Can AI trade crypto for you? Fahad gave six models the same task with one rule: no human interaction, not even for a single trade.
GPT lost over 62%, Grok about 45% and Claude about 30%, while only Qwen (about 22%) and DeepSeek (about 5%) finished green. The video then takes that result apart with a use simulator, a Tesla memory test, AI scam cases and a Pine Script indicator that went from a 14% win rate to 57%.
Fahad's summary: 'AI is not the brain. AI is basically your hands.'
Key Takeaways
- Remove the human and most models lost money, while the two winners made gains a patient trader could match without any AI.
- Automate the repetitive stages like level marking, entry triggers and trend accumulation, but write the rules yourself first.
- Same trades, different use: discipline at 2x made money while 15x lost, so size down before you chase bigger multiples.
- AI recall can look like insight, and fraudsters borrow that look, so if you cannot explain the logic behind a promise, skip it.
- AI can draft the indicator in minutes, but the setting changes that took the win rate from 14% to 57% came from the trader.
GPT lost over 62% while DeepSeek made about 5%
Six models, ChatGPT, Gemini, Claude, Grok, DeepSeek and Qwen, traded crypto under one rule: no human interaction, not even for a single trade. GPT lost over 62%, Grok about 45% and Claude about 30%. Qwen gained about 22% and DeepSeek about 5%.
Fahad opens with the question most viewers are already asking, will AI replace the trader, then flips it: 'Do you need AI to do this? Because you can do it yourself too.'
Losing 50% or 60% does not take a machine, and any retail account with a bit of use manages it unaided. Three of the five reported results are losses, and the two winners are modest. DeepSeek's roughly 5% is a number a patient human can produce.
The experiment is not a case for handing your fund to a model, because nothing replaced the judgment that was removed.
Small print: the video does not cover Gemini's figure, and the starting capital is not clearly stated, so the table shows percentages only.
Six AI models, same crypto task, no human interaction
| AI model | Result |
|---|---|
| GPT | Loss of over 62% |
| Grok | Loss of about 45% |
| Claude | Loss of about 30% |
| Qwen | Gain of about 22% |
| DeepSeek | Gain of about 5% |
Levels and entries can be automated, but the rules stay yours
Marking support and resistance and waiting for entry triggers are mechanical jobs, so a bot can take them over. Fahad's Hit Point Levels indicator auto-marks time-based and volume or liquidity-based levels, and Stakar accumulates in aggressive uptrends and sells dumps. The rules still come from the trader.
Fahad splits the day into three stages:
- Marking levels: Hit Point Levels draws the highest-probability zones using time-based and volume or liquidity-based methods. No more hours at the chart.
- Waiting for entries: a buy or sell trigger fires on the right confirmation, so you are not staring at price.
- Riding the trend: Stakar accumulates in an aggressive uptrend and sells in a dump, aiming at the bigger moves.
His thesis sits on top of all that: 'AI is not the brain. AI is basically your hands. Your experience, you are the brain.' Hands do what the brain defines.
AI assistants also cannot place trades or enforce stops by themselves. They need an exchange API or a bot platform to execute, as Bitsgap's breakdown of ChatGPT, Claude and Gemini for crypto trading explains.
Fahad does not oversell his own setup either. 'Honestly, nothing here is 100% assured.' For more on where AI helps and where it does not, see our overview of AI in trading strategies.
A 2x disciplined trader beat a 15x trader over 300 trades
In the use simulator, a disciplined trader taking three trades a day at 2x finished above +10%. A use-only trader at 15x, same three trades a day, finished about 3% down on the same 300 previous trades. More use did not mean more profit.
Fahad was direct about the belief behind this. 'The concept in your mind that higher use will make more profit, that's actually not true.' Exchanges offer up to 1,000x or 200x, and traders routinely pick the maximum.
Use simulator: discipline versus use only
| Trader type | Use | Trades per day | Result |
|---|---|---|---|
| Disciplined trader | 2x | 3 | Over +10% |
| Use-only trader | 15x | 3 | About -3% |
| AI bot in the simulator | Not stated | Not stated | $100 start, about $91 to $92 end |
The sample is 300 previous trades, and Fahad notes the numbers would shift slightly on a new run. The gap between the two rows is what matters, not the exact decimal.
In the same simulator the AI bots sat deep in red, ending around $91 to $92 from a $100 start.
Fahad's line on it: 'Use is one of the major reasons you blow your account.'
If you want a worked example of surviving the sizing question, our guide to passing a funded account challenge without blowing it covers the same discipline from the risk side.
AI answers from memory, and scammers sell that confidence
AI answers correctly when it has a stored answer and gives a confident wrong one when it does not. ChatGPT said yes to buying Tesla at about $200 in January 2020, then said no once its memory was erased. Scammers exploit the same borrowed authority with deepfakes and AI-branded schemes.
Tesla trades above $400 now, so the yes was right and the later no was wrong. Fahad's point is that the first reply was recall, not insight: 'If AI doesn't have a stored answer, it'll just give you a wrong answer with high confidence.'
A chart screenshot with a request for a buy call, stop or target mostly returns a description of trend structure, because AI cannot know the future trade. Fahad's blunt version: 'Just tell me, the right answer is no.'
On the fraud side Fahad sounds frustrated: 'Scammers' favourite word is AI.' The cases he covers:
- A 79-year-old Bengaluru woman lost about Rs 35,00,000 to a deepfake Narayana Murthy video.
- An IAS officer lost Rs 3,37,00,000 to an AI-branded investment scheme.
- A Pune businessman lost Rs 1,00,00,000 after a deepfake of a finance minister.
His filter for any claim of 99% or 80% accuracy: 'Just walk away until you can understand the logic. If you don't understand that, just walk away from it.'
A sensible habit is to test any automated claim on historical data first, as CoinQuant's guide to building and backtesting a ChatGPT crypto strategy describes.
Same code, better settings: a 14% win rate became 57%
A Pine Script V6 indicator that ChatGPT built from Fahad's rules in about 5 minutes showed a 14% win rate at first. Tuning only the settings, with no change to the code or logic, lifted it to 57%. The human still owns the rules and the tuning.
The workflow is four steps: idea, rules, indicator, then a webhook as the bridge from signal to broker. Fahad supplied the rules. Rule one uses volume delta from a lower timeframe for each bar, with a bearish criterion and its bullish opposite.
He pasted the output into TradingView through the Pine Editor. A free version of any AI is enough for this part.
Then came the tuning, and Fahad could hardly hide his excitement: 'The win rate is literally at 57% now. I did not change anything in the code.'
His takeaway: 'Now basically it's on you. When you change some settings with your brain... That's exactly when you get to see the best kind of results.'
It is not a clean win story. One indicator still failed even with the rules Fahad gave it: 'It couldn't make any money at all.'
We covered the debugging side of this in our walkthrough of building a TradingView indicator with AI, and a 30 day test of AI on Bitcoin futures shows how long-run results compare.
The full AI to webhook to broker walkthrough comes in a follow-up if enough viewers comment 'auto pilot'.
Who writes the rules decides the result. Six models with no human input mostly lost, a 15x trader lost to a 2x trader on the same trades, and ChatGPT's Tesla answer flipped when its memory was erased. When Fahad tuned settings on a rough indicator, the win rate moved from 14% to 57% with the code untouched. Use AI for the mechanical work and keep your own judgment on the rules and the size. Fahad's own caveat applies to his setup too: nothing here is 100% assured. Join the Hitpoint community for live market coverage and real-time trade discussions.
Watch the full video: 6 AIs Were Given $10,000 to Trade... Here's What Happened on YouTube
Disclaimer: This article is for educational purposes only and is not investment advice. Crypto assets and virtual digital assets are unregulated in India, are highly volatile, and can lose value rapidly; there is no established regulatory recourse for losses from them. The price levels, setups and trades described are a review of market events that have already occurred, shared to illustrate a method of analysis rather than to recommend any trade. Trading With Sidhant LLP is not a SEBI-registered Investment Adviser or Research Analyst, and nothing here constitutes a recommendation to buy or sell any asset. Income from virtual digital assets is taxable in India at 30% plus applicable cess, with 1% TDS on transfers. Past performance is not indicative of future results. Do your own research before trading.
Frequently Asked Questions
- Can AI trade crypto for you without any human input?
- In the video's experiment, six AI models traded crypto with no human interaction and most lost money. GPT lost over 62%, Grok about 45% and Claude about 30%, while Qwen gained about 22% and DeepSeek about 5%. Fahad's conclusion is that AI works as the hands, but the trader has to supply the rules and judgment.
- Does higher use mean higher profit in crypto trading?
- No. In Fahad's simulator, a disciplined trader at 2x use finished above +10% over 300 previous trades, while a use-only trader at 15x finished about 3% down. Exchanges offer up to 1,000x or 200x, but extra use mostly adds risk. Fahad calls it one of the major reasons accounts blow up.
- Can ChatGPT give a reliable buy or sell call on a stock?
- No. In the Tesla test, ChatGPT said yes to a buy in January 2020 when Tesla was around $200, then said no after its memory was erased, even though Tesla now trades above $400. Fahad's point is that AI gives confident wrong answers when it has no stored answer, and a chart screenshot only gets a trend description.
- How do AI investment scams work in India?
- Scammers use the AI label and deepfake videos of trusted public figures to make fake promises look credible. The video cites a 79-year-old Bengaluru woman who lost about Rs 35,00,000 to a fake Narayana Murthy video, an IAS officer who lost Rs 3,37,00,000, and a Pune businessman who lost Rs 1,00,00,000.
- Can you build a TradingView indicator with a free AI tool?
- Yes. Fahad asked ChatGPT for a Pine Script V6 indicator, supplied his trading rules, and had working code in about 5 minutes, then pasted it into TradingView through the Pine Editor. He says a free version of any AI is enough. The first result showed a 14% win rate, and tuning only the settings lifted it to 57%.
Share this post
Why Bitcoin Open Interest Falls While Price Rises
Bitcoin pushed toward 63,700 while open interest fell, and Fahad kept his short valid. Here is the order flow read behind it.
20 Nov 2026
How to Trade Trapped Buyers in Bitcoin: Wait for the Close
Fahad saw buyers trapped at the top of Bitcoin, yet held off shorting until the candle closed. Here is why that wait matters.
19 Nov 2026
What to Trade When Bitcoin Is Stuck in a Range
Bitcoin printed four days of sideways candles since 14 November, so Fahad rotated to gold, silver and oil.
18 Nov 2026
Join our Telegram community.
Get post alerts, trading discussions, and market updates.
Get post alerts in your inbox.
New blog post? We'll email you. No spam, unsubscribe anytime.