Small inputs, clear boundaries.

The analyzer request can contain a public TikTok URL, audience, optional product context, goal, available filming material, transcript or on-screen text. If you request a revision, we also receive the selected draft reference, change reason and correction choices. Copy and review events record the analyzer surface, report and draft references, draft number, judgement and selected change reason. Revision records link the new report to its parent. These events do not send the draft text or private rewrite context and do not confirm adoption, filming or results. Anonymous product events contain a random visitor ID, a short session ID, the page path, coarse UTM values and event properties. The optional hook game records anonymous visits, choices, completion, copy actions and clicks to the analyzer. It does not send the nickname, share text or full page URL. Google sign-in sends the profile fields provided by Google. We do not ask for passwords, payment details, private account cookies or contact lists.

Public-link metadata is fetched through a protected server-side connector. In manual text mode that lookup is skipped. When you request model-assisted suggestions, your supplied text and product context, available post description and published subtitles are sent to the configured model provider through our server. A requested revision uses the original report context and your explicit correction choices. The browser never receives provider keys. Do not paste secrets, private customer data or confidential source code into the form.

The prototype stores browser visitor identifiers, server-side daily usage counts, hashed request identifiers and short event records for funnel measurement. Research logs record a hash of the query or source link and the operation status. Limited report diagnostics record service status, attempt counts, timing and available usage totals, without source text or your brief. Public video descriptions, counts and published subtitles may be cached and reused for up to 24 hours; that cache does not contain your audience, product context, supplied text or generated advice. Daily allowance protection uses a server-side keyed hash of the request IP and anonymous visitor ID; raw IP addresses are not written to these usage tables.

When revision is enabled for a completed report, we store an encrypted snapshot of its original input, source text, brief and drafts, with a 64 KiB limit and a 24-hour expiry. Short-lived encrypted revision results allow the same submission to be retrieved without starting another model request. Expired content cannot be used for revisions and is removed during subsequent cleanup requests. A secure, HTTP-only cookie binds anonymous revision access to the browser that received the report. Private rewrite references stay out of analytics, URLs, copied drafts and Markdown exports.

Operation references, hashes, limited correction choices, allowance records and execution status may remain after content expiry to prevent duplicate execution and reconcile uncertain results. They do not contain the source text, private brief or draft body.

If you sign in and save a report, its source link, summary and report payload, including any supplied product context, filming material, transcript, screen text and generated drafts, are stored in the account history database. The account database stores the Google subject and profile fields needed for sign-in, plus a hashed session token. Self-serve retention and deletion controls are still required before paid access opens.