Skip to content
rankr
Sign inStart free

Fairness and the use of AI

Rankr uses AI to build exercises and to draft scores, and a person at the employer makes every decision. This page says where the AI sits, what it never sees, and what that means under hiring law.

Last updated 2026-10-31

Where the AI is

  1. Building the exercise. A model writes the material, plants specific flaws, and writes the answer key. An interviewer reviews that key, edits it, and approves the item before any candidate sees it.
  2. Inside some exercises. In a pull request or document critique the candidate can ask an assistant that is scripted to miss some flaws and raise some false ones. Catching that is part of the exercise, and the candidate is told the assistant can be wrong.
  3. Drafting the score. After submission a model compares the candidate's work with the planted key and drafts a score, with a reason for each unit. An interviewer reads that draft with the evidence beside it, changes anything they disagree with, chooses a recommendation, and finalizes. Nothing is final without a person.

What the AI never sees or uses

  • No protected characteristics. Name, email and CV are never sent to the scorer, and the scorer is told nothing about the person beyond their work in the session.
  • No video, audio, keystroke patterns or proctoring. Rankr records what the candidate wrote, answered and asked, nothing else.
  • No learning from your candidates. Model providers are used through their APIs, which do not train on customer content.

Why scores are comparable

Every candidate on an item is scored against the same key, with the same weights, by the same rubric. The score lists each planted flaw with the outcome, caught alone, caught with the assistant, missed, or dismissed when the assistant raised it, and shows the candidate's own words as evidence. Finalized scores on an item can be compared side by side, and you can see where each candidate lands among those who finished.

Known limits

  • A drafted score can be wrong. That is why the evidence is shown and the interviewer finalizes.
  • Timed exercises, rapid fire above all, test whether confidence matches accuracy when there is no time to check. Interviewers can remove or extend time limits for a candidate who needs it.
  • Code exercises assume familiarity with the language of the item. Every exercise is tagged with the roles it suits, so candidates are not sent exercises outside their field.

Hiring regulation

AI used to evaluate candidates is treated as high risk under the EU AI Act, and some jurisdictions, including New York City, require notices and bias audits for automated employment decision tools. Rankr is built for the deployer obligations that follow: human oversight of every score, a notice to candidates that AI is used, logging of every session and score, and export of the data needed for an independent audit. Your organization remains the decision maker and the controller of candidate data; Rankr provides the tooling and the records.

What candidates are told

Before starting, a candidate sees who invited them, what the exercise asks, how long they have, that an AI drafts a score and a person at the organization decides, and where the privacy policy is.