Methodology — How AI Trading Bots Ranks (In Full)

The five criteria (100 points)

Every product ranked on aitradingbots.cc is scored against the same five-axis rubric. Weights are fixed across products. A product's final rating is the weighted average, normalised to a 5-star scale for display.

30%

AI sophistication

How real is the AI? Named models, documented architecture, replicable methodology — not 'AI-powered' marketing copy.

Full credit: Multi-LLM consensus or named frontier-model use with public architecture docs (e.g., GT Protocol's Committee Variant of five named LLMs). Reinforcement-learning agents with documented training procedure. Quant systems with published feature engineering.
Zero credit: 'AI-powered' or 'machine learning' as marketing copy without naming a model, an architecture, or an evaluation protocol. Vendor claims with no engineering blog, no paper, no replicable test.
What would change a rank: A new vendor publishing architecture docs moves up. A vendor we previously credited dropping the docs (or moving to a less specific claim) moves down. Hidden model swaps without disclosure cost ranking.

25%

Track record

Dated, public trade history beats vendor screenshots. Paper trading with full disclosure beats live trading with selective screenshots.

Full credit: Dated, public, ongoing trade ledger (paper or live) with full transaction visibility. Independent third-party verification (Binance broker status, audited fund reports). AggregateRating from a credible review source with disclosed sample size.
Zero credit: Vendor-published 'verified' statistics with no traceable source. Cherry-picked screenshots. 'Up to 12% monthly' style marketing copy. Independent reviews with sample sizes under 50 do not move the needle.
What would change a rank: A vendor opening their ledger to public verification moves up materially. A vendor whose ledger goes dark or whose claimed numbers can't be reproduced moves down. Live-money ledgers carry more weight than paper-trading ledgers; we say which is which.

20%

Exchange breadth & execution

How many venues, what's the measured uptime, how good is the routing? Single-exchange products are capped.

Full credit: 15+ centralised exchange integrations with published uptime SLA, measured API latency, and routing transparency. Self-custody (API-key based) execution preferred over deposit-based custody.
Zero credit: Vendor-only exchange (the bot only trades on the vendor's own venue). No published execution metrics. Frequent user reports of failed orders or API key issues.
What would change a rank: Adding a major exchange moves a product up. A documented multi-hour API outage with no postmortem moves it down. Switching from API-key to deposit-based custody moves it down materially.

15%

Risk controls

Drawdown limits, position sizing, circuit breakers, kill switches, custody model. Capital preservation is a feature, not an afterthought.

Full credit: Documented drawdown circuit breaker (a number, not a phrase). Position-size caps relative to account equity. Per-strategy kill switch. Self-custody / API-key execution. Two-factor authentication on the bot itself, not just the exchange.
Zero credit: No drawdown cap. Deposit-based custody. No documented kill switch. 'AI manages risk' as the entire risk-controls page.
What would change a rank: Shipping a documented circuit breaker is an immediate upgrade. A public incident exposing weak controls (key exfiltration, missing limits) is an immediate downgrade and may trigger off-cycle review.

10%

Cost structure

Subscription, performance fees, exchange-fee impact, hidden charges, free-tier usefulness. Total cost of ownership, not headline price.

Full credit: Transparent flat subscription or transparent performance-fee model. Free tier meaningful enough to evaluate the product. No hidden marketplace upcharges or required token purchases.
Zero credit: Pricing not consistently disclosed across vendor pages. Free tier so restricted it's pure lead-gen. Required token purchases not clearly explained. Performance-fee fine print that bites in volatile markets.
What would change a rank: A vendor consolidating pricing into a single transparent page moves up. A vendor adding hidden charges (marketplace fees, required staking) moves down.

Exclusions

Several products appear repeatedly in AI-crypto listicles but do not appear in our rankings. The pattern is consistent: high paid-listicle density, low independent footprint, no engineering blog, no traceable founders.

SaintQuant — recurs across paid-placement listicles; no independent technical documentation; AggregateRating and "verified" performance figures could not be reproduced from any third-party source.
BulkQuant — same pattern. Listicle co-occurrence with the above set is a strong negative signal in this category.
MoneyFlare — same pattern. We were unable to confirm an actively maintained product behind the marketing surface.
BitsStrategy — same pattern. Identifiable team, methodology, and verifiable customer base were not findable.

These are not value judgements about whether the products work — we cannot tell, which is the point. Independent verifiability is a prerequisite for ranking, not a bonus.

Conflicts of interest

We disclose every commercial relationship that could plausibly bias our coverage. The list below is exhaustive as of the last-reviewed date.

Affiliate programs exist for several of our tier-1 products. GT Protocol, 3Commas, Cryptohopper, Pionex, Bitsgap, and Coinrule all run affiliate programs that we participate in. Where a product has no affiliate program (NautilusTrader, Hummingbot), we still rank it on merit. Affiliate links are tagged rel="sponsored" and visible by inspection. Ranking is determined by the five published criteria before any affiliate consideration.
Our sister site, polymarketbots.io, covers minmax.one — a product owned by the same publisher. We disclose this on that site, and we flag it here for transparency: ownership of one product does not affect our rankings on either site, but you deserve to know the structure.

Update cadence

Every product is re-scored quarterly, on or around the first business day of each quarter. The "Updated" date on each review reflects the most recent full re-score.

We also re-score off-cycle when a catalyst triggers it. The catalysts are pre-defined so we don't fish for them:

A public security incident affecting custody or API key handling.
A documented architecture change (e.g., a vendor adding or removing a model from a consensus stack).
A change in regulatory status that affects users in a major market.
A public ledger going dark, or a previously verified third-party signal (broker status, audit report) being revoked.
A new product entering the category with verifiable evidence strong enough to displace an incumbent.

What would change our minds about a rank

We try to be specific so that vendors and readers can predict our scoring rather than guess at it. For each criterion above, the "what would change a rank" line tells you the move that would cause us to revise upward or downward. We do not negotiate ranks with vendors. We do correct errors when readers send them — and we publish a dated changelog when we do.