AI STOCK SCREENER
An AI stock screener that reads the filings
An AI stock screener is a research tool that filters a stock universe with quantitative criteria, then uses AI to evaluate what the numbers alone can't. RockTurners applies that pattern to primary sources: fixed screens narrow ~4,000 U.S.-listed companies, and an AI evaluator reads the survivors' SEC filings — returning a verdict on fit to a published investing framework, every claim cited, every gap named, never an action to take.
UPDATED July 24, 2026
THE PATTERN
Two stages, like any screener — plus the reading most screeners skip
Stage 1 is familiar: quantitative filters — return on capital, balance-sheet strength, payout coverage, whatever the strategy's published framework requires — run against the SEC EDGAR operating universe, identically every time. Most screeners stop there and hand back a list.
Stage 2 is the part that makes it an AI stock screener worth the name: an evaluator reads the survivors' actual filings — annual and quarterly reports, proxy statements, insider forms — against a fixed rubric, and returns one of three verdicts on framework fit: Gemstone, Geode, or Sediment. The full pipeline is on How it works; a complete sample verdict is dissected on Anatomy of a verdict.
THE DIFFERENCE
How RockTurners differs from a typical AI screener
| Feature | Typical AI screener | RockTurners |
|---|---|---|
| Evidence base | Aggregated data feeds and model memory | SEC EDGAR primary filings, read per evaluation |
| Output | A score or rating | A cited verdict on fit to a published framework |
| Sources shown | Rarely, and seldom per claim | Every factual claim names its form, section, and date |
| Missing data | Usually invisible in the result | Named unknowns — enough of them cap the verdict |
| Action language | Ratings and action calls | None. Verdicts describe framework fit, not what to do |
Feature descriptions, not performance claims — how the output is built, so you can decide whether the construction earns your trust. The full case is on Why trust the output.
FAIR QUESTIONS
How is this different from a traditional stock screener?
A traditional screener stops at the numbers: it filters on reported metrics and returns a list. That list is where the real work begins — each name is a 200-page annual report to read. RockTurners' second stage does that reading, the same way every time, and shows its sources so you can check any claim against the filing yourself.
Can AI actually read SEC filings?
Only with discipline. The evaluator is given targeted sections of the filings — its citation vocabulary — and may only cite what it was given. Metrics are pre-computed rather than trusted to the model's memory, and anything the filings don't disclose is marked unknown instead of filled in. When the evidence can't support a confident verdict, the honest output is a lower one.
Is the output investment advice?
No. RockTurners publishes impersonal, general-circulation research: the same methodology, criteria, and scoring applied identically to every company, with no knowledge of who is reading. Verdicts describe how well filings fit a framework. What to do with that reading is between you and a licensed financial professional. More on the FAQ and the disclaimer.