Six Questions That Separate a Serious Trading Bot From a Sales Page

Before anything else, the disclosure: this blog is published by PraxAI, and we sell automation software for prop firm accounts. So when we write about how to choose a prop firm trading bot, we are not a neutral reviewer and we will not pretend to be one. That is precisely why what follows is a method rather than a ranking. We are not going to name competitors or tell you which product is best. We are going to hand you the questions a buyer should ask, knowing you will point them at us too.

If you want the broader landscape of automated tools for evaluations, the main read on this site is our overview of [AI trading bots for prop firms in 2026](/blog/best-ai-trading-bot-prop-firms-2026). This piece is narrower. It is the buyer side of the transaction, without relying on a sales page, a Discord testimonial or a screenshot you cannot verify.

One framing note. Almost every vendor now uses the word AI. Sometimes it means a model that learns from data, sometimes a fixed rule written in code with a marketing label on top. Both are legitimate. Describing the second as the first is not, so where the distinction matters below we mark it, and the longer version of that test is [whether any of these bots are actually AI](/blog/is-any-trading-bot-actually-ai).

This is the question that filters most of the market. Ask the vendor to describe, in language you could repeat to a friend, what the software does when it decides to trade. Not the outcome. The mechanics.

You are listening for specifics: what conditions it looks for, how many positions it can hold at once, whether it averages down, whether stops are attached at entry or managed later, and how it sizes a position relative to your risk setting. A vendor who built the thing answers in two minutes. A vendor reselling someone else's file usually cannot, and will redirect to results.

Refusal is the signal. Protecting a specific parameter set is reasonable. Refusing to say whether the strategy uses a martingale recovery sequence is not, because that single fact changes your risk profile more than anything else on the page. The forensic version of the same test, reading an answer against an actual trade history instead of taking it on trust, is in our list of [the red flags worth checking before you pay](/blog/prop-firm-bot-red-flags).

Prop firm rules are commonly structured around a daily loss limit, a maximum drawdown, minimum trading days and restrictions on news, weekends and lot sizes. The exact numbers vary by firm, account type and platform, they change over time, and you should confirm them on the firm's own site before relying on anything you read anywhere else.

That variability is the point. Software compatible with one firm's static drawdown may behave badly against another firm's trailing drawdown, even though nothing about the strategy changed. So the question is not whether the bot is allowed. It is whether the vendor ships a configuration for your firm and your account size.

Start from whether automation is permitted at all on the account you are buying, which our guide to [prop firms that allow EAs](/blog/prop-firms-that-allow-eas-2026) covers, and then ask about configuration. Vendors who take this seriously ship per firm settings files and explain what each one changes, which is the entire subject of [set files and what they actually do](/blog/setfiles-explained).

A profit curve is not evidence. It is a picture of a curve. Evidence is a methodology you can interrogate: which date range, which broker and spread assumptions, which slippage model, whether the parameters were chosen on the same data they are shown on, and what the results look like on a period the developer did not optimise against.

The gap between a tuned historical run and a live account is the most common disappointment in this category, and it is why we wrote about [why backtests and live EA results diverge](/blog/backtest-vs-live-ea-results). Ask for an out of sample window with explicit dates. Ask whether the results are from a demo or a funded account. Ask what the worst month looked like, because a vendor who cannot tell you has either not run it long enough or does not want to.

Then read the numbers correctly. Maximum drawdown, profit factor and average holding time say more about whether a strategy fits an evaluation than the headline return ever does, and a vendor who leads with the return has chosen which number you look at.

Every automated system has a bad day. What separates a tool you can live with from one you cannot is what it does on that day, and there are three sub questions to answer before you buy.

Does it stop? Ask whether there is a hard limit that halts trading, who defines it, and whether it is enforced in code or left to your discipline. Precision matters here: a loss limit that closes a position when a number is hit is a coded lock, not artificial intelligence, and a vendor who blurs that line is telling you something about the rest of their claims.

Does it tell you? Ask whether you get an alert when the software halts, disconnects or hits a limit, and where that alert arrives. Can you intervene? Ask whether you can close positions by hand without breaking the software's internal state. Losing an account rarely comes from one bad trade. It usually comes from the sequence described in [why funded traders lose the account](/blog/why-funded-traders-lose-the-account), where nobody was watching and nothing stopped.

Rules move. Firms adjust drawdown mechanics, consistency requirements and news policies with little notice, a pattern we covered in [what to do when a prop firm changes the rules](/blog/when-prop-firms-change-the-rules). A tool compatible in January is not automatically compatible in September.

So ask about the human side. Who answers support, on what channel, and in what typical timeframe. Ask what happened the last time a firm changed a rule, specifically, with a date. Ask whether updates are included or charged separately.

Commercial terms belong in the same conversation, because they are where surprises live. Ask exactly what you are buying: a lifetime licence, a subscription, or a licence tied to one account number. Ask how many accounts it covers and whether a second funded account costs more. Ask what the refund window is, what voids it, and who decides. Get it in writing. A vendor comfortable with these questions answers in a paragraph. A vendor who is not will answer with enthusiasm instead.

Some claims are not aggressive marketing. They are disqualifying, because they cannot be true of any trading product sold to the public.

Guaranteed income belongs in this category. So does guaranteed evaluation approval, because passing depends on market conditions, your account rules and execution, none of which a vendor controls. So does a win rate that would be remarkable in an institutional context being offered casually to retail buyers, any framing in which loss is treated as something the software removes, and any structure where the vendor needs your account credentials to trade for you.

Two softer signals are worth weighing. The first is results with no dates, no account context and no losing period shown. The second is a testimonial culture with no way to verify anything and no acknowledgment that some customers did not pass. Only 1 to 3 percent of funded traders keep the account long term. Marketing that implies otherwise is describing a different universe than the data does.

Publishing this list obliges us to be measured by it, so here is where we stand, including the parts that are not ideal.

On mechanics, the configuration we validate on gold takes one position at a time, with no martingale and no grid, and a news filter pauses around high impact events. On limits, PraxAI GUARD is a set of thresholds you define, enforced in code: it closes a position when your limit is reached and blocks an entry that would breach a ceiling. It is not artificial intelligence and we do not describe it as such. On commercial terms, the licence is 497 dollars once, covers unlimited accounts and carries a 7 day money back guarantee, with rule updates targeted within 48 hours.

Where we are weaker: our cTrader cBot is new and in validation with the first customers, so if cTrader is your platform, treat it as early rather than settled. MetaTrader 5 is the mature path. For futures, we will not tell you automation is permitted at any specific firm, because that is the firm's policy to state and not ours. Ask them in writing.

None of that makes us the right choice for you. It makes the answers checkable, which is the only thing this checklist was ever asking for.