Its page promises a one-to-two risk-to-reward. In my tester, it risks $40 to make $16. And it won’t open a single trade before 2016, so I gave it 2012. This is the My AI Gold EA review I wanted to read before paying $799.95: what it does well, what the page doesn’t tell you, and four checks you can run on any AI EA.
Quick note first: I build EAs too, gold included, at DoItTrading. I have no relationship with the developer of My AI Gold. Everything below is a historical test of version 1.82, the one on sale when I recorded it, not a forecast. Not financial advice. If the developer thinks I got something wrong, write to me and I’ll correct it.
What you’re buying
The listing talks about “advanced neural network engineering” built with “machine learning frameworks from GitHub”. $799.95, and under the price: next ten copies only, then $999. It said that on October 10, and it said the same, at the same price, on October 21. It was published on September 4, and by October 19 it was number three among MetaTrader 5 experts on the Market, with thousands of demo downloads.
How it trades: at the start of every hour it looks at gold and decides. Every trade gets a stop loss $40 away on the gold price and a take profit $25 away. When a trade is $16 up, it locks that in and trails, so most winners close right there, at about +$16.
What it does well
- No grid, no martingale, no averaging. Every one of its 1,459 trades in my test had a stop loss.
- Consistent. It trades the same on 15 minutes, one hour or four hours: the same trades.
- Every year positive at a fixed lot, from 2019 to 2026, with a profit factor of 2.25.
- It survives the stress test. In the same engine and with the same $1,000 and lot I used for Quantum Queen X, Queen X lost the whole account in one history out of five. My AI Gold, in none.
The 1:2 ratio, the other way round
The page says “a solid one-to-two risk-to-reward ratio”, so “you do not need an unrealistic, sky-high winrate”. In my test, a full stop loss lost $40.22, and most winners closed at about $16. The tester report, counting every trade: an average win of $14.92 and an average loss of $20.44. It risks more than it makes.
To break even with that, it needs to win about 58% of its trades. It wins 75%. So it makes money, but because it wins three trades out of four, not because of the ratio. Its own live signal says the same: an average win of $113 and an average loss of $224.
The default risk
This is the part that can cost you real money. In the product’s group chat, the developer is clear: “Default settings are high risk.”
- By default it trades 0.01 lots for every $300 of balance. One full stop loss is 13% of the account.
- Two trades open at once is normal: more than a third of its entries. Two stops at once is 27%.
- That’s why its backtest curve looks the way it does. $10,000 on default settings, compounding, ends at $16 million, with a 54% drawdown on the way: minus 39% in 2020, minus 48% in 2021.
- The “+1,000%” track record is a MetaTrader 4 account, though the EA is sold for MetaTrader 5. It started with $300, and on October 10 it was at +1,604% after a 43.6% balance drawdown.
- The settings file the developer shared as the one used on the signals: one stop loss is 10% of the account.
Is My AI Gold really AI?
In the group chat, the developer explains it: classical logic, “trend structure and pullback conditions”, finds the setup, then “the neural module” scores it, and only setups above a threshold are traded.
That fits what I measured. Keep its exits and its hours and pick random entries: profit factor 1.04. Use a simple trend-and-pullback rule: 1.01. Its own picks: 2.22. All the edge is in which candles it picks. With The Gold Reaper, I found its exact rules; here, no simple rule and no mix of 400 classic conditions reproduces its entries. And when I shift the whole price history by a week, it repeats almost 99% of them, so it reads price, not the calendar. Probably a real model.
I tried to build one too: six designs trained on history, with the rules written before each test, and none beat the plain strategy on years it hadn’t seen. That’s why the next part matters.
The 2016 wall
A model trained on history knows the years it was trained on. The years that matter are the ones it hasn’t seen. My AI Gold’s backtest starts in 2016. I ran it from 2012, and it opens nothing until January 4, 2016.
Is that the data or a date? I moved real gold prices four years back: 62 of its trades now fell before 2016, and it took none of them, while the rest were 99% the same. So I went the other way: real gold from 2012 to 2015, relabeled as 2020 to 2023 so the EA would trade it. First I checked the method on years it does trade, and it repeated a real run almost exactly.
| Gold prices from | Profit factor |
|---|---|
| 2012 to 2015 (the years it won’t show you) | 1.28 (the buys lose) |
| 2016 to 2019 | 1.55 |
| 2019 to now | 2.25 |
Still positive in the years it doesn’t show you. But a lot less.
The live signal against the backtest
From May to October, the live signal made 114 trades. The backtest of the version on sale today made 96 in the same months. 89 of them are the same trades, with the same results: profit factor 3.02 live and 3.03 in the tester. So the tester reproduces the live account.
Then there are 25 trades the live account took and today’s version doesn’t. Together they lost money, with a profit factor of 0.58. Today’s version skips almost half of the live losses, 11 of 24, and only one in six of the wins. Live, all trades: profit factor 1.84. The backtest: 2.69. Why? I can’t tell you: another broker, MetaTrader 4, maybe an older version. What I can tell you is that the backtest of those months looks like the live account, minus a lot of its losses.
One more line from the page: “Broker Independent”. On another broker’s prices, only 77% of its entries matched, even if the total result was similar. One buyer wrote in a review that it’s “highly sensitive to each broker’s specific conditions”.
My AI Gold EA review in one table
| Question | What I found (v1.82) |
|---|---|
| Stop loss, grid, martingale | Stop on all 1,459 trades; no grid, no martingale |
| Risk-to-reward in the tester | Average win $14.92, average loss $20.44 (not 1:2) |
| One stop loss on default settings | 13% of the account; two at once, 27% |
| Is it AI? | Probably a real model: its picks PF 2.22 vs random 1.04 |
| Years before 2016 | It won’t trade them; on 2012-2015 gold, PF 1.28 |
| Live vs backtest, May-October | PF 1.84 live vs 2.69 backtest; 25 live trades missing (PF 0.58) |
My score: 5 out of 10 on its default settings
It’s well built: a stop loss on every trade, no grid, and probably a real model. On the years it shows you, it’s strong. But on default settings one stop loss takes 13% of the account, the ratio is the opposite of what the page says, and its backtest looks better than its live account and better than the years it won’t trade.
One setup holds up: the developer’s prop firm set, one trade at a time, 1.6% per stop. In my stress engine it reaches +8% before −5% in 94 histories out of 100. On 2012-2015 gold, that same set drops to a profit factor of 1.23. If you run it: a small fixed lot, one trade at a time, and expect less than the backtest.
Before you buy any AI EA, check four things
- In the tester report, compare the average profit trade with the average loss trade. Not the ratio on the page.
- Run the backtest again with a fixed lot, so compounding can’t hide the drawdown.
- Set the start date years before their chart begins. If it trades nothing, ask why.
- Open the live signal and backtest the same months. Compare them trade by trade.
Here’s how to backtest an EA on MT5 without fooling yourself, and how a Monte Carlo simulation shows you the histories a single backtest doesn’t. The rest of the series: The Gold Reaper (it held up) and Quantum Queen X (it didn’t).
The video version of this test is on my YouTube channel. Here’s the previous one in the series, where a grid EA wins 1,589 baskets and then loses one night:
If you test on gold, the data matters. I use Axi‘s XAUUSD history for my tests. Affiliate link; I may earn a commission if you open an account through it. And if you were thinking of the prop firm set: on allocated-capital programs, read the rules first. Axi Select, for example, only allows a third-party EA for generating signals, with trades placed manually.
A good AI shows you the years it hasn’t seen. Ask to see them. Which EA should I break next? Tell me, and get the next test first: join the DoItTrading newsletter.
Historical tests of My AI Gold v1.82, not a forecast. Live results can differ. Not affiliated with the developer. Trading involves risk, and this is not financial advice.
Frequently Asked Questions
Is My AI Gold worth it?
On its default settings, I give it 5 out of 10. It has a stop loss on every trade, no grid and probably a real model, but one stop loss is 13% of the account by default, its real ratio is the opposite of 1:2, and its backtest looks better than its live account.
Is My AI Gold really AI?
Probably yes. With its exits and hours, random entries made a profit factor of 1.04 and a simple trend-and-pullback rule 1.01, while its own picks made 2.22. No simple rule I tried reproduced its entries, and it kept almost 99% of them when I shifted the price history by a week.
What risk-to-reward does My AI Gold really have?
In the tester report the average win was $14.92 and the average loss $20.44, so it risks more than it makes. It needs about 58% winners to break even and wins about 75%, which is why it’s profitable.
Why doesn’t My AI Gold trade before 2016?
In my tests it opened nothing before January 4, 2016. When I ran real 2012-2015 gold prices relabeled as later years, it traded them with a profit factor of 1.28, against 2.25 from 2019 onwards.
Is the My AI Gold live signal the same as the backtest?
Mostly. From May to October, 89 trades matched with the same results. But the live account also took 25 trades the current version doesn’t, and together they lost money, so the live profit factor was 1.84 against 2.69 in the backtest.
Are you affiliated with the developer?
No. I build and sell my own EAs at DoItTrading, including a gold strategy, and I said so at the start of the review.
