Most people open a 10-K and ask AI to summarize it. Wrong verb. Ask what changed, what matters, and what needs verification instead.
All briefs
Wall Street Prompt · LinkedIn one-page
Prompt Before You Read A 10-K
LinkedIn one-page
Do not ask AI to summarize a 10-K. Ask what changed, what matters, and what needs verification.
LinkedIn1080 × 1350
References & validationExact post missing
Reference creator resolved, but the exact source post was not captured. Capture the specific post (not just the profile) before treating as validated.
Reference creator: Ruben Hassid / Superhuman
Post content
Dense one-page content — sections, tables and frameworks
The Mistake Everyone Makes
Summarize is the wrong verb. A summary flattens the document. It strips the signal you actually came for. Here is how the lazy prompt fails.
| Lazy prompt | Failure mode | Disciplined prompt |
|---|---|---|
| Summarize this 10-K | Generic recap of the business model you already know | List what materially changed vs the prior year 10-K |
| What does this company do | Marketing copy from Item 1, zero edge | Flag new risk factors added or removed since last filing |
| Is this a good stock | Confident opinion with no basis | Extract figures I must verify against the financial statements |
| Give me the highlights | AI picks what sounds important, not what is | Surface language changes in MD&A versus prior period |
Why it matters: A 10-K is read for deltas and disclosures, not a plot summary. Ask the model to do the work a junior analyst does, not the work a press release already did.
The Three Questions Framework
Every 10-K read reduces to three questions. Structure the prompt around them. Run them in order.
| Question | What you are hunting | Where it lives in the filing |
|---|---|---|
| What changed | Year over year deltas in risk, language, segments, accounting | Risk Factors, MD&A, Notes, segment tables |
| What matters | Items that move the thesis: margins, debt, concentration, litigation | Income statement, balance sheet, liquidity section, Item 3 |
| What needs verification | Any number or claim before you trust it | Financial statements, footnotes, auditor report |
Key: AI proposes. You verify. The model never gets the last word on a number.
Copy-Ready Prompt Stack
Paste the filing text, then run these in sequence. Each builds on the last. Keep the source document attached so the model cites against it.
- 1. CHANGED: Compare this 10-K to the prior year. List added, removed, and reworded risk factors. Note any new accounting policies.
- 2. MATTERS: Identify the five items most likely to affect the investment thesis. For each, cite the section and the figure.
- 3. VERIFY: Output every quantitative claim you made as a checklist with source line items so I can confirm against the statements.
- 4. GAPS: Tell me what this filing does NOT disclose that a careful analyst would want to know.
- 5. QUESTIONS: Draft five questions I should bring to the next earnings call based on this filing.
Tip: Prompt 4 is the one most people skip. Absence of disclosure is often the signal.
Human vs AI: Who Owns What
Draw the line before you start. The model handles volume. You handle judgment. Confusing the two is how mistakes enter the IC memo.
| Task | AI builds | Human judges |
|---|---|---|
| Reading 200 pages fast | Yes | No |
| Spotting year over year language shifts | Yes | Confirms relevance |
| Deciding what is material | Drafts a list | Final call |
| Trusting a reported number | Never | Always verifies to source |
| Position sizing and risk | No | Yes, fully |
Human judges. AI builds. The 10-K read is leverage on your reading speed, not a replacement for your reasoning.
Verification Discipline
A model will state a wrong number with full confidence. Build the check into the workflow, not into your hope.
| Claim type | Risk | Verification step |
|---|---|---|
| Revenue or margin figure | Hallucinated or mismatched period | Trace to the income statement, confirm the period |
| Debt or liquidity stat | Net vs gross confusion | Check balance sheet and the liquidity footnote |
| Risk factor characterization | Overstated or softened | Read the actual language, not the paraphrase |
| Year over year delta | Compared against wrong base year | Confirm both filings are the years you think |
Heads up: If you cannot trace a figure to a line item, the figure does not exist yet. Do not put it in the memo.
Caption
LinkedIn post copy
Visual design notes
- Near-black forest green background (#0B0F0D base) with a single teal-green accent (#1FB07A) used only for the titleKeyword, the subtitle banner, and table header rows. No second accent color.
- Heavy condensed sans headline (Anton or Archivo Black) for the title, left-aligned. titleKeyword 'Before' filled in the green accent while the rest stays off-white.
- Subtitle bar: full-width teal-on-dark banner directly under the headline, single line, sentence case, generous letter spacing.
- Dense 2 and 3 column tables with thin 1px green hairline dividers, alternating row tint at 4% opacity, monospace or tabular figures for any numbers, tight row height for density.
- Section headings in small-caps green label style with a short rule beneath; body and table text in a clean grotesque (Inter or Söhne) at high contrast off-white.
- Insert a small three-node diagram for the Three Questions Framework: CHANGED to MATTERS to VERIFY as connected pills, left-aligned, accent arrows.
- Footer locked to bottom with a thin top rule; WSP wordmark small-caps left, page treated as a single tall 1080x1350 column with consistent 64px side margins.
Production checklist
- ☐Design the 1080x1350 one-pager in the WSP template: near-black forest bg, single green accent, condensed headline with 'Before' emphasized, teal subtitle banner.
- ☐Build the five sections with the four tables (Mistake, Three Questions, Human vs AI, Verification) plus the copy-ready prompt stack as a numbered list; use tabular figures and green hairline dividers.
- ☐Add the CHANGED to MATTERS to VERIFY three-node diagram between sections two and three.
- ☐Add header and footer branding: WSP wordmark, Dave Wang attribution, follow and repost line.
- ☐Proof for voice and compliance: no buy/sell/hold, no targets, no em dashes, includes 'Educational only. Not investment advice.'
- ☐Export PNG at 1080x1350 for the feed and a PDF version for DM delivery of the full prompt stack.
Follow Dave Wang · Wall Street Prompt · For more AI investing workflows · COMMENT & REPOST
CTA
Save this prompt stack for your next 10-K. Comment CHANGED and I will send the full copy-paste version.