Do AI Trading Bots Actually Work? The First Week, Month and Year

Jay Sharma · Founder, Botsfolio

Published October 19, 2026

"Do AI trading bots work?" is the most common question people ask before they let one near their money. In Google's suggestions alone it shows up in hundreds of forms. Most answers come from companies selling a bot, and they show a good month or a backtest with no bad days in it.

The honest answer depends on when you look. So instead of one number, here is how often an AI portfolio was up after a week, a month, three months, a year and two years, from 94 different starting dates, with every bad start left in.

The short answer

Yes, an AI can trade crypto profitably, but not every week or every month, and most bots sold online have no record that would show it either way. In our test, the AI was up about half the time after a week or a month, which is no better than holding. The difference showed up over a year or more, and mostly in how much less it lost in bad years.

Two things decide whether a bot "works" for you:

• The time you give it. One week or one month tells you almost nothing about any trading system, good or bad.
• What you compare it with. Making money in a rising market proves little. The test is whether it did better than simply holding the same coins, after fees, including the worst stretch.

How often the AI was up: week 1 to year 2

For every monthly start from January 2018 to October 2025, we measured where Botsfolio's BTC & ETH Cycle Trader stood after each stretch of time, next to holding BTC and ETH with the same money:

In the first month, the AI and holding were up equally often. From three months on, the AI was up more often, and the gap grew with time. Not one of the 82 two-year windows ended below its starting money. Holding finished below it in 30% of them.

The typical result, and the worst

How often a portfolio is up is half the story. The other half is how far it can fall:

The typical results are close: over one or two years, holding and the AI made similar amounts at the median. Over any single year, the AI finished ahead of holding in 57% of starts. Where it differed was the bad starts. Its worst year lost 45%; holding's worst year lost 77%. After two years, its worst start was still up 3.3% while holding's worst was down 65%.

That pattern is what a working AI portfolio looks like: not a winner every month, but smaller losses when the market falls, which adds up over time. Held from each start to October 2026, it finished ahead of holding from all 94, partly because the test ends in a falling year for both coins.

Why the first month tells you almost nothing

If you start an AI portfolio and it is red after a week, that is normal. In our test, 43% of starts were below their starting money after one week, and 46% after one month. The same was true for holding.

Short stretches are mostly noise. A week of crypto prices can swing 20% either way, and no trading rule can control that. What a rule can control is what happens across many weeks: how much it keeps when prices rise and how much it gives back when they fall.

A fair test of any AI trading bot takes six to twelve months, ideally including a falling market. That is also why paper trading first makes sense: you can watch the record build on live prices without any money at risk. Our guide on how long to paper trade covers what to look for, and paper trading vs real money covers the switch.

What lower risk changes

Botsfolio AI can also be told to keep falls small. That version runs as the BTC & ETH Steady Trader. Tested the same way:

At none of these checkpoints was it down more than about 6%, but it also made much less: a typical +10.9% after a year. A small-loss AI suits people who mainly want to stop losing. It doesn't suit people who want the most growth.

What it looks like live in 2026

Backtests show the range. The forward record on live prices shows what is happening now:

2026 has been a falling year for Bitcoin and Ethereum. As of October 19, 2026, the Cycle Trader is up about 24% for the year while holding the same coins is down about 14%, and its biggest drop was under 9% against about 47% for holding. In a fast rally, holding usually climbs faster. The month by month story is in buy and hold vs an AI-managed crypto portfolio.

Do bots beat people?

On Hyperliquid, a crypto exchange where every trade is public, we sorted 1,844 accounts that had traded at least $50,000 into humans, bots and accounts we couldn't classify, by how they traded. 42.9% of the bot accounts made money over their active period, against 31.4% of the 1,076 human traders. Bots did better, but most of them still lost, and 40% of them lost 90% or more of their largest balance at some point. Automation alone doesn't make a strategy work.

We also gave Botsfolio AI the same money and dates as each of the 1,076 human traders. Tested on their exact dates with spot coins and no leverage, it made money in 88.6% of the windows. The traders used real money, mostly with leverage. The full comparison is in AI vs 1,076 human traders, and the trader side on its own is in how many crypto traders are profitable.

Why most bots sold online don't work

The bot is rarely the problem. The record behind it is:

• Promises instead of a record. The US CFTC lists guaranteed or very high returns as the first warning sign of an AI trading scam. A real record shows losing months.
• Backtests that peeked or skipped fees. A strategy tuned on the past, tested without fees, or one that used tomorrow's prices looks great until it trades live. We explain the five common causes in why trading bots fail live.
• A chatbot picking trades. In nof1's Alpha Arena, six AI models traded $10,000 of real money each and four of the six lost money. Our own test of a frontier model picking trades found no skill. Details in can ChatGPT trade crypto.
• Fees nobody counted. The same AI portfolio made 33.6% a year at a 0.10% fee and 17.1% at 0.90%. See how exchange fees change the result.

For the scam side, our guide to whether AI trading bots are legit covers the red flags.

How to tell if an AI trading bot works

Every Botsfolio AI portfolio shows these on its page: the live record next to holding, the biggest drop, and a backtest from every monthly start since 2018. Our guides to drawdown and the buy and hold comparison explain how to read them, and the portfolio methodology explains how they are tested.

How to try it without risk

• Watch first. Botsfolio AI trades live in public: every buy, sell and stop, next to holding the same coins.
• Paper trade. Build your own AI portfolio on the coins you pick and watch it trade on live prices for free.
• Real money when you're ready. Run the same AI on your own Coinbase, Kraken, Binance or Binance.US account. See which exchanges work in your country, how connecting works and pricing.

Method and limits

FAQ

Educational analysis, not financial advice. Backtests are tests on past prices and paper trading results are simulated. Trader data comes from Hyperliquid's public data. Trading crypto can lose money. Past performance does not predict future results.