Wall Street Prompt · Brand

5 prompt patterns that turn AI into a research analyst (not a chatbot)

Instagram post adapted from a SPECIFIC captured source post — exact-source traceability, like our YouTube flow.

Instagram static

Validated — exact source linked@stics.ai

https://www.instagram.com/p/DYR4eFhiNfb/

Observed format: Single static image: curiosity hook on top, a numbered list of 5 "secret codes/shortcuts" in the middle, and a save-and-setup instruction at the bottom.

Observed hook: "Most people use Claude like a basic chatbot. But with the right shortcut codes, it can match your style, think deeper, build tools..." (the "you're using this wrong, here are the secret codes" pattern).

Adaptation: Kept the winning skeleton: contrarian hook ("most people use it like a basic X"), a numbered 5-item list, and a low-friction CTA. Changed everything else. The source sold made-up "shortcut codes" (/godmode, /ghost, OODA) that its own commenters flagged as fake; WSP replaces them with 5 genuine, source-grounded analyst prompts tied to 10-Ks, risk factors, earnings calls, and comps. The setup-and-save instruction became a comment-ANALYST lead mechanic. Added the explicit anti-hype framing ("there is no hidden /godmode"), the "ground it in a primary document and cite" principle, and a clear educational-only, no-advice disclaimer. No em dashes, no guru hype, no copied wording or claims.

On-image text

On-image text

TOP: Most people use AI like a search box for stocks.
MIDDLE: 5 prompts that make it read a 10-K like an analyst:
1. "Pull revenue by segment for the last 3 years and flag what changed."
2. "List every risk factor that is new vs last year's 10-K."
3. "Summarize the earnings call: guidance, margins, what management dodged."
4. "Build a comp table: this company vs 3 named peers, same metrics."
5. "Show the bear case and the bull case, citing the filing line."
BOTTOM: Educational only. Not advice. Always verify against the primary source.

Caption

Paste under the post

Most people type "is this stock good?" into AI and get a confident, useless answer.

The fix is not a secret code. There is no hidden "/godmode." What actually changes the output is giving the model a real analyst task and a real source to work from.

Here are 5 prompts we use against a single 10-K or earnings transcript:

1. Segment revenue: "Pull revenue by segment for the last 3 fiscal years and flag what changed." Forces it to work from the numbers, not vibes.

2. Risk delta: "List every risk factor that is new vs last year's 10-K." This is where the real story usually hides.

3. Call breakdown: "Summarize the earnings call into guidance, margins, and any question management avoided answering."

4. Comps: "Build a comparison table of this company against 3 named peers on the same 4 metrics." You name the peers and the metrics so it cannot drift.

5. Two-sided: "Write the bear case and the bull case, and cite the filing line for each claim." Kills one-sided confidence.

The rule underneath all 5: paste the actual filing or transcript and tell it to cite. AI without a source is guessing. AI grounded in a primary document is doing analysis you can check.

This is an educational research workflow, not financial advice. No buy, sell, or hold. Verify every number against the original filing before you trust it.

Comment ANALYST and we will send the full one-page prompt sheet (all 5, copy-paste ready).

Layout

Vertical 4:5 static. Near-black background. Top third: hook line in large bold Montserrat white, with the contrast word ("search box") in green. Middle: a clean numbered 1-5 stack, each item a short green index number followed by the prompt in white quotes inside a thin-bordered card. Bottom: small muted disclaimer line plus a green CTA chip reading "Comment ANALYST".

Design notes

WSP near-black (#0B0F0E) background with accent green (#16C784 / spectrum green) for numbers, the hook contrast word, and the CTA chip. Montserrat throughout: Bold/ExtraBold for hook and numbers, Medium for prompt text, Regular small caps for the disclaimer. Generous spacing between the 5 cards, thin 1px green-tinted borders on each prompt card. Keep it text-led and calm, no icons-as-decoration, no neon glow, no fake "terminal" gimmicks. Disclaimer in muted gray (#8A8F8C) so it reads as honest, not buried.

Why this works

The "you're using it wrong, here are the power moves" list is a proven high-save, high-comment pattern: it promises a quick capability upgrade and uses a numbered list that is easy to screenshot. It fits WSP because we can deliver the SAME structure with real, verifiable prompts instead of fake cheat codes, which directly answers the credibility problem the source post got hammered for in its own comments (multiple users called the codes "fake" and "made-up"). The comment-keyword CTA captures intent and feeds the list, matching the source's setup-and-save mechanic.

CTA

Comment-keyword mechanic: "Comment ANALYST and we will send the full one-page prompt sheet." Adapts the source's save-and-setup CTA into a comment-to-DM lead trigger.