AI can produce a stock-price forecast. That does not establish that the forecast is reliable, that it will make money after costs, or that the service is suitable for your savings. A chart of successful predictions needs an explanation of how those results were produced.
For readers exploring investing after 40, the useful skill is evaluating the claim. You do not need to build a trading model or connect a brokerage account to ask the following questions.
Start with the evidence behind the chart
Ask whether the results came from actual trades, a demonstration account, or a backtest—a simulation using historical data. Request the dates, the complete results, and an explanation of which costs were included.
Hypothetical performance has limitations. The CFTC’s required disclosure for certain hypothetical commodity-trading results highlights that simulated trades were not actually executed, may inadequately reflect factors such as liquidity, and can benefit from hindsight. That rule concerns commodity trading; it also illustrates why a simulated result deserves careful scrutiny rather than being treated as a live track record. Read the disclosure in 17 CFR 4.41.
Ask what “accurate” actually means
A claim of 85% accuracy is incomplete without a definition. Was the model predicting tomorrow’s closing price, whether the price would rise, or whether a particular trade would earn money? How many predictions were evaluated? Were the unsuccessful ones included?
Winning frequently does not necessarily mean making a profit. In this invented example, eight trades each earn $5, while one loses $50. Eight of nine trades win, but the total is a $10 loss before costs. The example illustrates arithmetic, not a forecast or trading strategy.
Ask for returns after the costs relevant to the service, the size of losses, and a clearly identified comparison investment over the same dates. A percentage without its measurement method is a marketing claim you cannot properly evaluate.
Use these questions to examine a backtest
- What information was available at the moment each simulated decision was made?
- Was the model evaluated on a later period that was not used to build or tune it?
- Were losing periods and failed versions disclosed, or only the best-looking result?
- What assumptions were made about trading costs, available buyers and sellers, and the price at which orders would execute?
- What changed between the historical test and the service being sold today?
You are not expected to audit the software yourself. These questions help identify what evidence is missing and what a qualified reviewer would need to examine. “Our proprietary AI handles it” is not a substitute for an answer about the results.
Give a chatbot a narrow reading task
Copy the service’s public performance explanation, without personal account details, and try this:
List every measurable performance claim in this text. For each, identify the time period, comparison, costs, and whether the result is simulated or from actual trades. Use only information explicitly provided. Make a separate list of questions the text leaves unanswered.
Check the output against the original wording. The goal is a question list, not a chatbot’s verdict that the investment is safe.
Keep forecasts separate from decisions
The CFTC warns that AI cannot predict the future and that guaranteed-return bot offers can be fraudulent. See its AI trading-bot advisory.
Review our AI investing checklist for questions about the product, provider, and your intended use. An impressive prediction does not answer whether a particular investment belongs in your financial plan.
This article provides general education, not personalized investment, tax, or legal advice.
Reviewed October 2, 2026. Featured photo: Bradley Andrews / Unsplash; for illustration.
AI disclosure: This article was drafted and revised using AI. Read our full disclaimer.
