First-party measurement
The site records page views and article-engagement events in its own database. These records may include the page or article, time, anonymous visitor and session identifiers, device category, referring hostname, broad traffic-source category, link label, and normalized destination hostname and path.
For essays, the site estimates active reading time and records one-time milestones at approximately 25%, 50%, 75%, 90%, and completion. Time pauses when the page is hidden, unfocused, or inactive, and rapid scrolling is not treated as genuine reading. These measurements are approximate.
Traffic source is inferred from a referring hostname or limited campaign source information and grouped into categories such as direct, search, ChatGPT, LinkedIn, Substack, email, and other referrals. Full destination query strings are not retained for outbound-link analytics.
Outbound and book clicks
The site records clicks to external destinations, including links to the book, Substack, LinkedIn, related essays, and other external pages. For a book link, the site may record the source page or article and the reader's measured engagement state before the click.
A book click is not a book sale. dcognition.ai cannot observe whether a reader views, purchases, or otherwise acts on Amazon after leaving the site.
Coarse geography
When available, Cloudflare provides an approximate country and region code for a request. The site converts this directly into daily aggregate view counts. It does not store geographic information with a visitor or session identifier. It does not retain city, metropolitan area, postal code, coordinates, timezone, or raw IP address. Low-volume locations are combined or suppressed in the private Studio report.
Identifiers, cookies, and browser storage
Anonymous readers may receive random, first-party identifiers limited to dcognition.ai. They are not fingerprints and are not used across other sites.
- Visitor cookie
- A random identifier used for audience measurement and best-effort duplicate-Like prevention. Maximum duration: one year.
- Session cookie
- A random identifier used to group activity during a reading session. Maximum duration: 30 minutes.
- Attribution cookie
- The broad traffic-source category and referring hostname used to preserve first-touch attribution during a session. Maximum duration: 30 minutes.
- Session storage
- Per-article reading milestones, active-reading seconds, and completion state used to avoid duplicate events and contextualize a book click. The site treats this state as expired after 30 minutes. It remains within the browser session.
The site does not use browser localStorage. The first-party cookies above are Secure, HttpOnly, SameSite=Lax, and host-only. If a reader is authenticated through ChatGPT, an authenticated identifier may be used instead of an anonymous visitor identifier for Likes or comments.
Likes and comments
Likes
A Like associates an article with an anonymous visitor identifier or, when applicable, an authenticated-user identifier. This allows the site to show whether that reader has already Liked the article and to prevent routine duplicates. Anonymous uniqueness remains best-effort because clearing cookies or changing devices creates a new identity. Unliking removes the corresponding Like record.
Comments and moderation
Submitting a comment requires a display name, email address, and comment text. Every new comment begins as pending and is reviewed before it can appear publicly. Moderators may approve, hide, restore, reject, mark as spam, or permanently delete a comment.
Commenter email addresses are private. They are never displayed in public HTML or returned by public comment APIs. Only the authorized owner can see them in the private Studio moderation area, where they are used for comment administration and identity verification for deletion requests.
If approved, the display name, comment text, date, and reply relationship may be shown publicly. Replies are limited to one level.
Do Not Track and Global Privacy Control
When the browser sends Do Not Track or Global Privacy Control, dcognition.ai does not record page-view analytics, reading engagement, outbound-click analytics, traffic attribution, or coarse geographic analytics.
A Like or comment deliberately submitted by the reader still requires the minimum identifiers and records needed to perform that action, prevent abuse, and support moderation. The separate analytics event that would ordinarily describe that action is not recorded when either privacy signal is present.
Retention and deletion
Comments and their private email addresses remain with the comment until the comment is permanently deleted. Likes remain until they are removed. Analytics, engagement events, moderation records, and daily geographic aggregates currently have no automatic expiration and are retained until they are manually deleted or aggregated. Short-lived rate-limit records are used to prevent abuse.
To request removal of a comment or deletion of information you submitted, send a private message through the Jiajie Zhang LinkedIn profile. Identify the article, display name, approximate submission date, and the action requested. For a request involving a comment or private email address, you may be asked to verify control of the email address originally submitted. Do not post that email address publicly.
Some anonymous analytics cannot reliably be connected back to a particular person. Requests will be honored where the relevant record can be reasonably identified and the requester's authority can be verified.
Services involved in operating the site
OpenAI ChatGPT Sites provides the site-management and hosting environment, including ChatGPT sign-in used to protect the private Studio administration area.
Cloudflare provides the Worker runtime, D1 database, R2 media storage, network delivery, security functions, and the country and region metadata used for coarse aggregate geography.
Pages with embedded media may load a player from YouTube, Vimeo, or Spotify. Those providers may receive ordinary request information under their own privacy terms. Amazon, Substack, LinkedIn, and other external destinations receive a request only when a reader follows a link to them; their subsequent activity is outside dcognition.ai and governed by their own policies.