Can AI Pick Stocks? What It Can and Cannot Do (2026)

Federico Kunze Küllmer is the founder of Moneda, one of the tools compared here.

Short answer: AI cannot reliably pick winning stocks, and no honest tool will promise you that. What it can do well is the slow part of research: reading company filings, pulling the numbers, laying out the case for and against a company, and flagging when something changes. Used that way, AI saves you hours and helps you ask better questions. The decision to buy or sell, and the risk that comes with it, stays with you.

Why "picking stocks" is the wrong question

A stock price reflects what millions of buyers and sellers already know and expect. To beat that consistently, you need to know something the market has not priced in yet, or be right about the future when most people are wrong. An AI model reads the same public information everyone else reads. It has no crystal ball.

So the useful question is a different one. Can AI help you understand a company faster and more clearly than you could alone? Yes, often by a wide margin.

Think of it like a very fast research junior. It will read a 200 page report in seconds and summarise it. It will also, now and then, get a number wrong with complete confidence. You would not let a junior place trades unsupervised, and the same goes here.

What AI does well

1. Reading the boring documents

Public companies in the US file a 10-K every year. That is the annual report sent to the regulator, with audited accounts, risks and how management sees the business. It is long and dry, and most investors skip it. An AI model will read it for you and answer specific questions about it.

Good questions to ask:

• "What are the three biggest risks this company lists in its latest 10-K, in plain English?"
• "How much of revenue comes from its largest customer?"
• "Did the company change how it reports any segment this year?"

2. Working out the numbers

AI can pull figures from filings and compute ratios. Two you will see often:

• P/E ratio (price to earnings): the share price divided by profit per share. A P/E of 30 means investors pay 30 times one year of profit for the stock.
• Free cash flow: the cash a business has left after paying for its operations and investments. Profit can be shaped by accounting choices. Cash is harder to fake.

The catch is that a general chatbot may recall a figure from memory instead of reading it from the source. That is where wrong numbers come from. A tool built for finance should work each figure out from the actual data and show you where it came from.

3. Giving both sides

Most stock content online is one-sided: either a hype thread or a doom thread. A well-built AI tool can lay out the bull case (why the stock could do well) and the bear case (why it could go badly) next to each other. Reading both is one of the cheapest ways to avoid buying on excitement alone.

Try: "Give me the strongest bull case and the strongest bear case for this company, with the numbers behind each."

4. Judging business quality

Some companies have a moat: a lasting advantage that keeps competitors away, like a strong brand, a network that gets better with more users, or costs nobody else can match. AI can work through frameworks for this and explain its reasoning.

It can also check quality of earnings, meaning whether reported profit is backed by real cash coming in, or propped up by one-off gains and accounting choices.

5. Finding ideas from a description

Screening used to mean filters and spreadsheets. Now you can describe what you want in a sentence: "US companies with growing free cash flow, little debt and a dividend." The AI turns that into a shortlist. A shortlist is where research starts. It is not a list of things to buy.

6. Watching and stress-testing

Once you own something, AI can track it and tell you when a filing, a result or a price move changes the picture. It can also run your portfolio through past crashes to show how it might have held up.

What happened when Der Spiegel let ChatGPT pick a portfolio

In June 2025 the German magazine Der Spiegel asked ChatGPT to build a portfolio that would beat the MSCI World, a broad index of large companies in developed countries. It reported the results in October 2026.

On paper, the AI won. From July 2025 to July 2026 its portfolio of four ETFs and three single stocks gained about 42.5%, against about 25.8% for an MSCI World ETF. The magazine was clear that this proves less than it seems:

• Some of the facts were wrong. Several of the fund ID numbers ChatGPT gave did not match the fund or did not exist. The journalists had to fix them before they could build the portfolio.
• It was one big bet, counted several times. Nvidia was a single stock in the portfolio and also a top holding of its AI ETF. Palantir and Broadcom sat in the portfolio and in its momentum ETF too. That is concentration (a lot of money riding on a few companies), not diversification.
• One good year is not a method. A fund manager the magazine asked argued that a good answer would have questioned the task itself, because no one can reliably plan to beat the index by picking well-known stocks.

The lesson matches the rest of this guide. Use AI to gather and check facts, ask it to show its sources, and look at what you actually own across all your funds before you add more of the same thing.

What AI cannot do

Be clear-eyed about this part.

It cannot predict prices. Short-term moves are driven by news, sentiment and flows that nobody forecasts reliably. Any tool claiming to know where a stock goes next week deserves suspicion.

It can be wrong and sound sure. Language models sometimes produce plausible numbers that are not in any source. Finance-focused tools reduce this by grounding answers in data and citing it, but the risk never drops to zero.

It does not know your life. Your tax situation, when you need the money and how you would feel watching your portfolio fall 30% are things that matter more than any ratio.

It only knows what it can read. Professional investors pay for broker research, expert calls and private data. Most consumer AI tools, Moneda included, do not have that. If your edge depends on it, a consumer tool will not give it to you.

Valuation is still judgement. A DCF (discounted cash flow) model estimates what a company is worth today from the cash it might make in future. AI can build one in seconds. But the answer depends on guesses about growth and risk, and small changes in those guesses swing the result a lot. The AI can do the maths. Picking the assumptions is your call.

How to use AI for stock research, step by step

1. Start with the business, not the price. Ask: "Explain how this company makes money, as if I had never heard of it."
2. Read the risks. Ask: "What does management say could go wrong, and has any of it started happening?"
3. Check the numbers against the source. Ask: "Show me revenue, profit and free cash flow for the last five years, and where each figure comes from."
4. Ask for both sides. Ask: "What would have to be true for this stock to fall by half?"
5. Compare it. Ask: "How does this compare with its two closest competitors on margins and debt?"
6. Fit it to your plan. Ask: "If I added this to my portfolio, how concentrated would I be in one sector?"
7. Decide yourself. Sleep on it. Then decide.

What to verify yourself

Even with a good tool, check these by hand before you put money in:

• Key figures. Open the cited filing and confirm two or three of the numbers the AI used.
• Dates. Make sure the data is from the latest quarter and not last year's report.
• The ticker. Similar names and share classes are easy to mix up. Confirm you are looking at the right listing.
• Fees and costs. For an ETF (exchange-traded fund, a basket of investments that trades like a single share), check the annual cost on the fund's own page.
• Your own limits. How much you can afford to lose is not something an AI should decide.

Where Moneda fits

Moneda is an app (mobile, plus web), built in Europe, where research and investing sit in one place: an AI that does the analysis and shows its sources, then you buy with your approval. Most tools do one or the other. Moneda AI is built on Numa 1, Moneda's financial AI engine, and it is designed around the strengths above while keeping the decision with you.

Moneda AI researches stocks, ETFs, crypto and a few commodities through ETFs (gold, silver and oil), using company filings and live market data. It works figures out from the data and shows the source behind each one. It gives the bull and the bear case, scores a company's 7 Powers (a framework for judging a moat), checks quality of earnings, and produces the Moneda Ratio, a descriptive 1 to 7 scorecard. It describes; it does not rate stocks as buys.

You can describe the kind of company you are looking for in plain language and get ideas back, or narrow the list with the stock screener. The agent can search SEC filings and earnings-call transcripts, so you can check what a company actually reported. The agent can build an illustrative portfolio from your goal, track it, tell you when something changes, and stress-test it against 2008, 2020 and 2022. You can benchmark a portfolio against gold, an index or any fund.

When you want to act, you can invest in the same app: US-listed shares and ETFs (including fractional shares) and crypto. Every trade is approved by you with your passkey, using your face or fingerprint. The agent never tells you to buy or sell.

If you already use Claude or ChatGPT, Moneda MCP connects them to your Moneda account with the permissions you choose, so you decide what each one can see and do. Payments are still approved in the app. A REST API and CLI are available on Premium and Ultra.

On accuracy, Moneda's own tests in October 2026 measured groundedness at 97.9% and claim verifiability at 92%. On the public set of the Vals AI Finance Agent Benchmark v2, Moneda scored 80%, run and graded by Moneda. For context, Vals AI's top published score is 65.4%, but that is on its private set, which is a different set of questions, so the two are not directly comparable. Moneda also scored 95.6% on FLARE-CFA. These are Moneda's own tests, not independent audits.

Honest limitation: Moneda invests in US-listed shares and ETFs only, so no European-listed stocks or funds.

Paid plans include commission-free trades each month (1, 5 or 10, depending on plan).

A note on general AI assistants

General assistants like ChatGPT, Claude and Gemini are very good at explaining concepts, summarising a document you paste in, or helping you think through a decision. For quick definitions or "explain this like I'm new", they are a fine place to start.

Where they need more care is live numbers. Unless the assistant is connected to real market data and shows its sources, treat any price or ratio it gives you as a draft to check. Connecting an assistant to a data source, through tools like MCP, closes much of that gap.

FAQ

Can AI predict which stocks will go up?

No. AI can analyse a company and spot patterns in its data, but no model reliably predicts future prices. Be wary of any tool that claims otherwise.

Is it safe to trust numbers from an AI chatbot?

Trust them only when they come with a source you can open. General chatbots sometimes produce figures that look right but are not in any filing. Always check a few key numbers yourself.

Should I let an AI trade for me automatically?

That depends on your comfort with risk, but most people are better off approving each trade themselves. In Moneda, every trade needs your passkey, so nothing happens without you.

What is the best first question to ask an AI about a stock?

"How does this company make money, and what could go wrong?" If you cannot answer both in your own words afterwards, keep researching.

Do I need to understand finance to use AI for research?

No. A good tool explains terms as it goes. Ask it to define anything unfamiliar, and to explain its reasoning step by step.

Can AI replace a financial adviser?

Not for personal advice. An AI does not know your full situation or carry responsibility for your outcome. It is a research tool, and it works best alongside your own judgement.

This guide is for information only and is not investment advice. Investing involves risk, and you can lose money.