Privacy and student data

The short version

• No student accounts. Students can use Semblen without signing up or giving a name or email address.
• What a student analyzes is stored for 180 days, with the network address and browser details it came from, then deleted.
• We do not sell personal data, show advertising, or share data for advertising.
• No education records. We don't ask for or keep names, rosters, grades or student ID numbers.
• Students shouldn't paste anything personal, such as their name or a classmate's, into text they analyze. If one does, write to us and we delete it.

What we store when a student analyzes something

• The text analyzed (a pasted text, or the paragraphs of the article or video transcript at a link), the results, the link, the time, the IP address and the browser's user-agent. Kept 180 days; a weekly job deletes what has passed that age.
• A cached copy so the same text isn't analyzed twice: a pasted text and its result for a day, under a one-way code of the text. An article's analysis (the article, not who asked) is cached for a day, or longer for articles on our own /daily and /government pages.
• A shared link, only if someone presses Share: the analysis, and for pasted text the text, for 180 days. Anyone with the link can open it until then.
• A count of analyses, not linked to any name. Without an account, Semblen allows 2 analyses a day per browser and 30 a day per network. A cookie, semblen_browser, holds a random code used only for that count.
• Visits, counted with Umami on our own server. Umami keeps a replay of a sample of visits for 30 days to see where the site is hard to use; anything typed into a form is hidden, and pasted text and its results are left out.

Improving the model. Our privacy policy lets us use submitted text and results to evaluate and improve our detection model, including as training examples. Text kept as an example is separated from the IP address and browser details, and kept while the model that learned from it is in use. To have something removed, write to us (below).

Who else processes it

Only the services that run the analysis, each bound by its own terms. The full list, with what each receives, is in the privacy policy.

• Anthropic (United States): text analyzed is sent to the Claude API for the explanations, the summary, the claim checks and the comparison with other outlets. Under Anthropic's commercial terms it does not train its models on this content.
• Exa (United States): a claim being checked is sent as a search query.
• Cloudflare: all traffic passes through it, and it runs a bot check before a signed-out analysis.
• Hetzner: our servers, and everything we store, are in Germany.
• Stripe: only if a teacher pays. It handles the card; we never see it.

FERPA

Semblen does not ask for, receive or keep education records. Students need no account; we collect no names, rosters, grades, student ID numbers or assignments. Used as described on this page, with students analyzing published articles, public statements or their own writing without personal details in it, a teacher shares no personally identifiable information from students' education records with Semblen.

Like any website, Semblen receives the IP address and browser details of each visit; we keep them with each analysis for 180 days and use them to run the site and stop abuse, not to identify students.

Age

Our terms require users to be at least 13. The site is not directed to children under 13 and we don't knowingly collect their data; if we learn we have, we delete it. Semblen is meant for college and high-school classes.

Deleting data, and questions

Write to [email protected]. Analyses aren't linked to accounts, so tell us roughly when and what was analyzed and we'll find it. We answer within one month.

For your administrator: a short data agreement you can print, sign and hand in.