In my first experiment, I tasked DeepSeek AI with configuring a BTC/USDT trading bot on TradeSanta to execute trades based on my personal market predictions and risk tolerance.
This time I decided to test a new hypothesis without limiting DeepSeek to my personal predictions and preferences. No micromanagement involved — it’s up to the AI to capitalize on market opportunities and choose what it considers the best crypto trading pair.
So, I started by giving the task to DeepSeek with the prompt:
“Choose the best trading pair and settings for TradeSanta bot yourself.”
To sharpen its edge, I activated AI’s advanced capability — DeepThink R1 and Search (which, admittedly, was a bit glitchy at the time, limiting its data to January 2025). I briefed it on TradeSanta’s bot mechanics, including what trading tools it has, the specifics of algorithmic trading bots’ work, and disclosed my $100 budget — a modest exposure for this experimental phase. Now, let’s see what I got in this experiment.
DeepSeek’s pick was longing SOL
After processing market trend, DeepSeek’s machine learning model gave me a detailed answer. Its choice was focused on a long position on SOL/USDT, using a DCA strategy. Here I would like to note that at the time of the prompt in early February, SOL had been trading bullishly, bouncing between $170 and $200, on the very day of the request, on February 18, the landscape started to shift, with the LIBRA memecoin scandal among the high-profile reasons for the pullback. But for the sake of an honest and clean experiment I followed DeepSeek choice completely.
Here’s how DeepSeek explained his choice:
“Based on the search results, Solana stands out as a high-potential candidate for short-term upside with moderate risk, supported by the company following factors:
- Liquidity & Trading Volume: Solids among the top 5 cryptocurrencies by liquidity, with a daily trading volume exceeding $1 billion and a narrow bid-ask spread (0.025-0.1%), ensuring efficient entry/exit for traders.
- Recent Performance: SOL surged 24.7% in January 2025 due to memecoin speculation and DEX activity on Solana, outpacing Ethereum’s DeFi growth.
- Analyst Forecasts: Analysts predict continued momentum for Solana-based DeFi and NFT projects, with potential price catalysts from institutional adoption and Layer 2 scaling advancements
- Risk Profile: While volatile, SOL’s liquidity and institutional backing mitigate downside risks compared to smaller altcoins”
Moving on, I also got the exact settings to create the bot on TradeSanta’s website. This time the take profit level is at 2.5% and trailing take profit — 1.5%. 2% step for extra orders capped at 4 extra orders maximum, and an 8% Stop Loss. The order size is $20 according with my investment of $100. The model also chose RSI technical indicator for more accurate entry. So with these settings, I launched my DCA on SOL/USDT on February 18.
Results of trading using AI
On the 18th of February the bot entered the market by buying $20 worth of SOL when it was trading at $170.53 and put the strategy live, starting to track the take profit level, which was set at 2.5% targeting around $174.79. However, the market had other ideas and plans. Solana’s token began to fall due to multiple factors, including the fading meme coin frenzy, Solana’s March 1 token unlock, which raised fears of oversupply, and the network’s worsening reputation due to its ties to high-profile scams. Together, these factors have significantly dampened Solana’s market sentiment.
But DeepSeek has taken a dip into considerations in its recommendations algorithms, and in case of a 2% pullback in SOL, the bot was ready to place up to 3 extra orders. So when SOL dropped by the set percentage, the trading robot placed the first extra order, and after a further 2% fall – another order. In total, 3 buy orders were placed on February 18 when SOL reached $170.53, $167.25 and $163.83 respectively. As a result of this — the average price of my position was around $167.20 and so now to reach the take profit level the price of SOL had to be $171.6 instead of initial $174.79.
The next day, SOL recovered slightly, and our DCA bot was able to reach the take profit level – after which it activated the trailing take profit.
So, as you remember, our trailing was monitoring for that moment when SOL would drop by 1.5% from SOL maximum price reached after take profit level. So the price managed to climb up to $174.12 before pulling back by 1.5% from the peak to $171.29 – that’s when the bot was triggered close my position.
So let’s crunch the numbers and see the result. Within 1 day my bot placed 3 buy orders: 0.117 SOL + 0.119 SOL + 0.121 SOL ≈ 0.358 SOL which in total cost me 60USDT. The bot sold at the moment when the selling price was $171.29. Thus, converting into USDT we get: 171.29USDT × 0.358 SOL ≈ 61.32 USDT.
So my profit amounted to $61.32 – $60 ≈ 1.32.
To complete the picture of the robot performance, let’s see the profitability for 1 day of trading and extrapolate it into a monthly and annual profitability using a simple interest (considering the performance would be the same for the whole period of time which of course is more of a hypothetical than a real situation on the market):
Profit (%) = (Profit / Investment) × 100 = (1.32 / 60) × 100 ≈ 2.2%.
Per month: 2.2% × 30 = 66%
Per year: 2.38% × 365 = 803%
Trading results analysis by DeepSeek AI
I couldn’t resist asking DeepSeek: “Are you satisfied with this trade?”
After AI reviewed the data, it responded with surprising self-criticism: “No. A 1.8% return per day underperforms SOL’s typical 5–10% daily volatility — even if it was recommended by me.” And DeepSeek suggested the following change: “To optimize, increase Take Profit to 5–7%, set Trailing Take Profit at 3%, and focus on mid-term momentum.”
Naturally, I wanted to test these adjustments, so I ran a backtest using historical SOL price action for the same period the bot traded. The revised strategy included:
- 5% Take Profit with a 3% trailing mechanism
- 8% Stop Loss
- 4 extra buy orders at 2% price intervals
This backtest scenario demonstrates how strategy optimization would have played out — the bot would still have entered at $170.53, allocated $20 and placed 2 extra orders at $167.25 and $163.83 during the dips. Then, 3 days later, a take profit level of 5% would have been triggered when SOL’s value rose to $180.00. At this point the trailing mechanism would have been triggered and the bot would have sold the position at $174.00 (trailing 3%).
As a result, the same amount of coins would have been bought: 0.117 SOL + 0.119 SOL + 0.122 SOL = 0.358 SOL, which would have been sold at $174.00, so the total profit would have been 0.358 SOL × $174.00 ≈ $62.292. with the profit for 3 trading days would be $2,292.
Given the initial investment of $60 and the profit for 3 days of $2,292, total return for 3 days:
(2,292/60) × 100 = 3,820%
And daily profit, assuming equal daily portfolio returns would be:
3,820% ÷ 3 days ≈ 1,273.33% per day
Ironically, this updated approach offered by DeepSeek would demonstrate weaker performance and produce a lower daily return than the original setup.
The takeaway: AI’s strengths… and blind spots
The second experiment is over, and it is time to stop, analyze and reflect on whether or not using DeepSeek in setting the bot can be a way to go.
A key issue that emerged from this experiment was DeepSeek’s inability to account for critical market factors that influenced SOL’s downward trend after I launched the bot. Specifically, it overlooked events like upcoming Solana’s March 1 token unlock, which raised fears of oversupply, and the network’s worsening reputation due to its ties to high-profile scams. This occurred because, at the time of prompting, DeepSeek experienced issues with its Search functionality and consequently analyzed data only up to January 2025.
On the other hand, DeepSeek was objective in assessing the outcome of its chosen trading scenario. However, when attempting to optimize the strategy, it failed again which can also be the results of not having fresh market data to be able to backtest the proposed new trading settings. So it looks like that until the search functionality is fixed and DeepSeek is able to access and process up to date data you either have to feed it all the necessary data or doublecheck its choices using your market knowledge.
Anyway, let’s keep experimenting with AI and automated trading. Stay tuned!
FAQ
Can DeepSeek AI choose profitable trading strategies on TradeSanta without human input?
In this experiment, the bot delivered a 2.2% profit, but DeepSeek failed to analyze current market conditions due to a broken Search feature. Until that’s fixed, you’ll need to supply data or double-check its picks.4o