How We Test AI Trading Bots: Our Review Methodology

Quick Answer

AITradingBotReview tests AI trading bots through a minimum 30-day live trading window using real money on exchanges where readers can actually deploy them — currently OKX, with Bitget and Bybit testing scheduled. We track 14 quantitative metrics covering performance, risk, and trust, and we apply a six-component weighted rating formula in which Risk Controls counts for 30%, Live Performance 25%, Transparency and Evidence 20%, Fees 10%, UX and Setup 10%, and Support and History 5%. We publish floating losses alongside closed profits because closed-trade-only reporting can hide drawdown. Verified evidence means exchange dashboards, API exports, and live YouTube streams — not vendor marketing screenshots. We disclose affiliate relationships and editorial conflicts of interest in every review.

Why We Wrote This Methodology First

Most AI trading bot reviews online have one obvious problem: the reviewer never actually traded with the bot. Pages get written, screenshots get borrowed, affiliate links get inserted, and a rating gets assigned — sometimes within hours of the writer first encountering the product. We have been on the receiving end of those reviews, and we built AITradingBotReview to be the opposite.

Crypto AI trading is a high-stakes category. Readers who follow our recommendations are putting real capital — sometimes their savings — into automated systems running on leveraged futures markets. A bad review costs them money. A dishonest review costs them more. So before we publish a single bot rating, we want readers to know exactly how we test, what we measure, what we count as evidence, and where we have potential conflicts that might bias the result.

This page is the source of truth for every rating elsewhere on this site. If you read a bot review here and the methodology behind a claim is unclear, this page should answer it. If it does not, contact us and we will fix the gap.

“

A bot that looks excellent in a strong directional week often collapses when the market reverses. Thirty days is the minimum window that exposes a bot to multiple regime shifts.

Testing Protocol

Our protocol is intentionally narrow. We do not try to test every bot on every exchange — we go deep on a few pairings, document what we find, and move to the next.

Minimum Duration: 30 Days

Every bot must run for a minimum of 30 consecutive days under real-money conditions before it earns a rating. Most automated trading strategies work in some market regimes and fail in others — a Martingale-adjacent strategy can post a perfect closed-trade record during a sideways week and bleed during a strong trend. Thirty days is the lower bound that exposes a bot to enough regime variation to draw conclusions. When floating positions remain open at the 30-day mark, we extend testing until those positions close, then publish the full sequence.

Minimum Capital: $1,000 USDT

We fund every test account with a minimum of $1,000 USDT. Bot performance changes meaningfully with account size — position sizing logic, fee drag, and drawdown tolerance all scale non-linearly. Testing a bot with $50 in a demo account tells us almost nothing about how the bot behaves with $5,000 in live capital. $1,000 is the floor where the math starts to resemble what real users experience.

Exchanges Tested

OKX is our primary testing exchange today, selected for Fast API support, strong futures liquidity, and proof-of-reserves transparency. Bitget and Bybit testing are scheduled — we will not publish bot reviews specific to those exchanges until we have at least 30 days of live data on them. Exchange-side variation matters: a bot that works well on OKX may behave differently on Bitget due to API rate limits, fee structure, or liquidity depth.

API Permission Scope

Every bot we test connects through Fast API or trade-only API permissions. Withdrawal authorization is never granted on the API key — if a bot requires withdrawal access to function, the review ends there with a permanent failed rating. We document the exact permission scope per bot in the review.

14 Metrics We Track Daily

During every active test, we log fourteen metrics daily. They fall into three groups: performance (what the bot is producing), risk (how exposed the account is), and trust (what protections are in place). The full list:

Performance Metrics

• Closed P&L — total profit or loss across positions that have been opened and closed during the test window.
• Floating P&L — current unrealized profit or loss on positions still open at the time of measurement.
• Win rate (closed trades only) — percentage of closed positions that closed in profit.
• Number of closed trades — total executed trades that completed during the window.
• Number of open positions — current count of unclosed positions (the floating P&L denominator).
• Average hold time — mean duration from open to close across closed trades.

Risk Metrics

• Maximum drawdown — the largest peak-to-trough equity decline during the test window.
• Leverage used — average and maximum leverage applied across positions.
• Maximum position size — the largest single-position exposure as a percentage of account equity.
• Per-coin stop loss — whether each individual coin position has an enforced hard stop, and at what threshold.
• Asset Guard / Equity Guard threshold — the account-level circuit breaker that halts trading on drawdown.

Trust Metrics

• Fee drag — total fees paid as a percentage of gross profit (a 20% performance fee with 80% gross profit becomes a 20% drag, not 16%).
• API permission scope — the exact permission set on the API key (read, trade, withdraw).
• Withdrawal authorization — whether the bot ever requests, requires, or attempts withdrawal-permission access. The answer must be "no."

“

Closed-trade-only reporting can hide drawdown. We require both numbers — closed P&L and current floating P&L — and we publish them together.

What Counts as Evidence

Not every screenshot deserves the same weight. Here is what we treat as primary evidence and what we treat as vendor-reported until verified.

✅ Counts as Verified Evidence

• Direct exchange dashboard screenshots with visible timestamps showing account balance, open positions, and P&L.
• Exchange API exports — CSV or JSON data pulled directly from the exchange's account history endpoint.
• Live trading sessions recorded or streamed where the exchange dashboard is visible end-to-end.
• YouTube Live exchange dashboard streams — our planned cadence is to broadcast a live OKX dashboard session at regular intervals so readers can verify our reported numbers in real time.

❌ Does Not Count as Evidence

• Vendor marketing screenshots — promotional dashboards under vendor control are vendor-reported by definition.
• Affiliate dashboard numbers — affiliate metrics measure referrals, not user outcomes.
• Closed-trade-only win rate without floating P&L — a 100% closed-trade win rate is not a complete picture if open losing positions are not also disclosed.
• Screenshots without timestamps or context — a profit number alone, with no date, exchange, or capital reference, is unverifiable.
• Single-week performance windows — anything under our 30-day minimum.

The Rating Formula

Every bot rating on this site is a weighted sum of six components. We publish the component scores alongside the final rating so readers can see which parts of a bot are strong and which are weak — a "good overall" rating that papers over weak risk controls is not useful.

Why Risk Controls Count Most

Risk Controls carries the heaviest weight at 30% because losing money is the most expensive failure mode in this category. A bot that performs well in good months but lacks circuit breakers can erase a year of returns in a single bad week. Asset Guard, Equity Guard, per-coin stop losses, position sizing rules, and API permission scope all live under this component. We score conservatively here — vendors who downplay risk controls or make them optional lose points heavily.

Why Performance Is Only 25%

Live performance is critical, but it is not the highest-weighted component. The reason: performance over a 30-day window is partially a function of market regime luck. A bot that posted a 10% return during a sideways month might post a 20% drawdown during a trending month. We weight performance lower than risk to penalize bots that produced good numbers in the test window without the structural protections to survive a bad regime.

How We Compare to Typical Affiliate Review Sites

For context on why this methodology exists, here is how our approach contrasts with the affiliate-driven review sites that dominate the first page of search results for "AI trading bot review" queries:

We are not claiming the moral high ground — we run affiliate links too. The difference is that we want readers to know exactly what we tested, how we tested it, and where we may be biased before they decide whether to trust the rating.

Editorial Disclosure

AITradingBotReview earns affiliate commissions when readers sign up for products through our links. Members of our editorial team have professional relationships within the crypto trading bot industry, including consulting and advisory roles. We disclose this because it could affect how you weigh our reviews. We mitigate potential conflicts by:

• Publishing this methodology page in full and linking to it from every review.
• Showing floating losses alongside profits in every test result we publish.
• Linking to independent third-party criticisms (Trustpilot, technical reviewers) within every bot review.
• Maintaining ratings even when they sit below industry hype levels — we have not raised a rating to match vendor marketing claims, and we will not.
• Refusing affiliate-only reviews — every rating on this site comes from a real test we ran with our own capital.

What We Do Not Verify

To stay honest, here is the inverse: things our methodology does not check. If a review on this site depends on any of the following, we mark it explicitly as vendor-reported.

• Vendor user counts and aggregate profit claims — we cannot independently verify "240,000 users" or "$60 million in collective profits" without exchange-side audit access.
• Vendor security audits — we read published audit reports but we are not auditors ourselves.
• Long-term track records past our test window — we cite vendor-published historical data with attribution but we did not personally run those years.
• Other users' results — we test our own accounts. Public user testimonials are anecdotal and we treat them accordingly.

Ken & Our YouTube Companion Content

On-site reviews and ratings are determined by the AITradingBotReview editorial team. Our companion video content — including step-by-step setup tutorials, monthly performance updates, and live exchange dashboard streams — is presented by Ken on the @TraderAgentClub YouTube channel and on TraderAgent.club.

Ken's video format is conversational and first-person — appropriate for video. The site you are reading now uses brand-as-presenter framing because our reviews are editorial, not personal. The same person can run multiple platforms with different formats; we just want to be clear about which voice belongs where.

Living Document

This methodology will evolve. As we test more bots, expand to more exchanges, and learn from cases where our framework missed something, we will update this page and date the change. Readers should always feel free to contact us with methodology suggestions or evidence challenges.

Last updated: May 3, 2026 · Next review: Quarterly · Suggest a change: Contact