The record · as of
Can AI pick stocks? What 269 graded Claude picks show.
Not on this record so far: 33 of the 131 Claude picks that have closed since 1 October 2026 beat the S&P 500 over the same days (25%), and the average closed pick trailed it by 6.9 percentage points. The record is young: 138 picks are still open, 80 of them ahead of the S&P 500 so far, and none has yet run its full 63 trading sessions.
Closed
131
33 beat, 98 trailed
Beat the S&P 500
25%
of closed picks
Average vs S&P 500
−6.9 pp
per closed pick; median −5.9 pp
Still open
138
80 ahead, 48 behind, 10 level
Why the early results lean toward losers
A pick runs for up to 63 trading sessions. It can close sooner only two ways: through its kill line, or when the research cuts its conviction below medium. A close through the kill line means the stock fell to the price set in advance as the point where the thesis is wrong. So far 41 picks have closed through their kill line, and none of them beat the S&P 500; 90 closed on a conviction cut, and 33 of those beat it. No pick has reached its 63rd session yet; the first ones do on 31 December 2026, and until then the closed set is made only of these early exits while the picks that are working stay open.
| Why it closed | Closed | Beat | Average vs S&P 500 |
|---|---|---|---|
| Kill line | 41 | 0 of 41 | −17.3 pp |
| Conviction cut | 90 | 33 of 90 | −2.1 pp |
| Ran 63 sessions | 0 | — | — |
By conviction
The same record split by how strongly the research stood behind each pick when it opened.
| Conviction | Open | Closed | Beat | Average vs S&P 500 |
|---|---|---|---|---|
| Supreme | 0 | 1 | 0 of 1 | −0.6 pp |
| High | 3 | 6 | 1 of 6 | −4.6 pp |
| Medium | 135 | 124 | 32 of 124 | −7.0 pp |
What a kill line is
A kill line is the price, written down before the market decides, that would prove a thesis wrong: usually a daily or weekly close below a stated level. When a pick opens, its thesis, its conviction and its kill line are frozen at that close. If a later close breaks the line, the pick closes and is graded against the S&P 500 like any other.
All 269 picks on the record have a written thesis and a price kill line. Kill line in the glossary
How the record is kept
Claude, Anthropic's AI model, writes the research on each stock and states how strongly it stands behind the thesis. Every thesis stated at medium, high or supreme conviction becomes a pick, and every pick is graded against the S&P 500 (SPY, dividends included) over exactly the days it was open. The rule was written down on 1 October 2026, before any pick existed, and nothing is removed for how it turned out.
The rule in full · How the research is made · Other AI stock-picking experiments, compared
Common questions
- Can AI pick stocks that beat the market?
- Not on this record so far: 33 of the 131 Claude picks that have closed since 1 October 2026 beat the S&P 500 over the same days (25%), and the average closed pick trailed it by 6.9 percentage points. The record is young: 138 picks are still open, 80 of them ahead of the S&P 500 so far, and none has yet run its full 63 trading sessions.
- Why do the first closed picks lean toward losers?
- A pick closes before its 63rd trading session only two ways: through its kill line, or when the research cuts its conviction below medium. A close through the kill line means the stock fell to the price set in advance as the point where the thesis is wrong. So far 41 picks have closed through their kill line, and none of them beat the S&P 500; 90 closed on a conviction cut, and 33 of those beat it. No pick has reached its 63rd session yet; the first ones do on 31 December 2026, and until then the closed set is made only of these early exits while the picks that are working stay open.
- What is a kill line?
- The price, written down before the market decides, that would prove a thesis wrong, usually a daily or weekly close below a stated level. It is frozen when the pick opens. If a close breaks it, the pick closes and is graded against the S&P 500 like any other.
- Which AI makes these picks?
- Claude, Anthropic's AI model, writes the research on each stock and states how strongly it stands behind the thesis. Every thesis stated at medium conviction or higher becomes a pick at that day's close, under a rule fixed on 1 October 2026 before any pick existed.
- 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.