Ready to go further
Next: learn how to run the paper test properly
The paper trading page covers what to log, how long to run and why demo results are a best case.
What is included
- What it is, in plain words
- The formula or mechanism, with exact arithmetic
- How it hits grid, martingale and DCA bots
- A check you can run before you switch a bot on
Who it is for
- People who run or plan to run a trading bot
- Anyone reading a backtest or a bot sales page
- Readers who want the numbers, not the pitch
- Not advice, and not a signal service
Key facts
- An AI trading platform is a service, not a single method. It usually bundles market data, a model or rule that makes calls, a way to place orders, and a place to hold the account.
- Its claims come in layers: about the model (accuracy), about the results (returns) and about the service (speed, safety, support). Each layer needs its own evidence.
- Accuracy is not profit. A platform can be right on most calls and still lose money if the wrong calls are larger than the right ones, or if costs eat the small gains.
- Worked example (made-up): a platform claiming 70% accuracy was right on 70 of 100 paper signals and still ended -26.0% after costs (calculated by us on 29 Oct 2026, exact).
- The way to evaluate one is a paper run you log yourself. This page names no product and makes no claim about any.
What these platforms claim
The page that describes a typical platform lists many things. Sorting them by layer shows which ones can be tested.
- Model claims. "Predicts direction with high accuracy", "learns from every trade", "reads sentiment". These are statements about a method that you cannot see inside. See machine learning trading for what a fitted model can and cannot do.
- Result claims. A return figure, a win rate or an equity curve. These are the easiest to inflate, and they should always come with costs, the period and the number of versions tried.
- Service claims. Fast order routing, uptime, safe custody of funds, support. These are about the plumbing, not about whether the calls are any good.
- Convenience claims. "No code", "set up in minutes", "fully automatic". Convenience is real, but it does not change the risk of what the platform trades.
If you want a view of the software side alone, see AI trading software. If the platform is built around a chat assistant, see ChatGPT trading bot. This page looks at the platform as a whole and at how to compare its claims with evidence.
Questions that expose empty claims
A real answer is specific and can be checked. An empty one is a slogan. The questions are grouped by the part of the platform they test.
| Part | Ask | A checkable answer | An empty answer |
|---|---|---|---|
| Data | Where does the price data come from, and how far back? | Source, start date, gaps, how it is cleaned | "Institutional grade data" |
| Model | What does the model predict, over what time span? | Direction in the next hour, or size of move in a day, in plain words | "Predicts the market" |
| Accuracy | What counts as a correct call, and what is the size of a right call against a wrong one? | A table of gain per right call and loss per wrong call | One accuracy number with no sizes |
| Testing | Were costs included, and was there a period the model never saw? | Yes, with the costs listed and the unseen dates stated | "Backtested on ten years" |
| Live record | Is there an account statement for a real period? | Statements covering good and bad months | Screenshots, or "past results vary" |
| Orders | How are orders sent, and what happens on a failed order? | A described order flow, limits on size, retry rules | "Lightning fast execution" |
| Custody | Where do the funds sit, and can the platform move them? | Funds in your own account, with limited permissions | Funds pooled or moved by the platform |
| Fees | How does the platform get paid? | A clear fee, and whether it profits when you trade more | Unclear, or a cut of your deposit |
| Control | Can I stop it, and can I run it on paper first? | A switch you control and a paper mode on the same logic | Live only, or a demo with different logic |
A platform that cannot answer the accuracy and testing rows with numbers is unproven, however it is described. A promise of fixed returns or of no losing trades is a reason to stop, since no real trading method can honestly make one.
Worked example: accuracy is not profit
Everything here is made up. A platform says its model is 70% accurate. You run it on paper for 100 signals and record each result. A right call gains 0.3% of the amount traded and a wrong call loses 0.9%, a common enough shape when small gains are taken quickly and losses run. Costs (fee plus spread) are 0.2% a trade.
| Item | Value |
|---|---|
| Signals traded | 100 |
| Right calls | 70 (70%) |
| Wrong calls | 30 |
| Result from right calls | +21.0% |
| Result from wrong calls | -27.0% |
| Result before costs | -6.0% |
| Costs (100 x 0.2%) | -20.0% |
| Net result | -26.0% |
| Average per trade after costs | -0.26% |
| Share of right calls needed to break even | 91.7% |
Calculated by us on 29 Oct 2026, exact. The claim is true in the sense that 70 calls in 100 were right. It is also beside the point, because the break-even point is 91.7%: the average loss plus the cost, 1.1, divided by the average gain plus the average loss, 1.2. The advert gave one number and left out the three that decide the result. This is the same arithmetic as expectancy.
A hundred signals is also few. It is enough to catch a gross exaggeration, not to prove a good platform.
How to evaluate one with paper trading
- Write down the claims first, in your own words, before you start. Accuracy, return, drawdown, speed. Then you cannot move the goalposts afterwards.
- Run the platform on a paper or demo account and keep your own log of every signal, entry, exit and cost. See paper trading. Do not rely on the platform's own report.
- Compute the numbers that decide the result: average gain, average loss, win rate, result after costs and the worst fall from a peak. See drawdown.
- Run long enough to see more than one kind of market. Weeks to months, not days. A short run says little.
- Check the paper mode matches the live one. If they run different logic, a paper result tells you nothing about live.
- Treat paper results as a best case. Paper fills are cleaner than real ones. Slippage and thin markets cost more live.
How it hits bots
A platform often hosts bots of the usual kinds: grid, martingale, DCA, signal followers. The platform label does not change how those fail. A hosted martingale bot still has a margin wall, and a hosted grid still has a range it cannot leave. The platform adds two risks of its own: you depend on its servers staying up, and you trust its description of what the bot does. Check both before trusting size to it.
What this page does not tell you
It does not say any AI trading platform works or does not, and it names none. It cannot tell you how common empty claims are. It gives questions and a method, and the example numbers are made up.
Keep Going
Related Tools and Guides
Keep exploring
FAQ
Frequently Asked Questions
What is an AI trading platform?
A service that bundles data, a model or rule that makes calls, order placement and an account. The label does not say how it works or whether it makes money.
Does a high accuracy figure mean the platform is profitable?
No. If wrong calls cost more than right calls gain, or if costs are large, a platform can be right most of the time and still lose. Ask for the size of the gains and losses.
What should I ask a platform before trying it?
What it predicts, how it was tested with costs included, whether there is a live statement, where funds sit, how it is paid and whether you can run it on paper first.
How long should I paper trade a platform?
Long enough to cover more than one kind of market, usually weeks to months. A short run only catches gross exaggeration.
Is a platform the same as a trading bot?
No. A bot is the program that places trades by rules. A platform is the wider service that may host bots, supply data and hold the account. The bot inside still fails in the usual ways.
Are the numbers on this page real?
Example numbers are made up to show how the maths works, and the page says so. A figure only counts as exact when the page says it was calculated on a stated date. Nothing you enter in a calculator leaves your browser.
Can I use this to decide whether to run a bot?
It can show whether a bot's numbers hold together and how much an account can lose. It cannot say what the market will do, and it is not advice. Use it to ask better questions, then read the terms of your own broker or exchange.
Is this financial advice?
No. It is general education. The example numbers are made up and nothing here is a recommendation about any platform.
Know the Risk Before You Run It
Free calculators and guides, added every week. Leave your details and we will email you when new ones go up.
Trading carries a risk of loss