AI gave you a clean summary of the 10-K. You felt finished. You were not. A summary restates what a document says. Research tests whether it is true. Educational only. Not investment advice.
All briefs
Wall Street Prompt · LinkedIn one-page
AI Summary Is Not AI Research
LinkedIn one-page
A summary tells you what a document says. Research tells you whether it is true.
LinkedIn1080 × 1350
References & validationReference missing
No specific creator reference resolved. Add an observed post before production.
Reference creator: dataAlpha.ai / Daloopa
No reference links resolved.
Post content
Dense one-page content — sections, tables and frameworks
The Clean Output Trap
A summary reads well. That is exactly why it fools you. Polish is not proof. The model compresses a 10-K into five tidy bullets and your brain registers closure. You feel done. You are not done. You have read a description, not built a case.
- Clean prose signals confidence the underlying analysis has not earned.
- A summary hides what it dropped. The risk factor it skipped does not show up as a blank line.
- Fluency is the disguise. The smoother the paragraph, the less you check it.
Why it matters: position sizing off a summary means you are sizing off the model's editing choices, not the company's numbers.
Summary vs Research: The Tie-Out
Same input. Different job. One restates. One interrogates.
| Dimension | AI Summary | AI-Assisted Research |
|---|---|---|
| Goal | Restate what the document says | Test whether it holds up |
| Source link | None or implied | Every claim tied to a page, line, or filing |
| Numbers | Quoted as given | Recomputed and reconciled to comps |
| Contradictions | Smoothed over | Surfaced and forced to a resolution |
| Output | Bullets you trust | Evidence you can defend in an IC memo |
| Failure mode | Confident and wrong | Slow but auditable |
Key: if you cannot click from a sentence back to its source, it is a summary wearing a research costume.
The Three Disciplines a Summary Skips
Real research is not a longer summary. It is three jobs a summary never does.
| Discipline | The Question | What You Actually Do |
|---|---|---|
| Source tie-out | Where exactly does this come from? | Map each claim to a filing page, transcript line, or data field |
| Verification | Do the numbers reconcile? | Recompute margins, growth, multiples against the raw 10-K and comps |
| Challenge | What would make this wrong? | Force a counter-case from short reports, prior calls, and missing context |
Why it matters: skip tie-out and you cannot trace an error. Skip verification and you inherit the model's math. Skip challenge and you only hear the bull case.
The Disciplined Workflow
Turn a raw summary into something you can underwrite. Five steps, no shortcuts.
| Step | Action | Output |
|---|---|---|
| 1. Extract | Pull claims, not prose. One line each. | A claim list with no narrative |
| 2. Tie out | Attach a source to every claim. | Each claim linked to filing or transcript |
| 3. Recompute | Rebuild key numbers from raw data. | Reconciled figures vs reported |
| 4. Challenge | Ask the model for the bear case and the missing context. | A documented counter-view |
| 5. Judge | You decide what survives. | An evidence set, not an opinion |
Human judges. AI builds. The model assembles the evidence. You make the call.
Red Flags You Are Reading a Summary
Heads up. These tells mean you have an answer with no audit trail.
- No page numbers, no line references, no filing dates.
- Every claim sounds equally certain. Real research has confidence gradients.
- Round numbers that never get recomputed against the source.
- Zero contradictions. No real company is that clean across a year of calls.
- It tells you the conclusion before it shows you the work.
Tip: prompt for the weakest claim in the analysis and the evidence against it. A summary cannot answer that. Research can.
Caption
LinkedIn post copy
Visual design notes
- Near-black forest background (#0B0F0D to #0E1512), ONE green accent (#1FBF75 / WSP teal-green) used only for the emphasized word, table header rules, and the subtitle banner. Everything else off-white (#E8EDEA) and muted gray (#8A9692).
- Heavy condensed headline (Anton or Druk Wide style) for the title, with 'Research' as the only word in the green accent. Title left-aligned, top of canvas, large enough to anchor the page.
- Subtitle as a thin teal banner bar directly under the title: full-width left-aligned strip, dark-green fill, off-white text. This frames the whole post.
- Four dense 2-3 column tables are the spine of the layout. Use thin 1px green top-border on each header row, alternating row tint (transparent vs +4% white) for scan-ability. Monospace or tabular figures for any numbers so columns align.
- Section headings in small-caps condensed, numbered or with a short green tick mark to the left. Keep generous vertical rhythm so the density reads as structured, not cramped.
- Add one small diagram in the 'Disciplined Workflow' section: a horizontal 5-node flow (Extract to Judge) with green connector arrows, the final 'Judge' node outlined in green to signal human control.
- 'Why it matters' / 'Key' / 'Tip' notes set apart in a left-border callout strip (2px green left rule, slightly indented), so the teacher framing is visually distinct from body copy.
- Footer locked to bottom margin, single line, muted gray with the WSP wordmark in green. Maintain consistent left margin with all section content for a clean editorial grid. Everything left-aligned, no centered text.
Production checklist
- ☐Design the 1080x1350 one-pager in the WSP template: near-black/forest bg, single green accent, condensed headline with 'Research' emphasized, teal subtitle banner.
- ☐Build the four tables (Summary vs Research, Three Disciplines, Disciplined Workflow, plus the red-flag bullet block) with aligned tabular figures, green header rules, and alternating row tint.
- ☐Add the 5-node Extract-to-Judge workflow diagram with green connectors and an outlined 'Judge' node to reinforce 'Human judges, AI builds.'
- ☐Style the 'Why it matters / Key / Tip' callouts with a 2px green left-border strip and consistent indentation.
- ☐Add header and footer branding: WSP wordmark in green, footer line locked to bottom margin, Dave Wang attribution.
- ☐Proof for voice and compliance: short declarative sentences, no em dashes, no buy/sell/hold, no targets, confirm 'Educational only' line is present in caption.
- ☐Export PNG at 1080x1350 for the LinkedIn feed and a PDF version for DM and carousel reuse.
Follow Dave Wang · Wall Street Prompt · For more AI investing workflows · COMMENT & REPOST
CTA
Save this before your next earnings season. Then comment SOURCE and I will share the tie-out prompt I use to turn a model summary into an auditable evidence set.