AI trading: step-by-step guides, bot setup, pros and cons

Most crypto users are willing to let AI agents manage a slice of their portfolio, according to a CoinGecko study.

The company surveyed 2,632 market participants and found that 87% would let artificial intelligence manage at least 10% of their holdings. One in seven would hand over the entire pot.

Which pill would you take?

1. Harness AI’s advantages and entrust your capital to the machine.
2. Shun the risk of a thoughtless mechanism draining your funds and stick to conservative, “manual” strategies.

If it is the first, ForkLog must warn you:

All actions are performed by professionals; do not attempt.

How to turn ChatGPT into a trading adviser

The market moves too fast. A dopamine-chasing generation does not read analysis or long articles and does not wait for expert takes. The era of Telegram news channels and TikTok has taught people to decide on the headline alone—whose content, more often than not, answers the only question that seems to matter: long or short.

Modern tech blasts dozens of headlines at consumers within seconds. Donald Trump speaks in America at 16:01—by 16:02 half the crypto traders in the CIS know about it. Fake news slips through, as do stories that are hard to read as a clear buy or sell signal.

ChatGPT can help judge the likely price impact of a headline—if you prompt it properly.

Step 1. Gather information

First, get the news. You can use:

• news websites such as ForkLog,
• social networks like X (a trove of crypto headlines; much of crypto starts here), Telegram and others. On X, try hashtags #Bitcoin, #Ethereum, #CryptoNews or more specific tags by topic;
• news aggregators such as Google News and Feedly.

Suppose you see a headline saying the Zora team was suspected of selling tokens ahead of the airdrop. What might that do to the token’s price? Ask ChatGPT.

Step 2. Open ChatGPT

You can use any AI chatbot you like—Grok, DeepSeek, Claude. We will use ChatGPT for illustration.

Open the assistant in the web version or the app and enter a suitable prompt.

You are a financial expert who has traded on the cryptocurrency market for many years. You read that the Zora team was suspected of selling tokens ahead of the airdrop. It is claimed they sent almost the entire drop to their own wallets. Some of these addresses were linked to the Bybit and KuCoin exchanges, raising suspicions of corruption in the allocation of assets. Wallets holding more than 100 million ZORA were also observed. After analysing this information, as well as any other data related to the project, would you buy or sell the ZORA token? Explain why.

ChatGPT’s answer was unambiguous: sell.

To diversify AI views, try the same prompt with another chatbot and compare the results.

Step 3. Ask follow-ups

If the first answer is not enough, dig deeper. Follow-up prompts can yield more specifics.

What if we wait for ZORA to drop and buy the bottom? What price levels make sense for accumulation?

For greater confidence, analyse similar cases. Handily, you do not need to scour the internet—ChatGPT will help. Prompt it as follows:

Do you know similar cases involving scandals and a large share of tokens held by a small group? What happened to those projects?

ChatGPT cited:

• Internet Computer (ICP) — heavy insider selling after listing drove the price from $700 to $3. Token concentration wrecked market reputation for good;
• SushiSwap (SUSHI) — the project’s creator, Chef Nomi, sold most of the tokens right after launch, causing an outcry. Under pressure he returned the funds, and governance passed to external developers. SUSHI recovered six months after the crash and saw a few rallies, but never matched Uniswap’s success;
• LooksRare (LOOKS) — a huge share of tokens went to insiders via staking pools. There was persistent selling into the market. The price fell from $7 to $0.1. The project still runs, but has lost almost all its market cap;
• Aptos (APT) — before launch many complained about extreme token centralisation among VCs and the team, but gradual unlocks and ecosystem development kept the project afloat.

News divorced from market context may not deliver good results. Market regime matters. Ask ChatGPT about that, too:

Given the ZORA situation and the fact that bitcoin is in an uptrend, how would you trade it? Answer briefly.

Sam Altman’s chatbot stays categorical—avoid ZORA or even short it. Better to focus on other, more reliable assets.

Step 4. Practice

Armed with the information, you can trade. Once again: AI is no panacea. It can be wrong, so practise with small sums. Over time refine your prompts and feed in more context. For example:

The ZORA team denied concentrating tokens in its own hands and provided evidence. Should I buy the token?

It may seem obvious to buy right after such news, but the AI remains more measured.

One more disclaimer

Tips

When trading with AI, keep the basics in mind:

• be specific — vague prompts like “what is a good trade” rarely work. State the headline and assets of interest clearly;
• double-check — verify ChatGPT’s output against other sources;
• speed — crypto moves extremely fast; supply up-to-date inputs. To get crypto news quickly, subscribe to ForkLog’s Telegram channel — https://t.me/forklog;
• risk — do not size up trades aggressively on AI’s say-so;
• stop-losses — they help protect capital from sharp drawdowns and cap potential losses.

Grok: more than just signals

Grok can do more than interpret crypto headlines. Its analytical and automation chops are stronger. Though not built for trading, its ability to evaluate data, spot patterns and read trends tempts traders to test it for automated strategies.

The premise is simple: Grok makes decisions based on available data, stripping out human emotion and bias.

Does it work? Some report impressive results; others find the notion of entrusting money to AI odd.

Advantages of Grok:

• identifying market trends — crypto markets are driven by emotion, FOMO and FUD. Grok can scan social media, headlines and community chatter to gauge shifting sentiment, a key driver of volatility;
• spotting hidden patterns — machine-learning capabilities let Grok detect subtle correlations that traditional bots may miss. For example, the AI can link improving social sentiment with whale activity and anticipate a bullish impulse;
• flexibility — rather than follow static rules like “buy when RSI drops below 30”, Grok supports more nuanced strategies.

How to build a trading bot

Grok is not a typical crypto-trading bot. It cannot place orders or integrate with exchanges by itself. But it can help you build a smarter, faster and more flexible system—from drafting trading logic to parsing market sentiment.

In one example a user, instead of hand-coding each script, asked Grok to create buy/sell logic that accounts for slippage, take-profits and priority gas fees.

The generated script can be wired into DeFi tools or extended via API 0x and Uniswap.

In another case a user built an automated system to trigger trades when strict conditions are met. Grok not only buys and sells; it also writes portfolio rebalancing scripts and can simulate position behaviour under different volatility regimes.

Useful services

Because Grok 3 does not connect directly to crypto exchanges, it needs integration with third-party platforms that support API automation:

• 3Commas — a solid option for executing trades via automated strategies;
• TradingView — use it to generate trading signals via Pine Script;
• CryptoHopper — offers bespoke strategy-building tools with API integration;
• Zapier or Make.com can connect Grok’s analysis to trading platforms;
• technically adept traders can process the bot’s ideas via Python scripts that place trades based on AI recommendations;
• services like IFTTT can trigger basic trade actions from Grok 3’s sentiment analysis.

Trading strategy

Success depends on a sound strategy. Unlike traditional bots that rely solely on technical signals, a Grok 3–based bot can blend several factors:

• technical indicators: RSI, MACD, Bollinger Bands and so on;
• sentiment analysis: social trends, influencer views and news headlines;
• on-chain data: whale activity, exchange inflows/outflows and large wallet moves.

Testing

Before going live, run backtesting to gauge effectiveness. It helps you evaluate:

• trade accuracy: how often Grok 3’s calls lead to profitable outcomes;
• trade frequency: ensure the bot does not overtrade into losses;
• ways to improve the strategy.

Risk control

Grok 3 is not an infallible manager. Implement:

• stop-losses that close a trade after an excessive drawdown;
• position limits to cap losses;
• trailing stops to lock in gains during uptrends.

Limitations of Grok 3

Grok 3 has strengths and weaknesses. Having covered the former, note the latter:

• latency — crypto moves at lightning speed; Grok 3 can lag and produce stale, poor trades;
• forgetfulness — the chatbot will eventually forget prior sessions. A nightmare for crypto traders;
• bias — Grok 3 can lean on incomplete or distorted sources.

Step-by-step guide

At ForkLog’s request Grok prepared a detailed step-by-step guide (scroll down for the Russian version) to building and running a high-frequency trading bot on the Solana blockchain, aimed at minute-by-minute moves in the SOL/USDC pair. The bot uses xAI’s Grok 3 API to decide whether to buy or sell based on market analysis, while trading executes via Raydium, a decentralised exchange. The guide assumes basic programming skills.

The bot will:

• trade SOL/USDC on Raydium, profiting from small price movements within one-minute candles;
• use Grok 3 to generate buy/sell orders from market analysis;
• apply risk controls: $100 position size, $20 max loss per trade, $50 target profit, up to 10 trades;
• execute trades automatically on Solana.

Before you begin, make sure you have:

• a computer with stable internet (Linux, macOS or Windows with WSL recommended);
• basic knowledge of JavaScript/TypeScript, the Solana blockchain and working with APIs;
• funds: at least 1 SOL for fees and USDC trading capital ($1,000);
• a text editor: Visual Studio Code (VS Code) or similar.

Other AI bots

Crypto trading bots are programs that buy and sell automatically based on machine-learning algorithms rather than fixed rules. They can digest large volumes of historical and real-time inputs—price changes, order-book depth, volatility, social sentiment—to spot opportunities.

AI bots can adapt to conditions dynamically—for instance, increasing position size when confidence in an outcome rises.

A list of notable bots:

• Pionex and Bitsgap — good for dollar-cost averaging (DCA);
• Freqtrade and Jesse AI — suited to building predictive models in Python;
• Stoic by Cindicator — uses in-house research to automate portfolio rebalancing;
• Cryptohopper and Kryll — convenient for users who do not code.

How to set them up? Step-by-step:

1. Choose a platform with AI support that fits your needs. For example, Freqtrade, Traality and Jesse AI let you import machine-learning models. 3Commas, Pionex and Cryptohopper focus on user-friendly automation and visual strategy builders.
2. Connect the bot to an exchange via API. Configure security: withdrawals disabled, two-factor authentication, and IP allowlists.
3. Set your strategy — define pairs to trade, order size, stop-loss and take-profit rules, and maximum position exposure.
4. Backtest on historical data. 3Commas, Cryptohopper and Freqtrade support such tests.
5. Go live with a small amount of capital, monitor results and scale gradually.

Common mistakes

Some pitfalls can lead to losses. Watch out for:

• failing to test with a small account. Some bots look profitable on paper but lose money in the wild;
• weak risk control. Skipping stop-losses or sizing too large can wipe you out. Bots like Freqtrade and Traality let you set precise risk limits—use them;
• ignoring trading frictions. Tests often omit slippage and fees, which can dominate in high-frequency trading. Jesse AI and Freqtrade offer tools to better simulate these costs.
• no monitoring. A well-tuned bot is not a “money” button. It needs supervision and tweaks. 3Commas and Traality support real-time alerts;
• excessive leverage — it can lead to liquidation.
• the wrong strategy — DCA can work on falling markets; breakout bots often do not. Platforms like Stoic and Kryll provide filters or pause triggers to avoid misfires.

Trading smart

Platforms like Freqtrade, combined with cloud tools such as Google Vertex AI or AWS SageMaker, make it possible to build new systems using reinforcement learning and online model retraining to keep pace with shifting market dynamics.

Integrating large language models into trading workflows lets bots interpret unstructured information—central-bank releases, tokenomics updates, SEC filings or even Discord announcements—and turn it into actionable ideas.

But there is always a human behind any trading AI. Choosing the wrong strategy, taking too much risk, AI mistakes or plain bad luck can all cost money. Risk management remains the cornerstone of any approach. Do not neglect it, especially in a young, experimental niche like AI-assisted trading.