Millie Marconi

STOP WRITING BASIC PROMPTS AND EXPECTING LLMS TO DO A MIRACLE.

Most people never unlock even 20 percent of what LLMs can actually do. Someone spent 2 weeks testing every major model and finally found the 10 prompts that consistently produce insane results.

Here they are ↓

1/ Coding apps

Mega prompt you can use to turn any LLM into an expert programmer:

"
# ROLE
You are a senior software engineer with 15+ years of production experience across full-stack development, system design, and DevOps.

# TASK BREAKDOWN
For every coding request, structure

2/ Research analysis

I fed Gemini a 40-page research paper on transformer architectures.
Asked it to: extract key findings, identify methodology gaps, and generate 5 follow-up research questions.

Got back a synthesis that would've taken me 6 hours. Took 90 seconds.

The prompt:

3/ Writing complete technical docs

Most people use AI for drafting. Wrong approach.

I use it for complete documentation from scratch.

The prompt: "You're writing API documentation for developers. Include: authentication flow, all endpoints with request/response examples, error

4/ Data analysis and visualization

Uploaded a messy CSV with 10k rows of sales data to ChatGPT.

"Clean this data, identify trends by region and product category, flag anomalies, and write Python code to generate 3 visualizations that tell the story."

It caught a data entry

5/ Email automation at scale

Not just "write me an email."

"Draft 5 variations of a cold outreach email to SaaS founders. Each should: open with a specific observation about their company, position our solution around their current growth stage, include social proof, end with a

6/ Meeting notes to action items

I paste raw Zoom transcripts into Claude.

The prompt: "Extract: (1) all decisions made with owner and deadline, (2) open questions that need answers, (3) action items by person, (4) topics that need follow-up meetings. Format as a structured doc

7/ Content repurposing

Take one long-form piece, turn it into 10 different formats.

"Take this blog post and create: (1) Twitter thread with 8 tweets, (2) LinkedIn post, (3) 3 short TikTok scripts, (4) email newsletter version, (5) 5 pull quotes for graphics. Maintain core

8/ Code review and debugging

Paste broken code into Claude, add: "Review this code as a senior engineer doing a PR review. Identify: (1) bugs and why they'll cause issues, (2) performance bottlenecks, (3) security vulnerabilities, (4) readability improvements. Provide fixed

9/ Building complete automation workflows

The wildest one.

"Design an n8n workflow that: monitors my Gmail for invoices, extracts data using OCR, updates a Google Sheet, sends Slack notifications for amounts over $1000, and files receipts in Drive. Provide the complete JSON and

10/ SEO content clusters

This one prints money.

"Generate a complete SEO content cluster for 'AI automation tools.' Include: (1) pillar page outline with H2s and H3s, (2) 10 supporting articles with target keywords, (3) internal linking strategy, (4) meta descriptions for each.

AI makes content creation faster than ever, but it also makes guessing riskier than ever.

If you want to know what your audience will react to before you post, TestFeed gives you instant feedback from AI personas that think like your real users.

It’s the missing step between

I hope you've found this thread helpful.

Follow me @Yesterday_work_ for more.

Like/Repost the quote below if you can: