How I earned just under $1 using DeepSeek to help me choose the settings for my trading bot on TradeSanta, and how I could have avoided this pitfall.
DeepSeek shook up the AI world earlier this year, making even OpenAI nervous. Remember the market bloodbath in January 2025, when U.S. stocks briefly lost over $1 trillion? That chaos was triggered by DeepSeek’s R1 launch — a supposedly OpenAI-level model built for just $6 million.
Naturally, I was curious whether it could work with cryptocurrencies. So, I let it set up a DCA bot on TradeSanta. The result? A total profit of just under a dollar. Read on to find out what went wrong with this strategy.
Step one – the prompt matters more than you think
Before we start, an important disclaimer: I used DeepThink R1 to see if an AI could stick exactly to the rules I set by feeding it all the market info, trading pair and even tossed in my own forecasts.
But next time I’m gonna let DeepSeek loose. Instead of just R1, it’ll also use its search engine to dig up real-time data and figure out its own strategy from scratch.
If you think AI just magically spits out good answers, think again. As we all know it well now: a bad prompt = garbage output.
“Remember, AI operates based on the information and training it has received, and it may not fully grasp highly individualized or abstract concepts,” writes the Codecademy team, a platform offering free coding courses. Their article presents two case studies illustrating seamless human-AI collaboration, but we’ll keep it simple and share their key tips for writing effective prompts:
Clarity and specificity. The team advises avoiding ambiguity and being as specific as possible.
Context and background. Providing the right context significantly improves AI-generated responses. Sounds fair.
Conciseness and relevance. While context is important, brevity matters too. Overly long prompts can confuse AI or make it focus on less relevant aspects.
Instead of simply asking DeepSeek, “Hey, pick the best settings for my bot,” I followed Codecademy’s advice and crafted a well-structured, contextual, and concise prompt with one main goal: to find the best parameters for optimizing the bot.
Since the market was favoring Bitcoin’s growth at the time, I decided to go long on BTC using the DCA averaging strategy. Given that different providers have their own nuances in setting up strategies, whether it’s Grid or DCA, I explained how DCA works on TradeSanta to ensure that DeepSeek fully understood what it was dealing with.
I also clarified that my account was linked to the OKX exchange to be completely transparent with my AI partner. For test trading, I allocated $100.
For context, I explained that my market perspective was bullish, expecting Bitcoin to rise to $160,000–$200,000 in 2025. I also threw in some details about the market conditions at the time, such as geopolitical tensions and policy uncertainty.
Here, in these screenshots, you can see my final prompt:
The prompt worked — DeepSeek didn’t hesitate and provided everything according to my request. Though not on the first try — I initially got an error saying, “The server is busy.”
Step two – setting up the bot
So, here’s the action plan I received from DeepSeek:
Take Profit: 2%
Trailing Take Profit: 2%
Extra Order Step: 2%, with up to three extra orders
Order Size: $25
Stop-Loss: 10%
Indicators: Bollinger Bands and RSI to refine entry points
By the way, for those paying close attention, I didn’t feed all the available settings from TradeSanta — just the basics. For example, I didn’t select Martingale for order volume or Custom TradingView Start signals, as the latter requires the Maximum plan on TradeSanta.
Moving on — on February 7, I… well, we, with DeepSeek, launched our first bot.
Step three – watching the bot in action
On the very first day, the first order was executed, followed by an extra order. Telegram notifications kept me updated on every move. But after these buy orders, the bot went completely silent. Unfortunately, the market entered a downtrend, and the bot was waiting for the price to go lower to average out more or to go towards take profit.
I kept checking my TradeSanta account and monitoring the Telegram notification chat. Honestly, I even thought about stopping the bot, locking in the losses early and shifting funds to other strategies. But I chose to continue the experiment — it was more interesting to see how it played out.
On day six, the bot finally closed the trade.
The bot traded $50 as the price started moving toward take profit. Once it hit the take profit level, trailing take profit was triggered. However, the market reversed, and the price dropped, so the bot sold to lock in the profit. In the end, my profit was just $0.2.
Over six days, if we annualize this return, it comes out to 2% per month, or about 24% per year. However, I ended up with much less than the expected 2% per trade. That’s extremely low — not a great result.
So, where did DeepSeek go wrong?
The main issue was that DeepSeek suggested setting a trailing take profit, but the percentage matched the take profit itself. In case of a price drop, this almost guaranteed zero profit. If the trailing take profit had been lower than 2%, the price wouldn’t have needed to drop a full 2%, reducing potential losses.
The takeaway – what worked and what didn’t
Overall, the outcome did not meet expectations, and it’s clear that using identical values for the trailing take-profit and take-profit was a suboptimal approach by DeepSeek. Here’s what I learned:
– The bot effectively followed instructions – DeepSeek didn’t work magic, but it structured a reasonable setup.
– Trailing Take Profit can backfire – Since it was set at the same level as the main take profit (both at 2%), it resulted in a nearly neutral trade instead of maximizing gains.
– Annualizing the return gives about 24% per year, assuming similar performance. Not terrible, but also not a guaranteed goldmine.
So, can AI create a profitable bot? Well, kind of. DeepSeek did its job setting up a bot based on logical DCA principles. However, AI lacks market intuition and human control.
Would I use AI for bot setup again? Yes, but with manual adjustments. AI is great for structuring a baseline, but fine-tuning settings based on experience is still crucial.
In the end, this experiment reinforced a fundamental trading truth: automation is a tool, not a replacement for strategy. AI can assist, but human judgment remains key to optimizing results.
FAQ
Can AI help configure a trading bot?
Yes, but only if you craft your prompts wisely. To succeed, you need to be clear and specific, provide relevant context like market conditions and stay concise. While DeepSeek R1 can generate a settings guide, keep in mind that these settings don’t guarantee your bot will perform well. Success depends on real-time market dynamics, strategy adaptability, and ongoing optimization.