How to Read a 10-K With AI in 3 Minutes (SEC EDGAR, No Code)
Every US public company files its annual report (10-K) on SEC EDGAR for free. Download it, drop it into ChatGPT or Claude, and get the business model, three years of numbers, and the risk factors in about 3 minutes. Here is the exact routine, the copy-paste prompt, and the guardrails that stop the AI from making numbers up.
A 10-K is the annual report every US public company is required to file with the SEC: what the business does, audited financials, and a list of risks the company itself wrote down. Think of it as the company's sworn statement, as opposed to its marketing. It is the closest thing to ground truth a public company produces, and it sits free on SEC EDGAR. The catch is a couple hundred pages of legal language, which is exactly the kind of reading a chatbot does fast and you do not. The part the chatbot is bad at, knowing what to trust, stays with you. Here is the whole routine.
10-K, 10-Q, 8-K: which one to grab
The 10-K has siblings. The 10-Q is the quarterly version, unaudited but fresher, filed three times a year. The 8-K is the "something just happened" filing, published between reports when there is news that cannot wait. For a first look at any company, the 10-K is the one to grab; the other two are how you keep up afterward.
Step 1: pull the filing from EDGAR
Skip the finance apps and go to the source. Everything on EDGAR is free and there is no login.
- 01Search the company
sec.gov EDGAR search, by name or ticker
- 02Filter to 10-K
Pick the latest annual report
- 03Open the document
The main 10-K, not the exhibits
- 04Save it
Download or print to PDF
Two practical notes. First, the filing page lists a pile of exhibits; the document you want is the main 10-K itself. Second, if you only care about one topic, say a company's AI spending, EDGAR's full-text search lets you search phrases across filings, which beats reading anything cover to cover.
Step 2: drop it into the chatbot as-is
Do not clean the file up. Upload the whole thing to ChatGPT or Claude and let the model deal with the formatting. Size is a non-issue: ChatGPT takes files up to 512MB (about 2 million tokens of text) and Claude takes 30MB per file, up to 20 files per conversation. A typical 10-K is a few megabytes. Claude will also read the tables and charts inside a PDF if it is under 100 pages; longer than that, it reads the text.
One caveat if you are on a free plan: upload counts are capped (free ChatGPT allows only a few files per day), so batch your questions into one conversation per company.
The 3-minute prompt I actually use
Here is my saved prompt. Paste it right after the upload.
You are an analyst reading this 10-K. Using ONLY this filing, give me:
1. How the company makes money (top 3 segments by revenue share).
2. Revenue, operating income, and net income for the last 3 fiscal years,
in a table, exactly as reported.
3. Did results improve or worsen vs. the prior year, and what reason does
the company itself give?
4. The 3 risk factors the company lists that would worry me most as a
shareholder.
5. One sentence on whether debt looks heavy relative to earnings.
If something is not in the filing, write "not in filing." Cite the item
or section each answer came from.
Run that on a real 10-K and the answer comes back in this shape (company and figures below are an illustrative example, not a real filing):
1. Revenue mix (example): Cloud services 54% > Devices 31% > Advertising 15% (Item 1, Business)
2. Last 3 fiscal years (example, $M)
| FY2023 | FY2024 | FY2025 | |
|---|---|---|---|
| Revenue | 41,200 | 45,900 | 51,300 |
3. Revenue up two years running; management attributes it to "continued enterprise cloud adoption" (Item 7, MD&A)
4. Risk factors: customer concentration, foreign-exchange exposure, data-center cost inflation (Item 1A)
5. Price target or fair value: not in filing
That last line is the point. The "not in filing" instruction is what keeps the model from inventing a price target, and the citation request ("Item 7, MD&A") makes your spot-check ten times faster.
What to ask, and what never to ask
The dividing line is simple: the filing's own contents versus the future.
- 3-year revenue and income trend
- Segment mix and customer concentration
- Risks the company itself listed
- "Will the stock go up?"
- "What is a fair price target?"
- Comparisons to data not in the filing
FAQ
Where can I read a company 10-K for free?
On SEC EDGAR at sec.gov. Use the full-text search or the company search to find the filer, open the 10-K, and download or copy the document. EDGAR is the primary source; the summaries you see on finance apps are someone else's processed version of the same filing.
What is the difference between a 10-K and a 10-Q?
A 10-K is the annual report: audited numbers, the full business description, and risk factors, filed once a year. A 10-Q is the quarterly version: unaudited, thinner, but fresher, filed three times a year. Start with the 10-K to understand a company, then use 10-Qs to track how the year is going.
Can ChatGPT or Claude read a whole 10-K?
Yes. ChatGPT accepts files up to 512MB (about 2 million tokens of text) and Claude accepts up to 30MB per file, 20 files per chat. A typical 10-K lands well inside both limits. Claude also reads tables and charts inside PDFs under 100 pages. The constraint is not size; it is that you should only ask about what is actually in the filing.
Should I trust the numbers the AI pulls from a filing?
Treat them as a fast draft, not a final answer. AI can garble a figure from a long table or invent an answer when you ask about something the filing does not contain, like a price target. Before any money decision, check the two or three numbers that matter against the filing itself.
My honest take
I use this as a screening filter, not a verdict. Ten companies on a watchlist means ten 10-Ks nobody has time to read; three minutes each with AI tells me which two deserve a real read. I never trade off the summary alone. The summary is a map of where to look closer, and that is worth a lot on its own.