Every bad AI session starts with the same hidden cost.
You have to introduce yourself again.
Your role. Your audience. Your tone. Your standards. The format you always want. The things you never want.
I call this the reintroduction tax.
It is the work you do before the work.
Most people blame the model when the answer feels generic. In many cases, the model was never told what specific means for you.
The shift
This is why personalization is becoming one of the most important AI skills.
OpenAI now describes ChatGPT as something that works better when treated less like a search box and more like a collaborator. Its Academy guide points users toward Custom Instructions and Memory. Its newer memory controls show which past chats, files, saved memories, or connected apps shaped a response.
Claude has account instructions, project instructions, and skills. Gemini has personal context, instructions for every chat, and Gems with optional knowledge files.
The pattern is clear.
The best AI assistants are no longer just prompt boxes. They now have a setup layer.
Most users still leave that layer blank.
What personalization actually does
Personalization does not make the model smarter in the abstract.
It makes the model start from a better place.
A useful setup layer does three jobs.
1. It sets identity. What you do, who you serve, what your work looks like.
2. It sets standards. What good output means for you, not for a generic user.
3. It sets boundaries. When to ask, when to challenge, when to cite, and when to stop.
That is the difference between asking a stranger for help and working with someone who knows how you operate.
Custom instructions and memory are not the same thing.
Custom instructions are deliberate rules. Memory is accumulated context. Use instructions for stable preferences. Review memory so stale details do not quietly shape future answers.
This is the first step in building an AI second brain.
Not because the tool remembers everything.
Because it stops starting from zero.
1. OpenAI: Personalizing ChatGPT
The cleanest explanation of why context and direction change output quality. Use this for Custom Instructions and Memory basics.
Read it here
2. OpenAI: GPT-5.5 Instant personalization controls
The latest signal. ChatGPT is getting better at using past chats, files, and connected Gmail, with memory sources that show what shaped the response.
Read it here
3. Claude and Gemini personalization docs
Claude now separates profile instructions, project instructions, and skills. Gemini has personal context, instructions, and Gems.
Claude guide · Gemini instructions · Gemini Gems
Paste this into ChatGPT Custom Instructions, Claude instructions, Gemini personal context, or the baseline instructions for a Project or Gem.
Use this as my default working profile. Role: [I am...] Recurring work: [I regularly use AI for...] Audience: [My work is usually for...] Standards: Good output for me means: 1. [standard] 2. [standard] 3. [standard] Default response style: - Be direct. - Use concise structure. - Ask clarifying questions when the task, audience, or success criteria are unclear. - Challenge weak assumptions before drafting. - Prefer practical examples over abstract advice. Output formats I use often: - [format 1] - [format 2] - [format 3] Boundaries: Do not invent facts. Flag uncertainty. Tell me when source checking is needed. Do not use polished wording to hide a weak idea. First task: Summarize what you now know about how to help me. Then list what is still missing.
The last line matters.
It forces the assistant to show the gaps in its context before you rely on the answer.
What do you still explain to AI every week?
Your role? Your audience? Your product? Your tone? Your formatting rules? Your definition of good?
Reply with the one thing you repeat most. That belongs in your setup layer.
Your second brain starts with a setup layer.
I wrote the full AI Second Brain guide at tminusai.com/blog/ai-second-brain-chatgpt. It shows the Capture, Structure, Retrieve system.
The free Prompting Hub at tminusai.com/power-guides gives you 19 laws and 32 cheat codes. Free, no email required.
For the full system, Power Guide Pro gives you personalization, Projects, weekly workflows, and model choice. One purchase, permanent reference. Euro 59.99.
More next week.
Kapish
Enterprise AI Architect · T-Minus AI