Terms · version 2026-10-17
These are the terms for using Diverge. They are short on purpose. Where they touch your data, they say the same thing as the privacy page. Diverge is for adults: you must be 18 or older to use it.
Diverge has zero tolerance for objectionable content or abusive users. We review reports within 24 hours, remove violating content, and remove the accounts that posted it. Objectionable means harassment, threats, hate or slurs aimed at people, sexual content, graphic violence, content that encourages self-harm, spam, and anyone’s private details. Disagreeing, even sharply, is not objectionable — that is what Diverge is for.
Every post, reply and photo has Report in its menu; every profile has Report and Block; every conversation in Messages has Report and Block. Blocking hides each of you from the other at once. How to do each: the support page.
Safety checks. Posts, replies, messages and photos are checked by OpenAI’s moderation model for safety before they appear. Anything it flags is not published. Each post’s text, including any words in its photo, also goes to OpenAI to score its divergence. OpenAI processes them under an API agreement that does not use them to train its models.
Everything you write or say here stays yours. By posting in a public room — a contrarian truth, a take in the Ideas orbit, a reply, a minute off the floor, a take in the morning pool — you let Diverge show it, score it, quote it back to other people as something already said, and keep it on record for as long as the room exists. That permission covers running this app. It does not cover anything else unless you opt a comment into the dataset.
Content created in the iOS app is never added to the AI dataset — nor is anything from an account that has ever posted from the app.
Diverge is building a dataset of human reasoning to license to companies that build AI models. A comment goes into it only when all three are true: you are signed in, you have told us you are 18 or older, and you switched on Add to the dataset for that comment. Nothing is in it by default. Nothing written before 25 September 2026 is in it. Nothing you said out loud in a round is in it.
When you do opt a comment in, you give Diverge a worldwide, non-exclusive, royalty-free licence to copy it, include it in that dataset, and license it to others, including for training AI models. You still own it and can use it however you like.
What goes in: the text of the comment, its divergence reading, and the votes and replies it got. What never goes in: your name, your email, your account, your device id, or anything an AI model wrote.
You can take a comment back out at any time from its thread, and it is left out of every export after that. A copy already delivered before you took it out can’t be recalled.
Anyone can read the feed. On the website, a text truth can be posted without an account; in the iOS app, posting, replying and messaging need an account and your agreement to these terms. Replying, voting, reporting, following, posting photos or videos and choosing who sees a post need an account — one per person. You post under a handle you choose, never your real name, or as Anonymous, which hides your handle on that truth; anonymous truths earn no points. Don’t pick a handle that pretends to be someone else, or one that’s crude or hateful — those are refused or renamed.
Points, levels and streaks are counted by software from votes, replies and follows. They have no cash value and can be recounted if something was gamed. A truth that enough people report is hidden until a person reviews it, and we can restore it or take it down.
You need to be 18 or older to use Diverge, on the website and in the app. If you are under 18, don’t use it — and if you find an account that belongs to someone under 18, report it and we will remove it.
The judge, the divergence reading and every rating are measurements made by software. They are consistent, and they are not verdicts about you. The divergence score is how far a post sits from what an AI would write about the same question, measured with OpenAI. None of it is advice of any kind.
Argue as hard as you like. Don’t post anyone’s contact details, slurs, or threats, and don’t post text a machine wrote and call it yours — the gate on every room is there to keep this a place for human thought. We remove a post, and the account that posted it, when it breaks these terms — a removed account can’t post, reply or send messages again.
Diverge is provided as it is, and it will sometimes break. To the extent the law allows, we are not liable for losses that come from using it. We may change these terms; when we do, the version at the top changes, and a comment you already opted in stays under the version it was opted in under.
The word frequencies the divergence reading uses come from wordfreq by Robyn Speer, used under the Creative Commons Attribution-ShareAlike 4.0 licence.
