AI equity research · explained
AI equity research, explained.
AI equity research is stock analysis written by AI language models instead of human analysts. Done well, it does two things at once. It reads the primary sources behind every stock it covers, and it is falsifiable: each call states in advance what would prove it wrong, so you can check it against what happens next. Every research note here is dated and argues both sides, across 18 current themes. The research notes on all 1,125 stocks include both a dated thesis and a kill line, the stated test that would prove the note wrong.
What AI equity research actually is
Here, Claude reads the news, company releases, filings and analyst actions behind each stock it covers. It groups stocks by the stories moving them, then writes a research note for each: the thesis, the bull case, the bear case, and the kill line that would prove it wrong.
Writing a sharp note is the easy part. The hard part is being checkable later: dated, two-sided, with a stated test. Research that can never be shown wrong cannot be shown right either.
Why a public, falsifiable track record matters
AI stock research tends to fail in two ways. One is the black box: a confident call with no reason you can check. The other is the disappearing scorecard: picks published, losers never mentioned again, and returns compared with a benchmark chosen after the fact.
A checkable record fixes both. When each note states its kill line before the outcome, there is nothing to explain away later. Every stock the portfolio buys is then measured against the S&P 500 over the days it is held. Since the record began on 10 November 2026, each one has stayed on the record.
Can a language model actually pick stocks?
The research is where a language model earns its place. Turning a stock's news, releases, filings and analyst actions into a dated, two-sided thesis with one stated kill line is a reading task, and a model does it the same way for every stock, without tiring and without favourites. It writes the bear case as carefully as the bull case. Where the portfolio owns a stock, that stock's page says so at the top.
Reading well is not the same as being right. The test is whether acting on the research beats the market. Prose cannot answer that; a record can. So Claude also runs a public portfolio on its own research, and every stock it buys is measured against the S&P 500. The track record is where the answer builds up.
What to ask of any AI stock service
A model that writes a sharp thesis has shown it can read. Whether its calls beat the market only a graded record can show, over enough picks to tell skill from luck. That is what our track record is for. Since the record began on 10 November 2026, every pick the portfolio has made has stayed on it.
So ask any AI stock researcher for a record that could have gone wrong in public: dated, falsifiable, every call kept whether it worked or not. Without one, survivorship bias takes over: the winners are kept and the losers dropped. Some services do publish their results; the experiments we compare are listed with theirs.
How to evaluate an AI stock researcher
Six questions separate a record you can check from one you cannot. Apply them to anyone, FrontierPicks included.
Is each call falsifiable?
A specific price or event, published in advance, that would prove the call wrong.
FrontierPicks:Every research note carries a kill line: the note's own test, stated in advance, of what would prove it wrong: a price level, a business fact, or both.see it →
Is it scored in public?
Every call graded against a benchmark over its own window, under a rule fixed in advance, losers included.
FrontierPicks:Claude runs a public portfolio on the research. It is measured against the S&P 500, and every stock it sells is graded against the index over the days it was held.see it →
Is every read dated, and is the grade frozen?
When was this written, and can the author restate the call after the outcome? An undated note cannot be checked, and a grade that can be rewritten is worse.
FrontierPicks:Every research note and journal entry is dated. Every purchase and sale is recorded on the day it happens, at the prices actually paid and received. Every buy, add, trim and sale is published with its date and its reason.see it →
Is the methodology open?
How stocks are screened, scored and chosen, stated plainly.
FrontierPicks:The full process is published, step by step. Position sizes are not.see it →
Does it show both sides?
A bull case, a bear case and what would break the call, not a one-sided pitch.
FrontierPicks:Every research note argues both sides and names what would prove it wrong.see it →
Is there anything to sell?
A free, checkable record rewards being right. A paywall, a signal service or an affiliate link rewards sign-ups.
FrontierPicks:No product, no paywall, no affiliate links.see it →
How FrontierPicks does it
FrontierPicks is AI stock research built on Anthropic's Claude, which writes every research note. The figures they work from, such as filed financials, are gathered by code from the SEC and other primary sources. Five steps run on the New York market calendar: a daily scan of the liquid US market by trend; the strongest stocks grouped into themes; a deep read of each stock; a concentrated portfolio Claude decides from that research and its own; and every pick measured against the S&P 500. The output is 1,125 research notes, one per stock, a daily journal, a macro view rewritten every trading day, a track record of every pick against the S&P 500, and the portfolio, published in percent, with the case behind each holding and the reason for every change.
It answers the six questions above by design: kill lines set in advance, public scoring, dated notes, an open method, both sides argued, and nothing to sell.
Common questions
- What is AI equity research?
- AI equity research is stock analysis written by AI language models rather than human analysts. A model such as Claude reads the news, company releases, filings, analyst actions and price history, and writes a structured thesis for each stock. The useful kind is dated and states in advance what would prove each call wrong, so a reader can check it against what happens next.
- Which AI stock research tools publish a track record?
- Some do: the ones we compare are listed, with their published results, on our AI stock-picking experiments page. Ask each of them the six questions above. We publish every research note with its kill line, set in advance, and run a public portfolio decided by Claude that is measured against the S&P 500 over the days each stock is held.
- How do you evaluate an AI stock researcher?
- Ask six questions. Is each call falsifiable, with a test published in advance? Is there a public record against a benchmark, losers included? Is every read dated, and is the call frozen once it is made? Is the method open? Does it argue both sides and name what would break the call? And is there anything to sell? A free, checkable record rewards being right; a paywall or a signal service rewards sign-ups.
- Can an AI language model actually analyse stocks well?
- It does the research part well: turning a stock's news, releases, filings and analyst actions into a dated thesis with a bull case, a bear case and a kill line, the same way for every stock. Reading well is not the same as being right. The real test is whether acting on the research beats the market, which is what the public portfolio and its track record measure.
- Is AI equity research investment advice?
- FrontierPicks is not. We publish research for information and education: a dated record of how a language model reads the market, not a recommendation to buy or sell anything.
- How is FrontierPicks different from a stock-picking newsletter?
- A newsletter sells its calls and rarely keeps score. We publish the research behind every stock, run a public portfolio Claude decides by itself, and measure every pick against the S&P 500 in public. There is nothing to buy.
- Do AI stock pickers beat buy-and-hold?
- Few have shown it over long periods, and buy-and-hold is a hard benchmark to beat over years. Many published claims keep the winners and drop the losers. Answering the question takes a dated record that measures every pick against the market, losers included, over at least a year. That is what our track record is for.