Guide · ChatGPT

Make ChatGPT sound like you — and treat you like you.

Writing samples fix the prose. A personality tuning fixes everything else.

Quick answer: Use writing samples when you want ChatGPT to mimic your word choice and sentence rhythm on drafts. Use a personality tuning in Settings → Personalization → Custom Instructions when you want ChatGPT to match how you think (answer-first order, direct pushback, and option framing).
On this page

Two approaches. Two different "you."

Most guides recommend a writing-sample approach: paste five to ten examples of your writing and ask ChatGPT to summarize your style. That works well when the deliverable is a draft in your voice (emails, posts, or memos).

However, “making ChatGPT sound like you” also means stopping it from talking like a generic customer-service script during everyday problem-solving. That part depends on how you want to be treated: whether you want the conclusion first or the reasoning first, how blunt critique should be, and whether tangents are welcome. A communication tuning handles that in a few plain-English rules:

  • Writing samples win when the output is a finished draft that should read like you wrote it.
  • Communication tunings win when you are thinking out loud, debugging, or making decisions.

The 15-minute personality-first walkthrough.

  1. Take a quick test (~5 minutes): Try the free in-browser MBTI test (or pick from Big Five, Enneagram, DISC, and Attachment on the tests hub).
  2. Copy your compact block (~2 minutes): Pick your type in the library—or grab any of the 16 condensed blocks on one page.
  3. Paste into Custom Instructions (~1 minute): Open ChatGPT → Settings → Personalization → Custom Instructions and paste into Box 2 (How would you like ChatGPT to respond?). Start a new chat to test.

Here is the condensed ENFP block as an example:

Show up bright; I calibrate off your energy as much as your content, and a flat response signals you don't get it. Show ideas as connected webs, not linear numbered steps. Treat tangents as the thinking, not a distraction; follow them. Offer frameworks as scaffolding: "here's one way to think about it," never "the right approach is." Validate the idea before evaluating it; "that's interesting because…" then the critique. Use possibility language: "what if," "I'm curious about." Be playful; sterile reads as disinterest. No premature "that's impractical" verdicts, no forced closure. When unsure, open up rather than narrow; I'll close down when I'm ready.

Notice that every sentence is a concrete rule about energy, structure, tangents, and critique order.

Layer both. They don't compete.

You can combine both in Box 2 by scoping each rule clearly: “When advising me, lead with a recommendation and key tradeoffs. When drafting as me, use short sentences, active verbs, and one idea per paragraph.”

In ChatGPT Work on the web, you can also enable Reference my writing style from connected apps while keeping your communication tuning in Custom Instructions.

What changes in practice.

With the ENFP block above, bringing a half-formed idea no longer triggers a rigid five-step checklist that shuts down exploration. Instead, ChatGPT connects related angles first and frames critique as “here is one way to test that.”

Conversely, an INTJ tuning cuts the warmup completely: conclusion in sentence one, zero hedges, and at most one clarifying question. Same model, different standing orders.

Questions people ask.

Which approach is better, writing samples or a personality tuning?

They fix different problems. Writing samples are better when the output is the product; emails, posts, drafts that must read like you wrote them. A personality tuning is better when the conversation is the product; how ChatGPT paces, structures, and challenges its replies to you. If you only fix one, fix the one that's actually bothering you.

How long does the personality-first setup take?

About 15 minutes end to end: ten for the free in-browser MBTI test, two to copy the block for your type, one to paste it into Settings → Personalization → Custom instructions, and a couple to sanity-check a fresh chat. If you already know your type, it's under five minutes.

Do I need to know my MBTI type first?

You need a type, but not necessarily MBTI. AgentTune has free in-browser tests and matching tunings for five systems; MBTI, Big Five, Enneagram, DISC, and attachment style. MBTI is the common starting point because its 16 types map cleanly onto interaction rules, but any system that describes how you prefer to work will do.

Can I use both approaches at the same time?

You can combine conversation preferences with writing-style notes. Keep the rules distinct, check your app’s field limit and compare the results on your own tasks.

Doesn't ChatGPT memory already learn my style?

Memory collects facts; projects, preferences you've stated, recurring topics. It doesn't reliably converge on interaction style, and you can't inspect or edit what it inferred the way you can a pasted block. Custom instructions are explicit and yours. Use memory for context, a tuning for behavior.

What changed.

  • Clarified preference scope and removed universal field-limit claims; current setup controls vary by client.

Keep going.