How to prep for earnings with Claude or ChatGPT (using free sources)

Updated 15 October 2026 · 8 min read

AI is good at reading a 40-page earnings release fast. It is bad at knowing which release is current, keeping "adjusted" and GAAP numbers apart, and admitting when it doesn't have a number. Here is a workflow that plays to the strengths and guards against the weaknesses.

Step 1: Get the right documents (free)

Go to SEC EDGAR search and type the ticker.
Filter to form 8-K. The earnings release is the 8-K that lists Item 2.02 ("Results of Operations"). The release itself is usually Exhibit 99.1; open the filing's file list to find it. (SEC: How to read an 8-K)
Open every 8-K filed after that one. Companies often update guidance between reports, after a deal, a charge or a settlement.
Optional: the call transcript or prepared remarks from the company's investor-relations site, consensus numbers you found, and option prices if you want an implied move.

Real example of why step 3 matters. Johnson & Johnson reported Q2 2026 on 15 July. Two weeks later, in an 8-K filed 29 July 2026, it lowered 2026 adjusted EPS guidance from $11.60–$11.75 to $10.96–$11.11 because of two transactions (about $0.64). An AI summary built only from the Q2 release would have used guidance that was no longer current.

Step 2: Paste with a prompt that forces sources

Paste the release (and any later 8-Ks) into Claude or ChatGPT with instructions like these:

You are preparing a facts-only pre-earnings brief for [TICKER].
Use ONLY the documents I paste below. For every number, add a source tag
like [8-K 2026-07-15 Ex.99.1]. If a number is not in the documents,
write "not found in sources". Never estimate.
Keep GAAP and non-GAAP ("adjusted", "core", "organic") numbers in
separate, labelled rows. Use the LATEST guidance and state its date.
Sections: what the company does; last quarter vs a year ago (table);
guidance now vs previous (table, with the company's stated reason);
filed events since the last report; 3-5 metrics to watch;
5 neutral questions to listen for on the call.
Do not give buy/sell views, price targets or predictions.
[PASTE DOCUMENTS HERE]

Step 3: After the report, compare against the right guidance

Paste the new release and the previous one, and ask for: headline numbers (GAAP and adjusted separately), guidance before this report (with its date) vs the new guidance, what changed vs last quarter, and quotes showing how management's language changed.

Two checks that catch most mistakes:

Which guidance is "previous"? J&J's Q3 release (13 Oct 2026) raised adjusted EPS guidance to $11.05–$11.15, which it described as up $0.06 at the midpoint to $11.10. That comparison is with the 29 July guidance (midpoint $11.04), not the 15 July release (midpoint $11.68).
GAAP vs adjusted. In the same quarter J&J's GAAP diluted EPS was $0.09 while adjusted EPS was $2.59. The reconciliation table shows a $5,519 million litigation-related item. An AI that mixes the two will tell a very different story.

Step 4: Tone and Q&A (only with a transcript)

Ask the AI to quote management's words this quarter next to last quarter's, and to list 3–5 analyst questions with a one-line answer each. Insist on quotes. Without a transcript, the honest answer is "transcript not provided"; a press release only contains prepared quotes.

Implied move (if you have option prices)

A common rough estimate of the move the options market is pricing in is the at-the-money straddle: (ATM call price + ATM put price) ÷ stock price, using the first expiry after the report. It describes the expected size of a move, not its direction. Give the AI the prices and ask it to show the inputs.

The 5 mistakes AI makes most with earnings

Stale guidance: it uses the last release and misses a mid-quarter 8-K.
Mixed bases: it compares adjusted EPS with GAAP EPS, or reported with constant-currency growth.
Invented consensus: it "remembers" analyst estimates. Only use consensus numbers you supply, with their source.
Wrong periods: some companies use 12- or 13-week quarters with odd end dates. Use the company's own period names.
Made-up precision: numbers that appear nowhere in the documents. Ask for a source tag on every figure and spot-check them.

Want this done in one command?

The Earnings Prep Pack packages this workflow as two skills for Claude, Cursor, Codex and ChatGPT. It fetches the filings from EDGAR, always checks for 8-Ks since the last report, tags every number and includes a checker that flags figures not found in the source. Prefer to do it by hand? The prompt above is free to use.

Educational content, not investment advice. Company figures are quoted from the linked SEC filings. See our disclaimer.