QUANTUM AI • GUIDES & RESEARCH
A checklist for an AI trading claim
Separate the product, provider, method and evidence behind an AI trading claim before considering any financial decision.
This website is an informational resource. It does not operate a trading robot, take deposits, connect to an account or offer personal investment advice. This checklist does not certify any provider or system.
“AI trading” describes a marketing claim until the seller explains what the product actually does. It might refer to research software, a signal, an order tool, a strategy that trades automatically or a managed account operated by another company. Those arrangements expose a person to different contracts, costs and risks. Before considering any of them, write down the exact offer and separate five questions: what is the product, who provides it, how does it make decisions, what evidence supports the result, and who controls it when something goes wrong?
The UK Financial Conduct Authority warns that AI can be useful for research but can produce inaccurate information and cannot predict future market conditions or guarantee returns. Its online-trading scam guidance also describes professional-looking sites that use return claims to persuade people to invest. A polished dashboard or technical vocabulary therefore deserves investigation, not automatic confidence. Use the worksheet below to organise questions; use the provider’s actual documents and current official records for answers.
1. Define the offer before judging the technology
Start with the proposed agreement, not the advert. Is the offer only a tool that displays information? Does it send trade suggestions that you must approve? Can it place orders through a linked broker account? Would another person or company decide how your money is invested? Each arrangement changes who can act, what you pay for and what might happen if the service stops. An “AI” label does not specify any of those points.
Ask for the asset class and legal product: shares, funds, foreign exchange, derivatives or something else. Establish whether you would own an asset, hold a claim against an intermediary or enter a leveraged contract. Next identify the money path: who receives deposits, where the account is held, who is allowed to withdraw and what happens if you cancel the tool. If the agreement does not answer these basic questions, there is no sound basis for evaluating a performance claim. The automation overview explains the difference between information, decisions and order execution.
Write down every cost even if the advertisement says the software is free. Trading can involve spreads, commissions, financing, currency conversion, subscriptions and withdrawal charges. Some costs are paid to a software seller, others to a broker or other intermediary. A screenshot of a gross return does not show what a user would receive after costs or losses. The US Commodity Futures Trading Commission includes fees, spreads and subscriptions among the factors to examine when assessing trading-bot promotions; the UK product contract still governs a UK reader’s particular arrangement.
2. Identify the legal provider and relevant permission
Collect the legal entity name in the agreement, trading names, website address, company number, firm reference number and contact details. These can describe different entities. A software vendor may license a tool while a broker holds an account; an advertiser may be a third party. Ask which entity is responsible for each part of the arrangement. Do not infer that a statement about one company applies to every organisation shown on the page.
For a UK financial service, navigate independently to the FCA Firm Checker and compare the precise entity and activity with the proposed offer. The FCA says its checker is intended to show whether a firm is authorised and has permission for the service being considered. If the legal names or contact details do not match, stop and resolve the difference through an independently sourced channel. A borrowed reference number, familiar logo or similar company name may be used by a clone firm. Our risk checklist keeps the provider check separate from technical claims.
Authorisation, where relevant, does not establish that the tool makes money, that a strategy is suitable for a particular person or that compensation would apply to every loss. Likewise, a lack of a published warning is not proof that a scheme is legitimate. Record what the official source actually says and the date you checked it. If the proposed activity is unclear, seek regulator or qualified independent guidance rather than letting the seller interpret its own status.
3. Ask for a plain description of the method
A useful technical explanation should identify the inputs, the decision process and the action. What data are used? How current are they? Does a model generate a forecast, rank signals, apply a fixed rule or send an instruction to place an order? Are people reviewing decisions? Under which conditions does the system stop trading? These questions can be answered without demanding a provider’s proprietary code. A refusal to explain the basic operation makes the product difficult to assess.
Distinguish a model trained on historical prices from a live service operating in changing markets. A system can fit past data and still fail when conditions change, data arrive late, execution is poor or the underlying relationships break down. A strategy that showed a positive result in a simulation may not produce the same result with live orders. The FCA’s current AI research guidance explicitly notes that AI cannot predict sudden market events or future performance. This is a limitation to account for, not a claim that every use of automation is useless.
Ask whether the system changes itself after deployment and who approves changes. If the approach is adjusted repeatedly after losses, an old result may describe a different strategy. Also ask how orders are limited: position size, exposure, margin, maximum loss, prohibited products and human override. A provider should be able to identify the person or system responsible for those controls. The testing and evidence guide helps distinguish a demonstration from an operational record.
4. Classify every performance number
Label each chart or figure as a backtest, a paper-trading simulation, a selected example, a live account statement or an independently examined record. Do not combine them. Ask for the period covered, the full sequence of gains and losses, the starting capital, position sizes, fees, spreads and the assumptions used. A single “win rate” does not reveal the size of losses or the risk of a large drawdown. A chart that shows only successful trades cannot support a claim about the overall strategy.
For backtests, ask when the strategy was designed and whether the test data were genuinely separate from the data used to develop it. Ask whether delisted assets, failed orders, funding costs, slippage and periods of market stress were included. These are editorial questions about the reliability of evidence; they do not make a simulation predictive. For a live record, establish which legal account produced it and whether it shows deposits, withdrawals and all costs. Do not accept an account balance on a web dashboard as independent proof that funds can be withdrawn.
The FCA has warned that online trading scams can show apparent early returns before encouraging a person to invest more. The CFTC warns that fraudsters have used the AI label to promote unreasonable or guaranteed returns. Those advisories do not prove that a particular product is fraudulent, but they show why marketing figures need independent context. Treat “guaranteed”, “risk free”, “daily profit” and unexplained urgency as reasons to stop, not as targets to validate by sending money.
5. Map account control and possible loss
Work out what access the software or provider would receive. Can it only read market data? Can it place or cancel orders? Can someone change limits, transfer assets or withdraw money? What is the process for revoking access, disconnecting an application or closing an account? Keep these questions separate; permission to trade is not the same as permission to move funds, and a dashboard switch may not terminate a legal agreement.
Check the downside in the underlying product. If leverage or margin is involved, find the close-out terms and any applicable loss protections in the actual contract. Identify who bears losses caused by an interrupted connection, a bad signal or delayed execution. Ask whether the service has a documented outage process, record export and complaint route. A technical description alone will not answer contractual liability questions. A provider’s support promises should appear in its terms or other verifiable documentation.
Consider whether you could afford to lose the amount at risk without relying on projected gains. Automation can change the speed or frequency of transactions; it does not remove market risk. The FCA says even an authorised firm with the correct permissions does not eliminate all risk. If the product or responsibilities remain unclear, a pause and qualified independent advice may be more useful than another demonstration video.
6. Keep an evidence sheet, including unknowns
For each material claim, write the exact wording, where it appeared, when you checked it, the document or independent record that supports it, and what remains unresolved. Separate facts from interpretation. A sensible note might say “the seller calls this AI, but the contract describes a signal subscription; live results were not provided”. Another might say “the firm name in the terms differs from the firm named in the advert; independent clarification needed”. Neither note should be upgraded to a recommendation to invest.
Recheck the evidence when a product, fee schedule, provider, strategy or account arrangement changes. Save the relevant version of the terms and any message that influenced your decision. This article reflects official FCA and CFTC consumer guidance checked on 27 September 2026. It has not tested a trading system or reviewed a named company. For a UK decision, return to current FCA materials and the exact provider record at the time you act.