What most “AI trading bots” actually run
Open the settings of most bots sold as AI and you find the same few mechanisms, each of them older than the label:
- Grid. Buy and sell at fixed steps inside a price range. At its best in a sideways market, painful when the market leaves the range.
- DCA. Buy more on a schedule or on each drop, lowering the average entry price. It works until the drop does not stop.
- Martingale. Increase size after each loss to recover it in one win. The loss it cannot recover is the one that ends the account.
- Signal rules. If this indicator crosses that level, open a position. Precise, and deaf to anything the rule did not mention.
These are legitimate tools, and we set rules against human judgment in copy trading vs trading bots. None of them is intelligence. They do the same thing in the same situation, forever, which is exactly their value and exactly their limit.
Where real models come in
Two techniques deserve the word more honestly. Statistical tuning uses recent data, such as volatility, to adjust a bot’s settings: the rules stay fixed, but their parameters move. Language models go further. Given the trade and its context, they return a decision, which means they can weigh things no rule anticipated.
That flexibility cuts both ways. A model that can reason about a situation can also reason its way into a bad trade and describe it convincingly. The more real the AI, the more the product needs a boundary the AI cannot cross.
Why the label is not harmless
In March 2024 the U.S. Securities and Exchange Commission charged two investment advisers, Delphia and Global Predictions, with making false and misleading statements about their use of artificial intelligence. They paid civil penalties of $225,000 and $175,000. It was the SEC’s first enforcement action over AI claims, and it settled something worth remembering as a buyer: in a financial product, “AI” is a claim someone can be held to.
Six questions to ask any AI trading product
- Which model makes the decision? A named model you can look up, or “our proprietary AI”?
- Can you see what it is asked? The instructions and the inputs, not just the output.
- What can it never do? A limit that sits outside the model, which no answer can override.
- What happens when it fails? Models time out and providers go down. Does the product stop, guess, or fall back to something defined?
- Can you compare it against no AI? On your own account, not in a marketing chart.
- Who is responsible, and does anyone promise returns? A promise of returns ends the evaluation on its own.
How CopyFi answers them
Our own product, held to the same six questions:
| How CopyFi answers | |
|---|---|
| Which model decides | One you choose: Anthropic, OpenAI, or your own compatible endpoint, with your own key. |
| What it is asked | Your instructions go with every decision, exactly as you wrote them. |
| What it can never do | Exceed the session’s leverage cap, size limit or direction filter. |
| When it fails | The session falls back to your strategy and keeps running. |
| A baseline without AI | Compare the model’s decisions against the strategy alone, on your own sessions. |
| Who is responsible | The decisions are your model’s. No returns are promised, by anyone. |
And where CopyFi is not AI at all: the strategy builder is rules, and copying a trader is human judgment. AI is optional and switched on per session. Without a model key, a session runs the strategy you built and nothing else. The full mechanism is in AI copy trading, with limits the model cannot break.
Questions people ask
Is a grid bot artificial intelligence?
No. A grid bot places orders at fixed steps inside a price range you set. It is useful, predictable automation, but it never decides anything its rules did not already specify.
Can an AI trading bot predict crypto prices?
No tool predicts prices reliably, and a product that claims it can, or promises returns, is the clearest warning sign in this category. A model can weigh context a fixed rule ignores. That is not the same as knowing what happens next.
How can I check whether a trading product really uses AI?
Ask which model makes the decision, what it is asked, what it can never do, what happens when it fails, and whether you can compare its decisions against a baseline without AI. A vague answer to the first question usually settles it.
Trading involves risk, with or without a model. Past performance does not guarantee future results, and CopyFi makes no investment recommendations.