The record · as of
Can AI pick stocks? A public record of Claude's picks.
Too early to say. After 4 trading days, Claude's portfolio is +0.4% since 10 November 2026 against the S&P 500's +0.2%: 0.2 percentage points ahead. What would settle it: a gap wider than luck alone produces, measured from the record's own daily moves once it has 21 trading days, checked after every close from the record's 252nd trading day, about a year in. Until then this page reports the gap and its date.
The portfolio against the S&P 500
The whole question in one picture. Both lines start at 100 just before the portfolio's first session and move with every daily close. The gap between them reflects two things: the stocks Claude picked, and how much of the portfolio was invested (about 60% at the 13 November 2026 close). While part of it is in cash, the portfolio moves less than the S&P 500 in both directions.
Show the daily figures
| Session | Portfolio | S&P 500 |
|---|---|---|
| 100.4 | 100.2 | |
| 99.9 | 100.6 | |
| 99.8 | 100.9 | |
| 100.0 | 99.6 | |
| Start | 100.0 | 100.0 |
What is being tested
One thing: whether an AI model, deciding alone, can choose stocks that do better than simply owning the index. Claude decides every purchase and sale. It writes down why it bought and what facts about the business would end the case, and all of it is published. Its reasons to sell are fixed in advance; among them are a new fact that ends the case and research that finds the stock fully priced. There is no stop-loss and no preset price target, and no trading rule stands in for its judgment.
The comparison is the S&P 500 (SPY, dividends included) over exactly the same days, for the portfolio as a whole and for every stock it sells. Since the record began on 10 November 2026, every pick has stayed on it, gains and losses alike.
How to read an early record
Slowly. The portfolio is meant to hold most of its few companies for quarters or years, so in its first months few picks have been sold, and the ones sold first are not a fair sample. Two kinds of sale come early. A shorter trade, bought for a dated event such as a results release, is sold once that event has passed: that is its plan, not a verdict. A position meant for quarters is sold early mostly when its case broke. The long-term winners, if there are any, are still open.
Every pick on the track record · How Claude decides · Other AI stock-picking experiments, compared
Common questions
- Can AI pick stocks that beat the market?
- Too early to say. After 4 trading days, Claude's portfolio is +0.4% since 10 November 2026 against the S&P 500's +0.2%: 0.2 percentage points ahead. What would settle it: a gap wider than luck alone produces, measured from the record's own daily moves once it has 21 trading days, checked after every close from the record's 252nd trading day, about a year in. Until then this page reports the gap and its date.
- Which AI makes these picks?
- Claude, Anthropic's AI model, decides every purchase and sale in the portfolio by itself, from company filings, results, news and its own research notes. It will rank every US-listed common stock and ADR that trades at least $5 million a day each week for the portfolio, on the second-to-last full trading day of each week, and research the best-ranked in depth. The first weekly ranking has not run yet. No person and no trading rule picks the stocks. It keeps to fixed limits, among them no borrowing, no short sales, no funds, at most twenty companies and no stock bought above 25% of the portfolio.
- How is a pick graded?
- A pick is a stock the portfolio bought. When the portfolio sells it, its return, from the prices it was bought and sold at, is compared with the S&P 500's over exactly the same days. More than the index: it beat. Otherwise: it trailed. Since the record began on 10 November 2026, no pick has been removed for how it turned out.
- How long before the record means anything?
- Longer than a few months. The portfolio is meant to hold most companies for quarters or years, so the first sales are a small and unrepresentative sample; its shorter trades close sooner, but they are few. The line that matters is the whole portfolio against the S&P 500, and the first answer on this page says what would settle it.
- Is this investment advice?
- No. It is a public record of how an AI model's stock picks have done against the S&P 500, published for research and education. It is not a recommendation to buy or sell anything.