Guide
An AI chat gave you a figure about a stock: how to check it
You ask an AI chat what it thinks of a stock, and it answers with confidence: a score, a revenue figure, a dividend. The answer is clear and well written… but nothing tells you how old that figure is, or where it comes from. This guide shows you how to check it with three questions, no finance degree required.
Updated
Why can an AI chat get a figure wrong?
A general-purpose assistant wasn’t built to track company accounts. It answers from what it read during its training, sometimes topped up with a web search. Two things can then go wrong.
- The figure may be old. A company’s accounts change every quarter, its share price every day. A figure that was right a year ago can be wrong today, and the assistant doesn’t always say so.
- The figure may have no source. When it can’t find the information, an assistant can produce a figure that looks plausible. It isn’t malice: it fills the gap, the way it would finish a sentence.
That’s no reason to stop using it. It’s a reason to check before you decide.
What questions should you ask a figure?
- How old is it? In investing, a figure without a date is worth nothing. Ask the assistant to state the date: the day of the price, the year of the accounts. If the answer stays vague (“recently”, “according to the latest data”), treat the figure as undated.
- Where does it come from? A serious figure has a source you can open: the company’s annual report, its investor website, or a service that says where it gets its data. “Public sources” is not a source.
- Does it fit with the rest? Compare it with a second place. If the assistant describes a very profitable company while its latest accounts show a loss, one of the two is wrong. Also ask the same question twice: if the answer changes while nothing else has, don’t rely on it.
An example: a figure, its date, its source
Take StockConviction’s public page for Microsoft, as we read it on 29 October 2026. It shows a soundness score of 84/100. At the top of the card, a line says “Data as of 29 October 2026”: that’s the date. At the bottom of the same card, the “How we rate a stock” link leads to our method, which says what the score is made of and where the data comes from: the accounts the company publishes, via a data provider the method names. All three questions have their answer on the page.
If you open Microsoft’s page today, you may see another score and another date. That’s normal, and it’s the whole lesson: a figure only makes sense with its date.
What does the StockConviction connector change?
If you use Claude or ChatGPT, you can plug the StockConviction connector into it. When you ask about a stock, the assistant can then read our analysis instead of answering from memory: an analysis recalculated every night from published accounts, which it receives with the date of the price and the date of the analysis. So you can ask it “how old is this figure?” and get a dated answer.
The analysis follows a single public method, the same for every stock: the company’s soundness first, its price second. The connector reads; it places no orders. It is Premium only.
In practice
Next time a chat gives you a figure about a stock, ask it the three questions: how old? where from? does it fit with the rest? If it can’t answer, go and check the source yourself. For a dated starting point, our list of quality stocks is recalculated every night. No figure, ours included, decides for you: we shed light, you decide.