System personas: six ready-made ways to be answered

A persona is a saved set of instructions that shapes how the assistant answers: the role it takes, the tone it uses, and the rules it holds itself to. Until now you had to write your own before you got any of that benefit, which meant most people never did. Every account now starts with six system personas, ready on day one.

The six

Engineering Partner treats you as a peer who already knows how to code. It reads the code you pasted before proposing anything, commits to one primary cause instead of listing five possibilities, and tells you which line to instrument to confirm it. Ask for the full file and you get the full file, with no "rest unchanged" gaps to fill in yourself.

Writing Partner returns the text and nothing else. No preamble, no "here's your revised version", no commentary buried inside the paragraph you were about to paste. Ask it to fix grammar and it fixes grammar rather than rewriting your argument in its own voice. If you are working in a second language, it corrects the English and hands it back clean.

Study Partner is built for the way people actually revise: you write down what you think is true and ask whether it holds. It answers that first, names the exact word that is wrong rather than restating the whole concept, and stops when your statement was already right. It also checks citation formats field by field, and says so plainly when it cannot verify a reference.

Thinking Partner pressure-tests a decision. It names the real decision, gives you two or three genuine options with what each one costs, then picks one and tells you the single condition that would change its mind. When two options both deserve a proper run rather than an argument, it will say so and suggest you branch and compare them.

Decision Partner is for the moment you have to commit. Where Thinking Partner opens a decision up, Decision Partner closes it down: it works the options, picks one, and then shows you what that pick rests on. Its answers close with the same four fields every time. The recommended path in one sentence. The key assumptions behind it, ranked with the most fragile first and each marked as controllable, partly controllable, or outside your hands. A fallback saying what to switch to if the first assumption breaks. And a Decision Risk of Low, Medium or High with the reason for it. The assumptions are listed precisely so you can disagree with one and ask again, which is a faster conversation than arguing with a recommendation whose reasoning is invisible.

A few things can cut that closing part short: a very long conversation, a hard length limit, or your own conversation instructions pulling the other way, and image requests do not carry it at all. When the arithmetic in an answer does not add up, a maths check may appear below the block.

Content and Social Writer drafts posts and emails in your voice rather than the cadence everyone recognises as machine-written. No "Here's the thing", no dramatic one-line paragraphs stacked down the page, no invented statistics. One version, plus a named alternative angle if a genuinely different one exists.

Using them

Pick a persona per message from the picker in the composer, or open the Personas page and set one as your default so every new chat starts with it. Because personas apply per branch, you can send the same question down two branches under two different personas and read the answers side by side. That is often faster than arguing with one of them.

Your own personas still work exactly as before, and sit in their own section on the Personas page. The built-in ones do not count against the persona limit on the Free plan.

Switching off the ones you don't want

Six extra entries in a dropdown is five too many if you only ever use one. Each system persona has a switch on the Personas page. Switch one off and it stops appearing in the persona picker in your conversations. Nothing is deleted, the change applies to your account alone, and you can switch it back on whenever you like.

The base Default persona is the one exception: it stays on, because it is what the assistant falls back to when a message carries no persona at all. If a system persona was set as your default for new chats and you switch it off, that default is cleared at the same time, so you never end up with a persona shaping your chats while being invisible everywhere else.

Where they came from

These are not guesses about who might use GPTree. Each one answers a group of people already doing that work in the product, and the rules inside them come from real conversations where the assistant got it wrong: deliverables arriving wrapped in commentary, debugging answers that restated the symptom instead of naming the defect, corrections that turned into apologies instead of fixes. If a persona feels opinionated, that is the point.