Privacy Policy

Last updated 1 October 2026

This platform lets a YouTube creator add reflective prompts to their own videos, and lets a viewer answer them. Two different people are involved and they are treated differently, so this policy covers each in turn.

The short version: a viewer's own answers, notes and reflections stay in their browser and are never uploaded. Usage events are sent only if the viewer switches sharing on, and even then they carry a per-channel random identifier rather than a name or an account. Creators sign in with Google and may connect their own YouTube channel; we use that access only to read their own channel and their own videos' captions.

For viewers

What stays on your device

Your answers to prompts, anything you write in the notebook, the intentions you set and their check-ins, and your own dashboard are all stored in your browser's extension storage on your machine. They are never uploaded as identified rows, and removing the extension removes them with it.

Nothing is shared until you say so

Usage sharing is off until you turn it on. You are asked once, above your first prompt, and until you answer, nothing is stored for sending and nothing is sent — a viewer who declines never has events written down in the first place.

If you switch sharing on, this is the whole list

a random session identifier, regenerated on every page load;
a viewer identifier, described below;
which channel the prompts belong to;
the kind of event (a prompt was shown, answered, skipped) and which prompt and video it referred to;
when it happened.

No name, email address, account, or identity derived from your IP address is collected or stored.

The viewer identifier, precisely

This stream is pseudonymous, not anonymous, and we would rather say so than use a word that flatters us. So a creator can tell a returning viewer from a new one, consenting events carry an identifier with these properties:

It is derived on your device, and the input never leaves it. Your browser holds one random secret. What is transmitted is a hash of that secret combined with the channel identifier. The secret itself is never sent.
It is different for every channel. Two creators receive different identifiers for the same person and cannot combine them — that is a property of the arithmetic, not a promise about our conduct.
Switching sharing off destroys the secret. Switching it back on mints a different identifier, so your earlier sessions and your later ones cannot be connected to each other.
It is never sent to an AI model and is never used to rate-limit or otherwise single out a request.

We are not told which video you are watching

To find the prompts for a video, the extension asks for everything matching a short prefix of the video identifier's hash, then picks the right one on your machine. Many videos share any given prefix, so the request itself does not reveal which one you opened.

Turning sharing off later

You can switch it off whenever you like, in the extension's settings. The secret is destroyed at that moment, so nothing further is sent. Events you already shared remain with the creator as aggregate usage — but because the secret they were derived from no longer exists, they cannot be tied back to you or to any identifier you might generate later.

What the creator can actually see

Creators see totals and patterns, never one person's answers. Any breakdown computed from fewer viewers than a fixed minimum is reported as insufficient data rather than displayed, and that floor is enforced on the server, so it cannot be lowered from the browser.

For creators

Signing in

Sign-in uses Google purely to establish who you are. We receive your email address and a Google account identifier, and we store them to keep your channel's content attached to your account. We never receive your password.

Connecting your YouTube channel

Connecting is optional; you can use the product without it by pasting transcripts yourself. If you do connect, you grant a YouTube scope (youtube.force-ssl) that we use for exactly three things:

confirming which channel is yours, so prompts can only be generated for videos you own;
listing your own videos, so you can pick one in the console;
downloading the caption track of your own videos, so prompts can be generated from what you said.

We do not post, comment, edit, delete, or change anything on your channel, and we do not read channels other than yours. Your access and refresh tokens are held server-side, are never sent to any browser, and are deleted when you disconnect — which also revokes our access at Google. You can revoke it yourself at any time from your Google account's permissions page.

Our use of information received from Google APIs adheres to the Google API Services User Data Policy, including its Limited Use requirements.

Your video transcripts and AI

To generate prompts, the transcript of the video you selected is sent to an AI provider (Google Gemini, OpenAI, or OpenRouter, depending on how the deployment is configured) along with instructions describing the style you asked for. This is your own content, sent at your request, when you press Generate.

Analytics on the creator console

The creator console — the signed-in screens where you generate, review and publish — carries a third-party web analytics tag, Yandex.Metrika (counter 111930255). It is on those console pages only: not in the browser extension, not on the public watch player, and not on any page a viewer sees.

Metrika's session-recording feature (Webvisor) is switched on, which means your interactions with the console — pages visited, clicks, scrolling, and the contents of the screen as you work — are recorded and replayable by us through Yandex's interface. Since the console shows your video titles, transcripts and analytics, assume those appear in a recording. Password-type fields are masked and are not recorded.

The sign-in and sign-in-callback pages carry no tag at all. They are excluded deliberately and the exclusion is enforced by a build check, because those pages' URLs briefly carry the token that stands for your session. The tag is also given only the origin and path of the page you are on, never the full address with its query string or fragment.

Yandex processes this data on its own infrastructure under its own privacy policy. If you would rather not be recorded, any tracker-blocking extension will stop it, and the console works normally without it.

The public website

The page that describes creai to people who have not signed in sets no cookies, carries no analytics tag, and loads no script, font or image from anyone else. Its own code makes no request beyond fetching the page's files. It collects nothing and has no form: its one button takes you to the sign-in page.

What we never send to an AI model

Exactly one feature sends viewer-derived data to a model: the written insights on the creator's analytics screen. It receives only aggregates — counts, the percentage of viewers who left at a point in a video, the creator's own prompt text. Individual responses, session identifiers, viewer identifiers and anything a viewer typed are excluded, and the exclusion is enforced in code by rebuilding the request from a fixed list of permitted fields rather than by filtering out known-bad ones.

How long data is kept

Shared usage events are kept for as long as the creator's account is active, because the analytics they support are cumulative. We do not currently run an automatic deletion schedule, and we would rather tell you that than imply one exists. A viewer's own answers and notes are not covered by this at all: they are on the viewer's device, under the viewer's control, and go when the extension does.

Children

This product is not directed at children, and we do not knowingly collect data from anyone under 13.

Changes

If this policy changes, the date at the top changes with it. Material changes to what is collected will also be surfaced in the product rather than only here.