Disclosure: this guide contains affiliate links to TradingView. If you sign up through them we may earn a commission, at no cost to you. Learn more. This is education, not financial advice.
"AI trading" means five very different things
An advert says "AI trading bot". A friend says "I use AI for trading". They probably mean completely different things. Self-driving cars have levels of autonomy, so we made the same ladder for trading. The higher you climb, the less work you do, and the more can go wrong without you noticing.
- Level0
You do everything
Your charts, your analysis, your clicks. The classic way.
- Level1
AI as your analyst
You ask an AI like TradingView's AI Copilot or Claude to summarise a chart, explain an indicator or challenge your idea. You still decide and click. Best for learning.
- Level2
Rules you wrote, alerts you act on
You turn your idea into clear rules (often with AI's help writing the code) and get an alert when they trigger. You check it, then place the trade yourself.
- Level3
Rules that place orders for you
Classic auto-trading. Your fixed rules send orders to a broker automatically. The rules don't think; they just follow instructions, including bad ones, at 3am.
- Level4
An AI agent decides and trades
A language model reads the news and prices, decides what to do and places orders itself. New, exciting and, as the next section shows, very unpredictable.
Our view: Levels 1 and 2 are where AI helps most people today. Level 3 is fine once you've proven your rules the boring way. Level 4 is for experiments with money you are genuinely happy to lose.
What happened when six AIs traded real money
In autumn 2025, a research project called Alpha Arena gave six leading AI models $10,000 each in real money. They all got the same instructions and the same data, and traded crypto on their own for about two weeks (18 October to 3 November).
| Model | Final balance | Result |
|---|---|---|
| Qwen 3 Max | $12,231 | +22% |
| DeepSeek | $10,489 | +5% |
| Claude Sonnet 4.5 | $5,799 | −42% |
| Gemini 2.5 Pro | $5,445 | −46% |
| Grok 4 | $4,208 | −58% |
| GPT-5 | $4,126 | −59% |
Four of the six lost money, most of them heavily. (Yes, that includes Claude, which we otherwise rate highly. Being brilliant at writing and coding doesn't make a model a good trader.) Two weeks is far too short to call anyone a winner, so treat the top spot as luck until proven otherwise.
The lesson: the models had identical information and still behaved completely differently. Some traded often and took big leveraged bets. Level 4 isn't "set and forget". It's "set and hope".
Sources: ForkLog · iWeaver season 1 results · Alpha Arena
The maths the adverts skip
Bot sellers love monthly numbers. "Just 10% a month!" sounds modest. Compounding says otherwise. Move the sliders.
🧮 Too good to be true?
What a "monthly return" claim would really mean.
Next, the number every bot advert shows off: the win rate. It means almost nothing on its own. A bot can win 80% of trades and still lose money if the losses are bigger than the wins. And fees and slippage (getting a slightly worse price than planned) quietly nibble every single trade.
🎯 Does this strategy actually make money?
Enter numbers from a backtest or a bot's sales page.
The default example wins 70% of the time and still loses money. That's not a trick; it's arithmetic. Berkshire Hathaway benchmark: 2024 shareholder letter (19.9% a year, 1965 to 2024, versus 10.4% for the S&P 500).
Why most great backtests are lying to you
A backtest replays your rules on past prices. It's essential. It's also the easiest place to fool yourself, and AI makes that easier, because it can try thousands of variations in minutes.
Researchers David Bailey and Marcos López de Prado showed why: test enough versions of a strategy and one will look amazing by pure chance. Pick that "best" version and it usually disappoints on new data. They call it backtest overfitting. In plain English: you taught your strategy the answers to last year's exam.
The backtest honesty checklist
How auto-trading with TradingView actually works
TradingView is where many people build and test their rules (Levels 2 and 3). It doesn't run a bot for you by itself. Instead, you chain a few pieces together:
- Write the rulesDescribe your idea to TradingView's AI Copilot (or Claude) and let it draft a Pine Script strategy. Read it; ask the Copilot to explain every line.
- Backtest honestlyRun it in the Strategy Tester with fees and slippage switched on. Use the checklist above.
- Paper tradeTrade it with pretend money in TradingView's paper trading for a few weeks. Live prices are the real test.
- Add an alert with a webhookAn alert can send a message to a web address when your rules trigger. Webhook alerts need a paid plan (Essential and up, at the time of writing).
- Connect a relay to your brokerA third-party service receives the webhook and places the order with your broker. Choose carefully: it can touch your money.
- Start tiny, with a kill switchUse the smallest position size, set a daily loss limit, and know exactly how to switch it all off.
Sources: TradingView: webhook alerts · TV Hub: plan limits
Prompts that make AI your sparring partner
The best use of AI in trading isn't finding trades. It's finding the holes in your ideas before the market does. Try these in TradingView's AI Copilot or Claude.
I'm thinking of buying this because [your reason]. Make the strongest possible case AGAINST this trade. What would have to happen for me to be wrong, and where on this chart would that show up first?
Here are my strategy's backtest results and settings: [paste]. List every reason these results might not hold up on new data. Check for repainting, too few trades, missing fees and settings that look suspiciously precise.
Here are my last 20 trades with entry, exit and my reason for each: [paste]. Find patterns in my mistakes. Am I cutting winners early, holding losers, or trading more after a loss? Be honest, not kind.
Add risk controls to this Pine Script strategy: a maximum daily loss, a maximum number of trades per day, and no new trades in the last 15 minutes of the session. Explain each change in plain English.
Automation doesn't change the odds
Studies of real traders are sobering. In Brazil, researchers followed everyone who started day trading futures between 2013 and 2015. Of those who kept going for more than 300 days, 97% lost money, and only 1.1% earned more than the minimum wage. In Taiwan, a long-running study found fewer than 1% of day traders were predictably profitable after fees.
A bot running a losing idea just loses faster, and never gets tired. AI is excellent at making you a better-informed, more disciplined trader. It is not a shortcut to beating people who do this full time with better data and faster machines.
Sources: Chague, De-Losso & Giovannetti, "Day Trading for a Living?" · Barber, Lee, Liu, Odean & Zhang
How to spot an AI trading scam
The US Commodity Futures Trading Commission has a blunt warning titled "AI Won't Turn Trading Bots into Money Machines". It describes scams promising returns of "tens of thousands of percent" or 100% win rates. Walk away if you see:
Guaranteed returns
Nobody can guarantee market returns. Not a human, not an AI.
"Never loses"
A 100% win rate means cherry-picked results, or worse.
Pressure to hurry
"Only 3 spots left!" is a sales tactic, not a market signal.
They hold your money
Deposits sent to them, or keys that can withdraw. Your money should stay in your own account.
DMs from strangers
Influencers and "mentors" who contact you first. Check their background with your regulator.
A secret algorithm
"Proprietary AI" with no testable rules and no independent record.
Pros & cons of trading with AI
👍 Where AI genuinely helps
- Learning technical analysis much faster
- Writing and explaining Pine Script without coding skills
- Testing ideas on years of data in minutes
- Removing emotion from following your rules
- Watching markets while you sleep (with alerts)
- A tireless critic of your own trade ideas
👎 Where it hurts
- Makes overfitting effortless
- Confident answers that can be wrong
- Can't predict news, shocks or crashes
- Bots repeat mistakes at machine speed
- Fees and subscriptions add up
- A magnet for scams using "AI" as bait
A 30-day plan to try AI trading safely
- Week 1: learnUse TradingView's AI Copilot or Claude as a teacher. Ask it to explain every indicator on your chart. No trading.
- Week 2: write one simple ruleOne idea, in one sentence. Have AI turn it into a strategy, then backtest it with fees on.
- Week 3: attack itUse the "Overfitting detective" prompt and the checklist. Most ideas die here, and that's a win: they died for free.
- Week 4: paper tradeFollow the surviving rule with pretend money and keep a journal. Only then think about small, real positions.
Frequently asked questions
Can AI predict the stock market?
No. It can spot patterns in past data and summarise information quickly, but prices react to surprises that nobody, human or machine, can reliably predict. Be very sceptical of anyone who says otherwise.
Is auto-trading legal?
Automating your own trades through your own broker account is generally legal, subject to your broker's rules. Selling signals or managing other people's money usually needs a licence. Check the rules in your country.
Can TradingView trade for me automatically?
Not on its own. TradingView handles charts, strategies, backtests and alerts. To place orders automatically, alerts are sent by webhook to a third-party service that talks to your broker. You can also trade manually through brokers connected to TradingView.
How much money do I need to start?
For learning, none: the free plan and paper trading cost nothing. With real money, start with an amount you could lose completely without it affecting your life.
Should I let an AI agent trade my account?
Not with money that matters. The Alpha Arena results show how differently, and how badly, top models can behave with real money. If you're curious, experiment with a tiny amount and strict limits.
Use AI to think better, not to stop thinking
AI is the best trading teacher and sparring partner ever made. Use it to learn faster, write and test clear rules, and find the flaws in your ideas. Be very wary of anything promising to do the thinking, and the winning, for you.
I asked my crystal ball to predict the market. It said "reply hazy, try backtesting". Wisest thing it's ever told me. I now trade only in treats, with a strict daily loss limit.
Practise with TradingView's AI Copilot
Free charts, paper trading and a Strategy Tester to put this guide into practice.
Try TradingView freeAlso read: Our TradingView review · TradingView AI Copilot guide · Claude AI review · The one-person AI studio · Does your business need an AI chatbot?