I tested the same finance research task across Perplexity, ChatGPT, and Claude. I expected a winner. I got a workflow instead. The tools are not competitors. They are stages. Save the one-pager and route your next task by stage, not by habit. Educational only. Not investment advice.
All briefs
Wall Street Prompt · LinkedIn one-page
3 AI Tools, 3 Stages of Research
LinkedIn one-page
I ran the same finance research task through Perplexity, ChatGPT, and Claude. The lesson was not which one wins. It was where each one earns its keep.
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: Matt Wolfe / FutureTools
Post content
Dense one-page content — sections, tables and frameworks
The Test: One Task, Three Tools
I gave all three the same job: research a mid-cap industrials name for an IC memo. Same prompt, same 90 minutes, same source discipline. I scored each on what stage of the workflow it actually moved forward.
| Tool | Where It Won | Where It Broke |
|---|---|---|
| Perplexity | Fast sourcing, live citations, finding primary docs | Shallow synthesis, no judgment on what matters |
| ChatGPT | Structuring the memo, drafting comps tables, reframing | Confident on stale data, weak on linked sources |
| Claude | Long-doc reasoning over 10-Ks and transcripts, nuance | Slower to surface fresh news, needs you to feed it docs |
Why it matters: A tool is not good or bad. It is good or bad at a stage. The mistake is asking one tool to run the whole pipeline.
Map Each Tool To A Stage
Research is not one job. It is four. Each stage rewards a different capability. Match the tool to the stage and the work compounds.
| Stage | Goal | Best Tool | What You Actually Do |
|---|---|---|---|
| 1. Sourcing | Find primary docs fast | Perplexity | Pull 10-K, last 4 transcripts, recent filings, citations |
| 2. Reading | Reason over long documents | Claude | Feed full transcripts, ask for management tone shifts |
| 3. Building | Structure comps and memo | ChatGPT | Draft comps table, IC memo skeleton, bull/bear framing |
| 4. Stress test | Attack your own thesis | Any (you lead) | Force the bear case, list what would break it |
Heads up: Stage 4 is not a tool task. It is a judgment task. AI builds the inputs. You decide what the inputs mean.
The Common Mistake And The Failure Mode
Most people pick one tool and force it through every stage. Here is what that costs you.
| Mistake | Failure Mode | Disciplined Fix |
|---|---|---|
| One tool for everything | You inherit its single blind spot on every memo | Route each stage to the tool built for it |
| Trusting unsourced synthesis | Confident numbers with no filing behind them | Demand a primary-source link for every figure |
| Skipping the read stage | You quote a summary, not the actual 10-K language | Paste the real document, read the management tone yourself |
| Accepting the first draft | A clean memo with an unexamined thesis | Run the bear case before you size anything |
Why it matters: A polished memo built on a stale or unsourced number is more dangerous than no memo. It feels like work product.
The Verification Layer (Non-Negotiable)
Every figure that reaches a comps table or an IC memo passes one checklist. No exceptions. This is the line between research and guessing.
- Trace it: every number links to a filing, transcript, or release.
- Date it: confirm the figure is from the latest reported period, not a model memory.
- Cross-check: if two tools disagree on a number, the filing breaks the tie.
- Read the footnote: adjusted vs reported, one-time items, share count basis.
- Own it: if you cannot defend the number from the primary source, it does not go in the memo.
Key: AI does not get a vote on what is true. The filing does. The tools assemble evidence. You adjudicate it.
The Workflow, Start To Finish
Same task, run the right way. This is the sequence I landed on after the test.
| Step | Tool | Output You Keep |
|---|---|---|
| Pull sources | Perplexity | Linked list of 10-K, transcripts, recent filings |
| Reason over docs | Claude | Margin trend, tone shifts, segment notes with quotes |
| Build the memo | ChatGPT | Comps table, bull/bear, IC memo draft |
| Verify every number | You + filings | Sourced, dated, cross-checked figures |
| Stress the thesis | You | Written bear case and the triggers that break it |
Why it matters: The handoffs are the edge. Each tool starts with cleaner inputs because the prior stage did its job. Human judges. AI builds.
Caption
LinkedIn post copy
Visual design notes
- Near-black forest-green background (#0B0F0D to #0F1512), ONE teal-green accent (#2FE3B0) used only for the title keyword, the subtitle banner, table header rules, and the footer divider. No second accent color anywhere.
- Heavy condensed sans headline (Anton or Archivo Black) for the title, left-aligned, with '3 Stages' in the teal accent. Subtitle sits in a thin teal-outlined banner bar directly beneath, light weight, sentence case.
- Dense 2 to 4 column tables with hairline (1px, 12% white) row dividers, generous left padding, no vertical gridlines. Table headers in teal small-caps, body text in off-white (#E8EFEA) at high density. Tools column in semibold.
- Each section heading in condensed uppercase with a short 40px teal tick mark to its left as a consistent anchor. Intro line in muted gray (#9AA8A0), one size down from body.
- 'Why it matters' / 'Heads up' / 'Key' notes set in a left-border callout (2px teal left rule, slightly indented, italic-adjacent weight) so they read as the teacher voice throughout.
- Stage-to-tool mapping (Section 2) and the final workflow (Section 5) can optionally render as a thin left-to-right arrow strip beneath the table: Sourcing -> Reading -> Building -> Stress test, to reinforce the pipeline mental model.
- Strict left alignment for all text. Comfortable vertical rhythm between sections (32-40px). Footer pinned to bottom with a full-width teal hairline above it. Margins 80px sides on the 1080x1350 canvas.
- WSP wordmark top-left small-caps, page format 1080x1350 portrait. Keep total color count to background + off-white + one teal accent + muted gray. Restraint signals finance, not hype.
Production checklist
- ☐Design the 1080x1350 one-pager in the WSP template: near-black forest background, single teal accent, condensed headline with '3 Stages' emphasized, teal subtitle banner bar.
- ☐Build all five sections with left tick-mark headings; render the four comparison/framework tables (Test results, Stage map, Mistake-Failure-Fix, Workflow) as dense left-aligned tables with teal small-caps headers and hairline row dividers.
- ☐Set the bullet list in Section 4 (Verification Layer) and style every 'Why it matters / Heads up / Key' note as a teal left-border callout for the teacher voice.
- ☐Add header WSP wordmark (top-left) and the full footer line with a teal hairline divider above it; confirm one accent color only and strict left alignment throughout.
- ☐Proof for voice and compliance: short declarative sentences, finance-grounded, no em dashes, no buy/sell/hold, no targets or returns, includes educational-only framing.
- ☐Export final asset as PNG at 1080x1350 and a matching PDF version for DM distribution; verify text legibility at LinkedIn feed scale.
Follow Dave Wang · Wall Street Prompt · For more AI investing workflows · COMMENT & REPOST
CTA
Save this and route your next research task by stage, not by habit. Comment "STAGES" and I will share the verification checklist as a one-pager.