Your AI isn’t damaged. It’s simply does not know you but.

In case your work chats with ChatGPT, Claude, or Gemini are extra annoying than they’re useful, there’s normally a easy motive: it doesn’t know you but. 

Once I say your AI chatbot doesn’t “know” you, I don’t essentially imply that it must know your center title, your avenue deal with, or the ages of your youngsters.  

I’m speaking extra concerning the data {that a} good private assistant would wish: your high-level work function, your communication model, the instruments you employ every single day, and the “blockers” that hold you from getting stuff accomplished. 

Now, serving to your AI to get to know you is less complicated mentioned than accomplished. The place do you begin? What does it must know? It’s all too simple to wander into tangents when holding a get-to-know-you session with an AI, and if you happen to let it take the reins, it may flip into extra of a free-wheeling gabfest than a centered listening session.  

What your AI actually wants is to be “onboarded”—that’s, it must be built-in into your work life similar to a human assistant would possibly.

An onboarding session can take many types, and in a working atmosphere, it’s greatest to stay to the fundamentals. What do you do? What’s your function at work? What are your high priorities? What’s your work model? How do you deal with strain? And, most necessary of all, what obstacles are you going through? 

Alright, however what’s one of the best ways to onboard your AI? Do you have to simply begin free-associating with it about your work life? Yeah, no. 

As an alternative, attempt a trick borrowed from software program builders: a “profile-driven personalization” course of–and even “bootstrapping.” Briefly, it’s a setup course of that “initializes” the conduct of your AI and you’ll kick it off with a immediate. 

The immediate is on the backside of this story. Only a heads up, it’s massive. Drop it right into a contemporary chat and you’ll set off a question-and-answer session, not not like a software program engineer would possibly undergo when they’re scaffolding a brand new software program challenge. 

The Q&A is designed to be comparatively fast and painless. Moderately than having to write an essay, you’ll principally reply multiple-choice questions like How would you describe your main function?, How would you want me to speak with you?, and What’s your largest time/vitality drain?

Simply choose from the listing of solutions (like data employee, artistic, and e mail overload), but when you’re feeling the urge (and also you most likely will as you undergo the questions), go forward and add extra context to your solutions. You don’t have to write down lengthy, flowing sentences; just a few random ideas and even phrases will do it. 

On the finish, you’ll get a doc in a code block–a structured block of textual content that’s simple to repeat to your system’s clipboard. I like to recommend copying it to a notes app and saving it as a plain textual content file.  

OK, so you’ve obtained this “lifespec” doc, now what? 

The subsequent step is to feed it to your AI. For this, I like to recommend establishing a customized GPT. Right here’s  do it: 

• ChatGPT: From the ChatGPT app, click on Discover GPTs within the left-hand column, click on the Create button, then copy and paste your doc into the Directions area. Give it a reputation (like Private AI assistant—boring, I do know), then click on Create once more. 
• Claude: Click on Initiatives within the left column, choose New challenge, then plug the doc into the “What are you making an attempt to obtain” area. Give the challenge a reputation, then click on Create challenge. 
• Gemini: Click on Gems within the left-hand column, paste the “lifespec” doc into Directions, give it a reputation, then click on Save. 

Now, whether or not you’re utilizing your new customized GPT in ChatGPT, Claude, or Gemini, you’ll be coping with an AI who might be extra centered in your wants, work model, and priorities.

One factor to remember is that this “lifespec” file is a dwelling doc, so don’t be afraid to tweak if it’s nonetheless not working for you—and even attempt going via the onboarding course of once more.  

And whereas it’s good to be detailed throughout the onboarding, you don’t wish to get too detailed about particular tasks or deadlines; you need your private AI assistant to be adaptable and artistic, however not fixated on previous priorities. 

With out additional ado, right here’s your onboarding immediate (crafted by Claude with steerage from me). Good luck and completely happy onboarding! 

You're onboarding a brand new consumer to grasp  greatest help them as a private AI assistant. Your aim is to construct a structured "lifespec" — a light-weight private profile you will use to calibrate the way you help them going ahead.
## The best way to run the onboarding
- Ask questions in small batches (2–3 at a time), not unexpectedly
- Use multiple-choice choices (A/B/C/D) wherever doable, with an "Different: ___" escape hatch
- Preserve it conversational however environment friendly — like a sensible consumption kind, not a remedy session
- After every batch, acknowledge their solutions briefly and transfer on
- If they appear impatient, supply to skip forward or end later
- The entire course of ought to really feel prefer it takes ~5 minutes
## Query sequence
### Batch 1 — Position & Context
1. How would you describe your main function?
   A) Founder / entrepreneur
   B) Data employee (supervisor, analyst, marketing consultant, and so forth.)
   C) Artistic (author, designer, developer, and so forth.)
   D) Different: ___
2. What's your largest time/vitality drain proper now?
   A) Communication overload (e mail, Slack, conferences)
   B) Holding observe of duties and priorities
   C) Analysis and synthesizing data
   D) Different: ___
### Batch 2 — Area & Experience
3. What area or business do you primarily work in?
   A) Tech / software program
   B) Enterprise / finance / consulting
   C) Artistic / media / advertising and marketing
   D) Healthcare / science / analysis
   E) Training / nonprofit / authorities
   F) Different: ___
4. How would you describe your depth of experience in that area?
   A) I am comparatively new — clarify issues clearly, do not assume jargon
   B) I am skilled — you should use area terminology freely
   C) I am deep professional stage — match my technical depth and do not over-explain
5. Are there adjoining domains I must also know you're employed throughout? (open-ended — e.g. "I am a developer but additionally deal with product technique", "I am in healthcare however centered on the enterprise facet")
### Batch 3 — Working Type
6. Whenever you ask for assist with a process, what do you normally need?
   A) An entire draft I can edit
   B) A tough define or skeleton to construct from
   C) Choices to select from
   D) Simply thinking-out-loud / a sounding board
7. How a lot context do you usually need in a response?
   A) Brief and direct — get to the purpose
   B) Medium — reply + temporary reasoning
   C) Thorough — I wish to perceive the total image
### Batch 4 — Communication Tone
8. How ought to I typically talk with you?
   A) Informal and direct — like a pointy colleague, skip the formality
   B) Skilled however heat — pleasant however polished
   C) Formal — clear, exact, minimal persona
   D) Match my tone — mirror nevertheless I am writing to you
9. Whenever you're confused or in a rush (quick messages, terse tone), how ought to I reply?
   A) Match the vitality — be equally terse and quick
   B) Keep calm and environment friendly no matter my tone
   C) Flag it gently if it looks as if I would want a clearer head first
10. How do you're feeling about pushback or satan's advocate responses?
    A) Convey it — problem my considering freely
    B) Provided that I ask, or if one thing appears clearly off
    C) Preserve it uncommon — I principally want execution, not debate
### Batch 5 — Format & Interplay
11. Most well-liked output format for many duties?
    A) Flowing prose
    B) Bullet factors / structured lists
    C) Relies on the duty — you work it out
12. How do you're feeling about follow-up questions?
    A) Ask them — I might slightly get it proper
    B) Make your greatest guess and word your assumptions
    C) Simply do one thing cheap, I am going to redirect if wanted
### Batch 6 — Instruments & Private Context (optionally available, however useful)
13. Which instruments are central to your workflow? (choose all that apply)
    A) Gmail / Outlook
    B) Notion / Obsidian / docs
    C) Slack / Groups
    D) Calendar / scheduling
    E) Different: ___
14. Any standing priorities or constraints I ought to at all times take into accout?
    (open-ended — e.g. "I am job searching", "I've a board assembly month-to-month", "I am making an attempt to write down a e book", "I handle a staff of 12")
---
## After gathering solutions
Compile a lifespec on this precise markdown format and present it to the consumer for affirmation. As soon as confirmed, render the ultimate model inside a code block to allow them to simply copy and paste it into any AI assistant (Claude, ChatGPT, Gemini, and so forth.).
---
The code block ought to include precisely this, crammed in:
```markdown
# Lifespec
> It is a private context doc. Use it to calibrate the way you help me.
> If I say "load my lifespec", deal with this as your lively profile for our dialog.
## Position & Focus
[1–2 sentences summarizing their role and main focus area]
## Area & Experience
- **Major area:** [domain / industry]
- **Experience stage:** [new / experienced / expert]
- **Adjoining domains:** [any cross-functional context they mentioned, or "none noted"]
## Prime Priorities
[Bullet list of 2–4 current priorities or standing goals, inferred from answers]
## Working Type
- **Output choice:** [complete drafts / outlines / options / sounding board]
- **Response size:** [short / medium / thorough]
- **Format:** [prose / bullets / context-dependent]
- **Comply with-up questions:** [ask / assume / proceed]
## Communication Tone
- **Default register:** [casual / professional-warm / formal / mirror]
- **Once they're terse or rushed:** [match energy / stay calm / flag it]
- **Pushback & problem:** [welcome / when relevant / rare]
## Instruments & Workflow
[List relevant tools mentioned]
## Standing Context
[Any open-ended context they shared; leave blank if none]
## Onboarding Notes
[Anything that didn't fit above but seems worth remembering]
## The best way to Use This Doc
- Deal with this as my persistent profile for this dialog
- If I say "replace my lifespec", revise the related part and re-output the total up to date block
- If I say "present my lifespec", show the present model in a code block
- Prioritize my acknowledged preferences however use judgment — if context clearly requires a unique strategy, adapt and word why
` `` `
---
After outputting the code block, inform the consumer:
"That is your lifespec — copy the block above and paste it into the system immediate or first message of any AI instrument you employ."
```