Why Most Crypto Traders Lose Money (And How AI Can Help)

By the end of this article you’ll have a five-question pre-trade checklist you can paste into ChatGPT (or Claude, or Grok) before every position. It won’t make you profitable. Nothing will guarantee that. But it will force you to answer, in writing, the exact questions that 84% of first-year crypto traders skip – the ones that separate a trade from a bet.

That’s the destination. To understand why the checklist matters, we have to walk backwards through the data on why most crypto traders lose money – and it’s less about picking bad coins than you’d think.

The numbers are worse than the memes suggest

Every crypto guide quotes some version of “90% of traders lose money.” That number is folk wisdom – nobody can source it cleanly. The actual peer-reviewed research is grimmer.

Barber, Lee, Liu and Odean’s landmark Taiwan Stock Exchange study (most recently updated in 2023) concluded that less than 1% of day traders are able to predictably and reliably earn profits after fees. A 2019 paper by Chague, De-Losso and Giovannetti went further: 97% of all individuals who persisted for more than 300 days lost money in Brazilian equity futures.

Regulator data tells the same story. A 2020 FINRA report found that 72% of day traders ended the year in a financial deficit. India’s securities regulator SEBI has published two damning reports back to back – over 70% of individual intra-day equity traders lost money in FY 2022-23, alongside a 300%+ surge in intra-day participation vs FY 2018-19 – and 91% of individual traders posted net losses in FY25. Bull-market inflows don’t help the crowd; they systematically feed the losing side.

Crypto-specific data lands in the same range. A 2025 survey of 1,005 retail crypto traders found 84% lost money in their first year, 1 in 3 quit within six months, and 58% lost almost all of their capital in that window.

The hidden reason the stats are actually understated

Here’s something no tutorial mentions: the numbers above are probably too kind. As one Swiss analysis of the academic literature put it, most traders simply quit quietly. They don’t show up in surveys. They delete the app and pretend it never happened. The published “loss rate” is the loss rate among people who stuck around long enough to be counted.

Think about what that means. If 84% of visible first-year traders lose, and a third of the invisible ones exit before any survey can reach them, the real base rate is closer to a coin flip that lands on “you lose” almost every time. The survivorship problem doesn’t make the data wrong – it makes it optimistic.

What actually causes the losses (backed by survey data, not vibes)

Every article says “emotions” and stops there. The NFTEvening survey actually asked traders which specific mistake cost them the most. Two answers dominated: poor research (55%) and FOMO (44%). Those aren’t surprises. The quieter but more damaging finding: 58% of new traders don’t seek professional advice when they’re losing, relying instead on friends and social media. The single most predictive question for whether a trader will still have capital in 12 months isn’t “do you use stop losses” – it’s “who reviews your trades besides you.” If the answer is nobody, you’re a sample size of one arguing with yourself.

That last pattern is the real killer. It’s not the first loss that ruins portfolios – it’s the doubling down that follows, without anyone qualified in the loop.

The five-question AI checklist (the actual tutorial)

Here’s the workflow. Before every trade above, say, 2% of your portfolio, open ChatGPT and paste this prompt with your trade details filled in. The point is not that ChatGPT will predict the market – it can’t, and I’ll explain why below. The point is that answering these questions in writing flushes out the specific mistakes the survey data flagged.

You are a skeptical trading coach. I am about to enter this position:

Asset: [TICKER]
Direction: [long/short]
Entry: [$price]
Size: [% of portfolio]
Stop loss: [$price or "none"]
Take profit: [$price or "none"]
My thesis in one sentence: [reason]

Answer these five questions bluntly:
1. What specific piece of evidence in my thesis is weakest?
2. If this trade goes to zero, what % of my annual income do I lose?
3. Am I entering because of new information, or because the price moved?
4. What would have to be true for me to be wrong - and how would I know?
5. Have I taken a similar trade in the last 30 days? If I don't know, that's my answer.

Do not tell me if the trade is good or bad. Only surface what I haven't thought through.

Question 3 catches FOMO. Question 4 catches confirmation bias. Question 5 catches the revenge-trading loop that quietly drains accounts between the big blowups. None of these require the model to know the current price of anything – which is important, because it doesn’t.

The gotcha nobody mentions: what AI can’t do here

ChatGPT will confidently answer questions about tokens that no longer exist. It does not access live market data unless specifically integrated with external APIs, and its training data is months to over a year stale depending on the model. Ask it about a memecoin that launched last week and it will either refuse or hallucinate – and both are dangerous when you’re about to click Buy.

There’s a subtler failure mode too. Due to misinterpretation of prompts, the model can make errors in analysis and surface incorrect suggestions. Paste a chart image and ask “is this bullish” and you’ll usually get “yes” – because the model reads your framing as confirmation-seeking. This is why the checklist prompt is deliberately worded to make the model resist you.

What to use AI for, and what to keep manual

Common pitfalls to avoid with this workflow

• Skipping the checklist on “obvious” trades. The obvious ones are the FOMO ones. That’s the whole point.
• Rewriting your thesis after seeing the AI’s response. If you edit your one-sentence thesis to “win” the checklist, you’re back to arguing with yourself.
• Asking the model to pick the trade. Different prompt, different job. The checklist audits your decision – it doesn’t replace it.
• Trusting fundamental research on new tokens. If the token launched after the model’s training cutoff, the model is guessing. Cross-check with CoinGecko or the project’s actual on-chain data.

Does this actually change results?

Honest answer: there’s no controlled study on “AI pre-trade checklists reduce retail losses.” That study doesn’t exist yet. What the survey data does point at is a specific chain of failures – no research, FOMO entry, no external review, no exit plan – and the checklist directly interrupts three of those four. After running this on my own trades for a few months, the finding isn’t that it stops bad trades. It stops bad trades I hadn’t thought about. That’s a smaller claim, but it’s the true one.

When NOT to use this approach

High-frequency traders: skip this entirely – you don’t have time to paste a prompt between candles. Pure DCA into Bitcoin on a schedule? Also overkill; the whole point of DCA is removing the decision. Perp liquidations, MEV, arbitrage – the model’s latency alone disqualifies it for sub-second execution windows.

The checklist is built for discretionary swing traders taking a few positions a week. That’s the population where the loss data is worst – and where a friction step has the most room to help.

FAQ

Is the “90% of crypto traders lose” number real?

No verified source for it. The real peer-reviewed numbers – Barber-Odean (under 1% profitable long-term), Chague et al. 2019 (97% of persistent traders lost), NFTEvening 2025 (84% first-year loss rate) – are actually worse.

Can I just let an AI trading bot do this for me instead?

You can, but you’re solving a different problem. A bot removes emotion from execution – it doesn’t remove it from strategy design. If your underlying strategy came from a YouTube video you watched at 2am, automating it just lets you lose money faster and without noticing. The checklist targets the decision layer, which is where the survey data says most losses actually originate. Bots and checklists aren’t substitutes; they operate at different stages.

Which AI model works best for the checklist prompt?

Any frontier model handles the prompt fine – free-tier ChatGPT, Claude, and Grok all produce usable output. Where a paid tier starts earning its cost: you want to upload a CSV of your last 30 trades and ask the model to find patterns in your losing positions, or you need persistent memory so it can track your thesis across sessions without you re-explaining context every time. For the basic five-question checklist alone, the free tier is enough. Upgrade when the friction of re-explaining your history gets annoying enough to pay to remove it.

One last thing: the checklist only works if you fill it out before clicking Buy – not after, when you’re already rationalizing. If you get to question 4 (what would have to be true for you to be wrong) and draw a blank, that’s the trade to skip.