Blog › Guides
How to screen stocks with AI, without writing code
A classic screener makes you translate your idea into boxes and sliders. With an AI assistant connected to real data you describe it the way you think it, and the assistant does the translating.
16 November 2026 · 5 minute read
What changes
In a normal screener you need to know what each metric is called, which unit it is in and where the filter lives. With an assistant connected over MCP to a fundamentals database, you say what you are after and it picks the metrics, builds the filters, runs them against the 25,000 companies and hands you the list. If something looks wrong, you tell it and it refines.
Five prompts to start with
"European companies with a ROE above 15%, no net debt and a P/E below their sector median."
"The 20 US companies whose revenue grew the most last quarter against the same quarter a year earlier, with a gross margin above 40%."
"Japanese companies worth less than their net cash."
"Australian dividend payers whose dividend is covered by free cash flow, not just by earnings."
"Companies whose ROIC beats their cost of capital in every capture we have."
What happens behind the scenes
It helps to know this to understand the answers. Usually the assistant first checks which metrics exist and what they are called. Then it runs the screen with your conditions turned into filters, for example roe > 0.15: percentages are stored as fractions, so 15% is 0.15. Finally it shows you the result and usually explains which filters it used. If it does not, ask it to: that is how you check it understood what you meant.
Four mistakes to avoid
- Comparing sectors that are nothing alike. A P/E of 12 is expensive for a bank and cheap for a software company. Always ask for the comparison against the sector median.
- Trusting a single year. A one-off profit, such as selling a division, makes a company look cheap when it is not. Ask for the figure's history before you believe it.
- Confusing cheap with good. Many companies trade cheaply because their business is getting worse. Once you have the list, ask why each one is cheap.
- Asking for too many conditions at once. With eight filters it is easy to end up with nothing. Start with two or three and add as you go.
From the list to an idea
The screen only gives you candidates. The useful part comes afterwards, in the conversation:
- "Of these, which have improved their margin over the last captures?"
- "Compare the top three with their two closest peers."
- "Which of them report results in the next two weeks?"
And before deciding anything, read the company's accounts. AI saves you hours of searching, but it does not replace understanding the business.
Connect Market Pi-Rates to your AI assistant and run your first plain-language screen. The free plan covers every company.
Try it freeInformation only, not investment advice.
