is where the ChatGPT-written bots we tested fell short. When we had ChatGPT write the 5 viral bots, every backtest ran, and 2 of them matched our engine's numbers exactly. The paper modes were a different story:

The BTC mean-reversion bot's paper mode only checked the Alpaca paper account and never placed an order.
The dip-ladder bot's paper mode only bought and never sold at its profit targets.
The multi-asset bot's paper mode dropped the ATR stop-losses and used a fixed 10% position size.
The regime-switching bot's paper mode was a manual buy/sell helper, not the strategy running on its own.

So after the first few signals, open the paper account's order history and check each order against the rules by hand. A paper record of the wrong strategy is worse than none, because it looks like proof. The full test notes are on the prompts page and in our ChatGPT and Claude bot guide.

Get the weekly receipts. One email a week with what the viral bots actually did, new tests, and broker AI-access changes. Free, plus the Bot Builder Starter Kit. Subscribe to the newsletter →

Step 5: Add realistic costs

If your bot logs its own simulated fills, make them pessimistic. Our engine uses the same rules for backtests and forward paper runs: signals are computed on a bar's close and filled at the next bar open, so the bot can't trade on a price it hasn't seen yet. Every fill is charged slippage of 5 basis points (0.05%) on SPY, QQQ and GLD and 10 basis points on Bitcoin and Ether, with $0 commissions. A bot that only looks good with perfect fills at the signal price isn't ready.

Step 6: Run it long enough to mean something

Count trades, not days. A viral 20/50 moving-average bot we tested made 17 trades in its whole backtest from 2019, about 2.2 a year. A few weeks of paper trading would show you almost nothing about it. Our own forward run is a good reminder: with data from 2026-10-16 through 2026-11-09, the bots on our board have 6 closed trades between them, so it's far too early to judge any of them.

Slow bots need months. Faster ones can show a pattern sooner, but even then you want dozens of trades before trusting a win rate, and you should watch the drawdown, not just the wins. A viral dip-ladder crypto bot we tested won 100% of its 309 backtest trades and still fell 44.0% from its peak, because it never sells at a loss (trade log).

Step 7: Compare paper with the backtest and with buy-and-hold

At the end of your paper run, ask two questions. First, did the paper trades match what the backtest would have done over the same dates? A big gap usually means a bug or a timeframe mismatch, like a bot backtested on daily bars but traded on hourly ones. Second, did it beat simply holding the same market over those dates? If not, the bot added risk without adding return. Our do AI trading bots work? guide shows how the 5 viral bots did against buy-and-hold.

Before you go live

Paper trading is a test, not a green light, and nothing here is a recommendation to trade. If you do move to real money, start much smaller than your paper account and turn on per-trade approvals where the broker offers them. Questrade, for example, uses push approval in app for every order; agent can't place, change or cancel on its own (Questrade). Keep checking the order history. To compare tools that run bots for you, with prices sourced from each company's own page, browse the AI trading tools directory. And before trusting anyone else's paper or live results, run through the 10 red flags.

FAQ

What is paper trading a bot?

Paper trading means running a trading bot's exact rules on live market data with simulated money. The bot makes its decisions and sends orders, but the fills are simulated, so no real money changes hands. It sits between a backtest (past data) and live trading (real money).

Where can I paper trade a trading bot for free?

Alpaca is the simplest start: paper trading is the default and the account comes with Yes, default (ALPACA_PAPER_TRADE=true); $100k simulated; ChatGPT plugin is paper-only for orders money. In our 15-broker tracker, 10 list some form of paper, demo or sandbox mode for API or AI access, including Webull, Tradier, moomoo and NinjaTrader. Check each broker's own docs before relying on it.

How long should I paper trade a bot before going live?

Long enough to see a meaningful number of trades, not a set number of days. A viral 20/50 moving-average bot we tested averaged about 2.2 trades a year in its backtest, so a few weeks of paper trading would show almost nothing. Compare the paper fills with what the backtest would have done over the same dates.

Is paper trading the same as real trading?

No. Alpaca's own docs say its paper trading doesn't account for market impact, slippage from latency, queue position, price improvement, regulatory fees or dividends. Real fills are usually a bit worse, so add a cost estimate and expect some gap.

Can ChatGPT write a bot that paper trades?

It can write one, but check what the paper mode actually does. In our tests of 5 ChatGPT-written bots, every backtest ran, but the paper modes often didn't follow the strategy: one never placed an order, one only bought, and one dropped the stop-losses. Always read the paper account's order history.

Get the weekly receipts. Subscribe free and get the Bot Builder Starter Kit (14-page PDF): 5 tested prompts, a hype-detector checklist, and safe paper-trading rules.

Form not showing? Subscribe on beehiiv