How to correct what GPTree assumes about you

Personalization is only useful while it is accurate. If GPTree keeps framing answers around the wrong niche, the wrong role, or a problem you solved months ago, something it carries between conversations is out of date. Two things are worth checking, and both are yours to change.

Correct it in the chat

The fastest fix is to say so, in the conversation, the way you would correct a colleague:

• "I am not an SEO specialist, that is not my focus."
• "I no longer run an agency. I do positioning work for founders now."
• "That is not my main problem any more. Leads never convert is the real one."

That updates the short profile GPTree carries into new conversations, the one holding your role or focus and the problem you said you were working on. State the accurate version alongside the correction and it replaces the old value. Reject the description without offering a replacement and the field is cleared instead, because working from no assumption beats working from a wrong one.

The change lands on the conversations you open afterwards, not only the one you are in, which is the point: a correction you make once should not need repeating in every new chat.

Fix the memories behind it

Separately from that profile, GPTree stores facts, decisions, patterns, and identity picked up as you work.

Identity memories are the ones behind "it keeps describing my business wrong", and they correct themselves from the same sentence you just typed. Say "I am not an AEO specialist, that is not my niche" and the stored belief it names is retired on the spot: it stays visible in the Inspector marked contradicted, and it stops being used in answers. State the accurate version, "I do positioning work for founder-led service businesses", and that becomes the live one. Corrections apply per topic, so fixing your role leaves what GPTree knows about your company alone.

Only what you write counts here. A description GPTree produces about you in an answer is never stored as identity, so you cannot end up corrected by the assistant's own guess.

For everything else, open the Memory Inspector. Filter by project, search text, layer, kind, or scope, then act on the rows that are wrong:

• Edit a memory whose detail drifted.
• Delete the ones that are simply stale.
• Disable anything you want out of the way but not gone.
• Pin the ones you would be annoyed to lose. A pinned memory is always eligible to be used, whatever its length or confidence.

Scope is worth a look while you are in there. A memory scoped to one conversation only applies there; project scope travels with the project; account scope follows you everywhere. A wrong account-scoped memory is the one most likely to be behind answers that feel generic in every chat.

Fix the persona that frames the answers

There is a third place an assumption can live. A persona is the standing brief GPTree works from: your tone, your depth, your field, who you serve. If your default persona says you specialise in something you do not, every conversation opens from that wrong footing, no matter how clean your profile and memories are.

Say so in the chat and GPTree can now revise the persona itself:

• "Update my persona, I am not an AEO specialist."
• "Change my persona so it stops opening with a summary."

Because a persona shapes every later conversation, the change is not silent. GPTree shows you an approval card naming the persona it wants to edit, and nothing is written until you approve it. Ask what it plans to change first if you want to see the wording before it lands.

Two limits worth knowing. System personas are shared by everyone, so they cannot be edited: ask for your own copy instead and GPTree will create one and set it as your default. And a revision replaces the persona's instructions rather than appending to them, so if there is wording you want kept, say so as part of the request.

You can always do the same thing by hand on the Personas page, where you can read a persona's full text, edit or delete your own, and set the default for new chats. You can also pick a different persona per message in the composer.

When the answers are still too generic

If the profile and the memories both look right and replies are still vague, the missing piece is usually context you have not given yet rather than context GPTree lost. Put the durable material where it belongs: project instructions and reference files for the things you would hand a new hire, memory for the decisions nobody thinks to write down.

And if you would rather a particular session left nothing behind at all, use a temporary conversation. It saves no memories, and the conversation is deleted after 24 hours of inactivity.