Privacy

This site counts page views through Plausible, which sets no cookies, stores nothing on your device, and builds no profile of you across visits or across sites. It records the page, the referrer, and a coarse country, browser and device type, and nothing that identifies you.

The personal data held is what you type into a form and send: a claim, a follow, a request for the release note, an inquiry, a purchase, or the reviews of a person the index pays to review labels. Email addresses are never published. The emails the index sends carry no tracking: no open pixel and no rewritten links.

Pages are served by Vercel, which keeps ordinary server logs including IP addresses for a short period, as any web host does.

The typeface is loaded from Google Fonts, so your browser makes a request to fonts.googleapis.com and Google receives your IP address. The analytics script is loaded from plausible.io, which receives the request and discards the address without storing it. Those two are the only third parties your browser contacts while reading. Everything else is served from this domain: the stylesheets, the scripts, the marks and the badges.

If you use the theme picker in the footer, your choice is kept in your browser’s local storage under the key theme. It never leaves your device and is not a cookie; clearing site data removes it. The operator’s own admin page uses a session cookie; no page a reader sees sets one.

The claim form collects your name, work email, company, an optional one-line description of the product and an optional message. It is stored in a database hosted by Neon and handled by this site’s own functions on Vercel, and a verification email is sent to the address you gave. Nothing is published until the address is verified.

If the claim is approved, the page shows your name, your company and the date, plus the description if you supplied one. Your email address is never published. It is used to verify the claim, to tell you when an edition publishes and when your product’s standing moves past the noise floor, and to send the vendor brief. Each of those emails carries a link to stop them.

The follow form on a product or category page collects one email address and the thing you asked to follow. A confirmation link is sent, signed and valid for three days; nothing is stored until it is clicked. Once confirmed, you get one email the morning each edition publishes, saying where that product or category moved and where it held.

Every one of those emails carries a stop link and a link to a page listing everything the address follows, with a stop for each and the release note on or off; the page can also be requested by address at /alerts/manage/, and the link it sends lasts thirty days. Stopping marks the follow as stopped and keeps that row, so the record shows that you asked and that you stopped; ask and the row is deleted instead.

The release-note form collects one email address. A confirmation link is sent, and on the click the address becomes a contact with Resend, the email service, opted in to the release note for this index. Every release note carries a link to a preference page where you can choose which of the index’s emails you get, or none.

The index pays people to review a sample of the labels: they read a model’s answer and settle what it recommends where two readings of it differ. They see only the model answers and the labels on them, never a claim, a follow or any other person’s data. Their work is checked as they go: a share of what they read are control items with a known answer and attention items with a plainly wrong side, and the method page reports how the reviewers agreed with the judge and with each other, as numbers.

For each reviewer the index holds a name, an email address, a rate, a personal link to the review room, and every review made: the labels chosen, any note, and how long it took. The operator sees reviewers by name; the site never does. Reviews are part of the index’s record and stay in it; a reviewer who stops can ask for their name and address to be removed, which leaves the reviews standing under a token.

Every email the index sends is logged: the address, the subject, which template it was, and what it was about, so that any address can be looked up and every email to it stopped from one place. Resend reports whether an email was delivered, bounced or marked as spam, and those events are logged against the send.

No email carries an open pixel or a rewritten link, on any of the index’s domains, so the index does not know whether you opened or clicked. The operator’s own sign-in link and the collector’s alerts are logged the same way as everything else.

The inquiry and correction forms collect what you type and are stored the same way, in the same database, and reach a person by email.

Purchases are handled entirely by Lemon Squeezy as merchant of record. They collect your payment and billing details under their own privacy policy; this site never sees a card number. What reaches the index is that a subscription exists, the email on the order and the plan, so that the pack can be delivered and the release note sent. Each download with a license key is logged by the key’s id and the email on the order. The same key opens the assistant endpoint; those requests are not logged beyond the host’s ordinary logs.

Vercel hosts the site and runs the functions behind its forms. Neon hosts the database that stores form submissions, the follows, the reviews and the email log, and the index’s own record. Resend sends the email and holds the release-note contacts and their preferences. Lemon Squeezy processes payments as merchant of record. GitHub runs the collection and the nightly backup check, and keeps an encrypted copy of the database for thirty days that only a key held off GitHub can open. Google Fonts serves the typeface. Plausible counts page views. Those are all of them.

The model providers receive the index’s own questions and nothing about any person. The training set for the index’s open judge is the models’ answers and the labels on them; it holds no personal data, and the machines that train on it hold nothing else.

A claim is kept while the claimed page stands, because it is what makes the page claimed. A follow is kept until you stop it, and then as the record that you stopped. A release-note contact is kept until you unsubscribe. An inquiry is kept while it is being answered and for a reasonable period after. The email log is kept as the record of what was sent. A reviewer’s reviews are kept as part of the index’s record.

You can ask what is held about you, ask for it to be corrected, ask for it to be deleted, or ask for a copy. Write to the address below. It is handled by a person, not a queue. Where something has to stay because it is the record, such as a review or a stopped follow, your name and address come out of it and the row stays under a token.

The index itself contains no personal data. It records what models answered about products, and the free record of each edition is published under CC BY 4.0 fourteen days after the edition compiles. A claim adds a name and a company to a product page, with the claimant’s consent, and nothing else about a person is ever published. Reviewers appear on the method page as numbers, never by name.

AI Indexes LLC, 2700 S Gilbert Rd, Ste 5 #234, Chandler, AZ 85286, United States. Anything about this page, or any notice under it, goes to index@hr-ai-index.com.

The other pages: Terms of Use · Disclaimers · Refunds.