BUYER'S GUIDE · 5 AUGUST 2026
How to choose a gold trading bot
Put these nine questions to any vendor, including us. Marketing copy cannot answer most of them, and what a vendor avoids tells you more than what they advertise.
What is the maximum drawdown, and over what period?
Ask this before you ask about returns. Any bot can show a good month, so ask what it cost to get there. A gain quoted without a drawdown figure is half a number. If the drawdown is missing from the marketing entirely, assume it is bad enough to hide.
What a good answer looks likeA drawdown figure quoted alongside every return figure, over a stated period.
Has it been tested on data it was not tuned on?
This question reveals the most. Almost every published backtest is in-sample: the settings were chosen by studying the same months the results come from, which guarantees a good-looking chart and proves nothing. Out-of-sample testing means picking the configuration on one period, then running it untouched on a different period the developer never looked at.
What a good answer looks likeTwo separate periods reported, with the out-of-sample one visibly weaker. Identical results across both usually means it was not really out-of-sample.
How many strategies actually trade, versus how many exist?
Strategy count is the most gamed number in this market. A bot advertising thirty strategies is telling you thirty exist. It is not telling you thirty are enabled, or that thirty were measured and kept. Ask how many are switched on by default, and what happened to the rest.
What a good answer looks likeA vendor who can name the strategies they removed and why. Anyone who has never disabled anything has never measured anything.
What is the minimum account size, and what happens below it?
Most gold bots quietly stop working on small accounts, for reasons of arithmetic rather than quality. Brokers enforce a minimum trade size, typically 0.01 lots, which is roughly $9 to $12.50 of risk on gold. Below about $700, that floor exceeds the risk your percentage setting asks for, so every trade clamps to the minimum and risks far more of the account than you configured.
What a good answer looks likeA vendor who volunteers a minimum balance and explains the lot-floor problem. Silence here means small accounts get a much riskier product than the settings imply.
Does it use martingale, grid, or averaging down?
These strategies produce smooth equity curves until the day they fail. Doubling into a loser wins repeatedly and then loses everything at once. They are also the strategy types prop firms most commonly ban outright, so if you intend to trade a funded account, a yes here rules the bot out.
What a good answer looks likeA direct no. If the answer is evasive, or the words "recovery mode" appear anywhere, it is a yes.
What does it do during high-impact news?
Gold gaps hardest and spreads widen most around CPI, NFP and FOMC releases. A bot that ignores the economic calendar takes the worst fills of the month at the moment your account can least afford them.
What a good answer looks likeA news filter that is on by default and pulls a real calendar, with nothing the user has to remember to set by hand.
Is there a hard daily loss limit?
Most accounts that fail go on a single bad day rather than in a slow decline. A daily lockdown stops trading once a loss threshold is crossed and does not resume until the next session, so the bot cannot keep digging once the day has gone wrong. On a prop firm account this maps directly onto the daily drawdown rule that ends most challenges.
What a good answer looks likeA configurable hard stop, and a vendor who tells you to set it below your prop firm threshold rather than at it.
Can I actually open the account this bot needs?
Buyers overlook this one, and it rules many of them out. Many gold bots are written for offshore brokers that cannot legally accept US residents, and US retail traders cannot trade spot gold CFDs at all. If you are in the United States, check this before anything else on this list.
What a good answer looks likeA vendor who states which brokers and jurisdictions the bot suits. Most stay silent because the honest answer costs them the sale.
What is the refund policy, and can you test before paying?
Any bot worth buying can be run on a demo account first. If a vendor resists demo testing, or the guarantee has conditions that make it unusable in practice, that tells you what they expect the first month to look like.
What a good answer looks likeA clear money-back window and encouragement to demo first.
Our own answers to the nine questions
Publishing this list without answering it ourselves would be cheap. Our answers, briefly:
Common questions
What is the most important thing to look for in a gold trading bot?
Out-of-sample testing. Almost every backtest you will see is in-sample: the settings were chosen by studying the same period the results come from, which guarantees a good-looking result and proves nothing about the future. A vendor who has tested the configuration on a separate period it never saw, and publishes that weaker result beside the good one, is telling you something real.
Are gold trading bots profitable?
Some are, over some periods, with a drawdown you have to be willing to sit through. A vendor promising consistent profit or a daily income figure is making a claim they cannot support, and the more precise the number, the less likely it is to be true. Treat "make $X per day" as a reason to walk away.
How much money do I need to start with a gold bot?
More than most vendors admit. The broker minimum of 0.01 lots costs roughly $9 to $12.50 of risk per gold trade, so on a small account every trade clamps to that floor regardless of your risk percentage. Our own bot runs from $100, but $1,000 is the number to plan on: every figure we publish was measured at $1,000, and its default fixed 0.02 lot does not shrink with a smaller account.
What claims should make me walk away?
Guaranteed profits, guaranteed prop firm passes, a specific daily income figure, a win rate above about 90%, returns quoted with no drawdown, and testimonials with performance claims but no verifiable source. Each of these is a sign the vendor is selling a story rather than a measured system.