School Data Protection Summary

For data protection officers and school leaders. Last updated: 18 August 2026. This page states facts about how the service handles data so your school can make its own assessment; it is a summary of, not a substitute for, our privacy policy.

What the service does

Is It AI? screens a piece of writing for patterns associated with AI-generated text and returns an AI-pattern score with flagged passages and reasons. Results are screening signals for a teacher to weigh alongside their own knowledge of the student; they are never proof that a student used AI, and we say so on the result itself. Our measured error rates and their limits are published on the methodology page.

When a teacher scans student work, the school remains the data controller for that work. The tables below set out exactly what we receive, where it goes and how long anything is kept, so that assessment can be made on facts.

What is sent where, and for how long

DataRecipientPurposeRetention
Submitted text (which may be student work)Anthropic (Claude API), plus our own servers in memoryAI-pattern analysis during the scanTransient. Held in memory for the duration of the scan, then discarded. We store no copy, in whole or in part. Anthropic processes it under its API data policy, which does not use API inputs to train models by default
Scan record (scan type, result, score, date)Our database (Prisma Postgres)Scan history, CSV and PDF export for account holders. Contains no words from the submissionUntil the account is deleted
Content fingerprint: a one-way SHA-256 hash of the normalised text, with scores and flagged-passage character positionsOur database (Prisma Postgres)De-duplication, so thirty scans of the same essay bill as one. The record contains no words from the text and the text cannot be reconstructed from it24 hours
Account data (email, name, password)Our database (Prisma Postgres)Authentication and account management; passwords are bcrypt-hashedUntil the account is deleted
IP address, as a one-way SHA-256 hashOur database (Prisma Postgres)Rate limiting for anonymous scans; the full IP is not stored24 hours
Payment detailsStripeBilling. We never see full card details, and Stripe never receives submitted textPer Stripe's policy; our payment records, 7 years (legal requirement)
Email address, for transactional emailResendWelcome, password reset and billing emails. Emails never contain submitted textFor delivery of each message

What is never stored

  • Submitted text is not stored, on any tier including free — no copies, no excerpts, no previews. Before 17 August 2026 a 200-character preview was retained with each scan record; that practice ended on 17 August 2026 and every stored preview was deleted.
  • Submitted text is not used to train any model, ours or anyone else's.
  • We build no profile of individual students. Scans are recorded against the account of the teacher or user who ran them, not against the author of the text.
  • Full IP addresses and full API keys are not stored; both are kept only as one-way hashes.

Facts relevant to your lawful-basis assessment

  • Students' text is processed transiently, for the single purpose of the scan, and then discarded. What persists is metadata about the scan, never its content.
  • The service is not designed to process special category data, and we do not intend or seek any such processing. Free-text submissions are not analysed for anything other than AI-writing patterns.
  • Because no text is retained, there is no accumulated repository of student work to secure, breach, hand over or delete at contract end.
  • Text is processed by Anthropic in the United States during the scan. Our privacy policy sets out the transfer safeguards; your review should weigh this transfer alongside the transient processing above.
  • Data subject rights: account holders can export their data (JSON) and delete their account, which deletes their scan history, from the dashboard.

Fair process for students

A detection result is a screening signal, never proof, and any school policy built on it should say the same. The JCQ's guidance on AI in assessments points towards process a student can see and answer: our results show which passages were flagged and why, so a conversation with the student starts from something both sides can read, and a decision never rests on a bare percentage. How the scores are produced, and the measured error rates behind them, are published on the methodology page.

Sub-processors

  • Anthropic (US) — AI analysis of submitted text during the scan
  • Vercel — hosting and infrastructure
  • Prisma Postgres — database for account data and scan metadata
  • Stripe — payment processing; never receives submitted text
  • Resend — transactional email; never contains submitted text

Contact

Questions from a DPO, or a request for anything above in writing for your records: hello@isitai.co.uk. The full policy is at isitai.co.uk/privacy.