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

Two approaches. Two different "you."

The standard advice — most how-to guides and Reddit threads — is corpus-first: collect five to ten samples of your writing, paste them in, and ask ChatGPT to describe your style, then reuse that description. It works, for what it targets. If the goal is drafts that read like I wrote them — your sentence length, your vocabulary, your comma habits — corpus-first is genuinely the better tool. A personality tuning won't teach ChatGPT your voice on the page.

But "sounds like me" usually means something wider than prose. It means the replies stop feeling like a customer-service script aimed at nobody. That part isn't in your writing samples — it's in how you like to be treated. Whether you want the answer first or the reasoning first. Whether pushback reads as respect or friction. Whether a tangent is a distraction or the actual thinking. That's personality-first territory: take a test, grab the tuning file for your type, paste it. No corpus to assemble, and the result is inspectable — a short list of plain-language rules you can read and edit, not a style description ChatGPT inferred and you have to take on faith.

Side by side:

• Corpus-first wins when the output is the deliverable — emails, posts, docs that must read as yours. It captures rhythm, vocabulary, and register that no personality type predicts.
• Personality-first wins when the conversation is the deliverable — thinking out loud, getting unblocked, making decisions. It sets pacing, directness, option count, critique order, and what happens to your tangents.
• Corpus-first costs more — you need samples you're willing to share and a judgment call on whether the inferred style description is right. Personality-first is a test plus a paste.

One more reason the default feels off: ChatGPT ships with a personality already. In AgentTune's research — about 2,200 personality-test runs across six frontier models — GPT-5.5 returned INTJ on all 100 MBTI runs. Untuned, it defaults to one interaction style for everyone. Unless you're that type, it's not yours.

The 15-minute personality-first walkthrough.

1. Take the test — 10 minutes. The free in-browser MBTI test, no signup. Already know your four letters? Skip ahead. Prefer a different lens? The tests hub also has Big Five, Enneagram, DISC, and attachment style, each with matching tunings.
2. Grab your tuning — 2 minutes. Find your type in the library — say ENFP — and copy the condensed custom-instructions block. All 16 MBTI blocks are collected in one page.
3. Paste it — 1 minute. ChatGPT → Settings → Personalization → Custom instructions → the How would you like ChatGPT to respond? field. Save, then start a new chat — instructions apply to new conversations, not old ones.

Here's the ENFP block, condensed from the full ENFP tuning:

Read it again as a stranger. Nothing in there is about ENFP trivia — every line is an instruction about treatment: energy, structure, tangents, critique order. That's the entire trick.

Layer both. They don't compete.

The two approaches write to different layers, so stack them. The tuning block goes in How would you like ChatGPT to respond? and governs interaction. Your style notes go in the same field after it, or in a project's instructions — two or three concrete observations from your corpus exercise: "short sentences, no semicolons, one idea per paragraph, sign off without a summary." Both fields hold 1,500 characters; a tuning block uses 400–700, which leaves room.

The division of labor is clean. When ChatGPT drafts an email as you, the corpus layer decides what the email sounds like. When ChatGPT talks to you about the email, the personality layer decides whether you get three options or one recommendation, critique before validation or after. Most people who bounce off corpus-only setups are missing that second layer — the drafts improved but the conversation still felt like a form letter.

What changes in practice.

Take the ENFP block above. Before: you bring a half-formed idea and get a numbered five-step implementation plan — structured, complete, and deflating, because you weren't asking for a plan yet. Tangents get politely steered back to the topic. Every suggestion arrives pre-hedged with feasibility caveats.

After: the first response engages with why the idea is interesting before weighing it. Related ideas get connected instead of filed separately. The critique still comes — the tuning asks for validation before evaluation, not instead of it — but in an order that keeps you thinking rather than defending. Structure shows up as "here's one way to think about it," which you can take or leave.

The same mechanism runs the other direction. An INTJ tuning deletes the warmup instead of adding energy: answer first, hedges cut, one follow-up question maximum. Same model, same knowledge — different standing orders about how you like to work. That's the sense of "sounds like you" that writing samples never touch.

Questions people ask.

Keep going.