Can AI find undervalued stocks?
Last updated August 2026
Short answer
The word doing the work in this question is undervalued. Cheap is an observation about a ratio. Undervalued is a claim that the market is wrong.
Screening is the solved part
Filtering thousands of listed companies by price-to-earnings, price-to-book, free cash flow yield, debt levels or margin trends is arithmetic on public data, and tools have done it since long before the current wave of AI.
Because it is cheap, everybody has the output. A screen that any retail platform runs for free is not information anyone else lacks.
A screen is a starting list. Treating it as a conclusion is where most of the damage happens.
Why cheap is usually correct
A low multiple is a statement about expectations. The market is saying it does not expect current earnings to persist, or it sees a risk it wants compensating for.
Sometimes that view is wrong, and the stock is a bargain. Often it is right, and the cheapness is the market pricing a decline in advance.
Distinguishing between these two is the entire discipline of value investing, and it depends on understanding the business rather than the ratio.
What AI adds after the screen
Reading the annual report, which most people do not do. A model can work through a 10-K and pull out how revenue is actually earned, what the risk factors say, and where the debt sits.
Comparing filings across years, which is where deterioration shows up first. Language about a product line that quietly changes is a signal a ratio never carries.
Explaining an unfamiliar industry well enough that you can form your own view. That is the honest version of the value AI provides here.
Try it in Walnut
Walnut reads the holdings you already own and answers questions about them against real filings, which is a narrower and more checkable job than picking winners.
What to be careful about
A confident valuation with no stated assumptions is not analysis. Ask what growth rate, margin and discount rate produced the number, then decide whether you believe them.
Backtested screens look excellent because they were tuned on the period they are shown against. A rule that worked on the last ten years has not been tested on anything.
Any product claiming its AI reliably finds mispriced stocks is making a performance claim. The CFTC has warned specifically that claims of high or guaranteed returns are red flags of fraud, and the SEC has charged advisers for overstating AI capability.
A sensible way to use it
Screen if you like, then take three names and read their filings with the model rather than asking it which to buy.
Write down what would have to be true for the stock to be worth more than it trades at, then check whether the filings support it.
Check the answer against the document. Every claim about a company that matters is in a filing you can open.
What a screen cannot see
Anything not yet in the financials: a contract about to be lost, a regulator preparing a decision, a founder who has quietly stopped showing up.
Accounting choices that make the ratios look better than the business. Two companies with identical margins can be capitalising very different things.
The reason the market disagrees with you, which is usually the single most valuable thing to find out, and it lives in the filings and the transcript rather than the screener.
Sources
Company filings are published by the SEC through EDGAR full-text search. Warnings about AI-based return claims are in the CFTC customer advisory of 25 January 2024 and SEC press release 2024-36. Walnut is informational and is not an investment adviser. This guide is educational and not personalized investment advice.
FAQ
What does undervalued actually mean?
+
That a company trades below what it is worth, which requires a view of what it is worth. A low price-to-earnings ratio is not that view. It says the market pays little for current earnings, which is sometimes an opportunity and sometimes an accurate assessment of a declining business.
Can AI screen for value?
+
Yes, and so can a free stock screener. Filtering by valuation ratios, margins, debt and growth is mechanical work that software has done for decades. The screen is not the hard part and it is not where an edge comes from.
What is a value trap?
+
A stock that is cheap for a reason that has not finished playing out: eroding demand, a legal overhang, an obsolete product, accounting that flatters the past. It stays cheap, or gets cheaper, and the low ratio that attracted you was the market pricing a decline you had not modelled.
So what is AI good for here?
+
Reading. Working through a 10-K, comparing this year's language against last year's, pulling out what management said about the thing you are worried about, and explaining a business you do not know. That is real work and it is checkable against the filing.
Can AI value a company for me?
+
It can run a discounted cash flow with assumptions you supply, and the output is only as good as those inputs. A valuation quoted without its growth rate, margin and discount rate is a number, not an argument.