Thread by @RubenHssd

I bet 99% of people who use ChatGPT don't know how to set it up to make it 10x more useful.

They obsess over prompts, but prompts are only 20% of the equation.

Setup is 80%.

In this thread, I'll show you how to (actually) set up your ChatGPT:

→ First, take a look at how much ChatGPT knows about you.

Then, delete everything.

Go to Settings > Personalization > Manage Memories > Delete All Memories.

Most users have 6+ months of random, contradictory memories that actively hurt performance.

→ The 13-Question Framework.

Use this prompt with GPT-5: "Design my ChatGPT digital twin. Ask 13 items one at a time:

Identity, Role, Audience, Outputs, Personality, Voice, Formatting, Values, Projects, Goals, Preferences, Do-Not List, Privacy.

Target <1300 words total."

Why 13 questions?

Each captures a critical dimension: who you are (identity), what you do (expertise), who you serve (audience), what you create (outputs), how you think (personality).

Also how you communicate (voice), what matters (values), what you're building (goals).

Remember, ChatGPT memory snippets max out at 350 characters each.

→ After answering the 13 questions, insert prompt:

"Create 13 snippets I can copy/paste to update memory. Each is 350 characters max. Cover everything without redundancy."

→ Then: "Maximize each snippet to the character limit."

This forces ChatGPT to compress your identity efficiently.

Every character counts toward building your digital twin.

→ Installation Method

Open GPT-5, then copy/paste this 13 times with each snippet:

"UPDATE YOUR MEMORY WITH THE FOLLOWING SNIPPET, EXACTLY THESE WORDS: [snippet]".

This builds your digital twin systematically instead of randomly.

Once ChatGPT knows you well, it becomes biased toward your patterns.

Temporary chats = raw model without customization.

Use when you need: unbiased perspectives, testing new approaches, or ChatGPT to challenge rather than affirm your thinking.

→ One-Shot Prompting Architecture

This is the 3-block structure that eliminates back-and-forth and cuts conversation time by 60% while improving quality:

Real power users spend 90% time on setup, 10% on prompts.

Back-and-forth conversations are the sign of poor setup, not sophisticated use.

Your goal is to provide Role + Example + Request in ONE message.

If you're iterating multiple times, your memory setup is incomplete.

Use temporary chat when you don't want AI remembering anything to prevent context bleeding into future responses.

Perfect for one-off questions where past context would muddy the answer.

→ When Few-Shot Beats One-Shot?

Rare cases: varied formats, nuanced categories, high-risk tasks.

Rule of thumb: Zero-shot = speed. One-shot = clarity + single format (80% of cases). Few-shot = complexity + variety (20% of cases).

Most people overuse few-shot and waste time.

Your ChatGPT setup determines 80% of output quality. Your individual prompts determine 20%.

Everyone teaches prompts.

Almost nobody teaches setup.

This is why most people get mediocre results despite "knowing AI", they're optimizing the wrong 20%.

After proper setup, lawyers use ChatGPT to draft contracts in their exact style.

Writers use it to capture trends and write briefs.

Researchers use it for deep search with context.

The difference? ChatGPT knows WHO they are before starting WHAT they want.

You deserve an assistant who understands your goals, voice, and preferences.

Setup first. Prompts second. Always.

Follow me @RubenHssd for more content like this in the future.