For information
AI Transparency Statement
Where the words come from, what the model is given, and what it cannot do. Published because you should not have to guess.
Version 1.0 · 14 August 2026
Your Blueprint is not written by a language model. It comes from fixed rules and wording written by a person, so the same answers always give the same reading.
Ask Your Blueprint and the Clarity Room are written by a language model, working from your Blueprint. It can be wrong while sounding certain.
Your content is never used to train any model, and no automated decision is made about you.
1 · What is written by a model, and what is not
Some of what you read on Exist Better is written by a language model. Some is not. The difference matters, so here it is plainly.
Your BlueprintNot written by a model. Your answers are interpreted by a fixed set of rules we wrote ourselves, and the wording you read was written by a person in advance. The same answers always produce the same reading. Nothing about it is improvised.
Ask Your BlueprintWritten by a language model, working from your Blueprint and from your question. It sweeps a wide range of disciplines before choosing which few are worth opening for you.
The Clarity RoomWritten by a language model, working from your Blueprint and from the conversation so far.
The rest of the siteWritten by a person. The pages, the questions, the practices in your daily ritual and every crisis route were all written and checked by hand.
2 · Which model, and what it is told
We use a large language model provided by OpenAI, accessed over their API. Which model changes as better ones become available, and we will not pretend that the choice is permanent.
What it is givenYour Blueprint, or the part of it that is relevant, and what you have written. In the Clarity Room it is also given that conversation's earlier turns, which is what allows it to remember.
What it is told to doTo write in a particular voice, to acknowledge before explaining, to describe experiences rather than people, and to avoid clinical language. Those instructions are ours and they are long.
What it is not givenYour name, your email address, your payment details, or anything from another person's account.
TrainingYour content is not used to train any model. OpenAI's API terms do not permit them to train on data sent through it, and we have not opted into anything that would change that.
3 · What it cannot do
A language model produces text that reads as though it were written by somebody who knows. It is worth holding onto how it actually works.
It can be wrongConfidently, and in a well-written sentence. Where something matters, particularly about health, check it with somebody qualified.
It does not know youIt knows what your Blueprint says and what you have written. It has not met you and it cannot see you.
It is not a clinicianIt cannot diagnose, prescribe, or assess risk. Where something belongs with a person, it is instructed to say so.
It variesAsk the same question twice and the wording will differ. Your Blueprint does not behave this way, deliberately.
4 · Safety comes before the model
Anything you write is read for signs of risk before a reply is composed and before your membership is even consulted. That order is deliberate: somebody writing in distress should never meet a paywall or a reflective question.
Where the read finds something, the reply is set aside and support is offered instead. That is done by rules we wrote, not by the model. Nobody is alerted in the moment, so the platform is not a way of reaching us urgently.
5 · What we will not do with it
PretendNothing on Exist Better will claim to be a person. Where a model wrote it, this document says so and the platform does not imply otherwise.
DiagnoseWe will not use a model to name a condition, assign a score, or tell somebody what is wrong with them.
Decide about youNo automated decision is made about your access, your money or your care. A person makes those.
Sell it onYour content is not used to train models, to build products, or for advertising. Not now and not later.
6 · If it gets you wrong
Any part of your Blueprint can be marked as not sounding like you. It is removed, recorded, and not said to you again. Anything a model has written can be reported the same way.
We read those reports. They are the most useful thing we receive, and they are how the wording improves. hello@existbetter.online, or the correction link on the reading itself.
Every one of our documents is at existbetter.online/legal.
