Privacy Policy update — the AI proficiency assessment
Posted September 23, 2026 · effective September 30, 2026
The Privacy Policy gains a clause setting out what the AI proficiency assessment collects, what it is used for, when it is destroyed, which processors handle it, and who can read the result report. Until now this was stated only on the consent screen shown before a sitting.
The effective date is seven days after publication. From that date the wording below appears in the Privacy Policy itself.
New clause
3-3. The AI Proficiency Assessment
What we collect: (1) the documents, code, notes and working plan you write during the sitting; (2) the full record of your conversation with the AI inside the screen (requests and responses); (3) the results of running and checking your code or document; (4) your answers to the situational-judgement questions (and your confidence rating, if you send one); (5) your verdicts in the AI proposal review; (6) your answers in the confirmation interview and your self-reported experience; (7) input-origin signals — the count and character length of paste and drop events, the number of characters typed, and the number of times and total time the screen was left (we do not collect what was pasted or what you did away from the screen, only counts and durations); (8) your sign-in email, account identifier and the submission time of each step.
Why we use it: only to score this assessment, to write the result report, and to prepare the confirmation interview questions. We do not use it for advertising or to recommend other services, and we do not give it to AI providers for training their models.
How long we keep it: the record of a sitting is destroyed 90 days after the final submission. The final submission is the time you submitted the confirmation interview; if you stopped before that, the time you submitted your work; if neither exists, the time you started. A reviewer opening or scoring the sitting does not extend this period. If you ask us to destroy it earlier, we empty the body of the record and leave a deletion marker, and from that moment the candidate view, the sharing links and the reviewer view are all closed. A result report the reviewer has downloaded is separate from our destruction; the organisation the reviewer belongs to keeps and destroys that copy under its own recruitment-records rules.
Processors and international transfers: · Anthropic, PBC (United States): generating the in-session AI assistant's replies, scoring, writing the narrative of the result report, checking the evidence, and writing the confirmation interview questions. · OpenAI, Inc. (United States): only for writing the narrative of the result report, and only when Anthropic does not respond. This applies only to sittings consented to on the consent screen that states it (from the September 21, 2026 version); records from sittings consented to before that are not sent to OpenAI. · Turso (a United States company; servers in the Tokyo region of Japan): storing the record of the sitting. The items transferred are (1) to (6) above, and the transfer happens over an API call at the moment the feature is used. The retention period is the one stated above. If you do not agree to the international transfer, you cannot sit this assessment.
Who can read the result report: only after the reviewer confirms the scoring and turns sharing on. There are two paths. (1) You read it in your own account. (2) The reviewer creates a viewing link, and whoever holds that link reads the report without signing in. A link expires 30 days after it is created, or when the record of the sitting is destroyed if that comes first; a sitting can have up to ten links, and the reviewer can revoke any of them at any time. The report does not carry the raw scoring draft, the reviewer's notes or the non-public validation material. We do not give the record of a sitting to anyone outside these two paths.
Access, correction and deletion: write to help@preterview.com from the email address of the account that sat the assessment. When you ask for a correction we check the record again and either correct the result report or explain why we are not changing it. What we provide for access is the result report and the submission summary; we do not provide the raw record of the sitting, because it contains the expected values of non-public validation cases.
Before this change: the Privacy Policy had no clause covering this assessment; the above was stated only on the consent screen shown before a sitting.
This revision is recorded in the revision history of the Privacy Policy. It applies from the effective date to sittings started before that date as well.