Most people treat AI like a smarter search bar. Ask a question, copy the answer, move on.
That is the slowest way to use it.
The leverage shows up when you point it at the same primary sources an analyst already reads: the 10-K, the earnings call, the comp set, the model you already built. Not "what do you think of this stock." Structured work on documents you can verify line by line.
4 moves we use inside a real research workflow:
1. Feed it the actual 10-K, not your memory of it. Ask it to pull the risk factors that changed versus last year and quote the exact language. You verify against the filing.
2. Turn an earnings call transcript into a guidance-vs-actual table. What management promised last quarter, what they delivered, where the wording got softer.
3. Build a comp screen by definition, not vibes. Same multiple, same period, same accounting basis, so you are comparing like for like before you trust the spread.
4. Stress-test your own assumptions. Have it argue the bear case against your model so the weak inputs surface before the market finds them.
None of this replaces your judgment. It removes the grunt work between you and the evidence.
Educational only. No buy, sell, or hold. No targets, no returns. Verify everything against the primary source.
Want the exact prompts plus the workflow checklist behind these 4 moves?
Comment ANALYST and we will send it over.