AI Act Art. 13

Instructions for use.

For recruiters and hiring managers who operate Resume Screening AI. What the system is for, what it will not do, where it fails, and how you overrule it. A score is not a hiring decision.

Last updated 20 August 2026. Provider: Ksaitor Media Pte. Ltd. These instructions are how the product is meant to be used. They are not a substitute for your own legal advice.

1. Intended use

Resume Screening AI is a first-pass tool for recruitment. You supply a job description and a set of resumes. The system parses those documents, scores each one against that job description, and shows a ranked list with a written rationale so a person can decide who to read next.

It is intended for recruiters, hiring managers, and workspace members screening applicants for a role they control. It is not a candidate-facing product and is not intended to talk to applicants.

Screening is currently paused for visitors in the EU and EEA. If you screen people who are in the EU from outside it, GDPR still applies to you as the employer. See the EU hiring-law page and the privacy policy.

2. What it does

  • Parse uploaded resumes (PDF, DOC, DOCX, and many image scans) and extract text.
  • Compare each resume to the job description you supply, using the same rubric for everyone in the batch.
  • Return a score from 0 to 100, a written rationale, and short strengths and weaknesses.
  • Strip name, email, and phone from the text sent to the ranking model. The recruiter view still shows the original name.
  • Optionally fill custom fields you set in Settings (for example years of experience), from the resume text only.
  • Accept resumes by upload, spreadsheet of URLs, or email to a job inbox.

Typical flow: paste or edit a job description, upload resumes (or email them in), wait for parsing and ranking, then review the list. You can re-rank after you change the job description.

Out of scope

What it does not do

If a hiring step is not on this list as something the product performs, assume a person has to do it.

  • Reject, advance, hire, or waitlist a candidate.
  • Email, message, or chat with a candidate on your behalf.
  • Verify that employment, education, or dates on the resume are true.
  • Read video, voice, photos of faces, or any biometric or emotion signal.
  • Decide promotions, terminations, shift allocation, or workplace monitoring. Those uses are out of scope.

3. Scores are not decisions

The number is a fit estimate against the job text you provided. It does not reject anyone, does not shortlist anyone, and does not contact anyone. The legal and practical hiring decision is yours.

Under GDPR Article 22, an automated score can still count as a "decision" if the human who sees it never really disagrees with it. Do not rubber-stamp the ranking. If you would not advance someone, you must be able to say why in your own words, including when that means ignoring a high score or reading a low one.

4. How to overrule

You overrule by what you do next, not by editing the stored number. There is no "correct the score" control.

  • Read the rationale, strengths, and weaknesses before you act on the rank.
  • Open the resume. If the text or the file disagrees with the score, trust the document and your judgement.
  • Advance, skip, or interview anyone, in any order, regardless of score.
  • Remove a resume from the job if it should not be in this batch.
  • Change the job description (clearer must-haves, location, seniority) and re-rank the batch.
  • Export the list if you need it in your ATS or a spreadsheet, then decide there.

Assign review to people who have the time and authority to disagree. A junior recruiter who is measured only on "follow the AI list" is not adequate oversight.

Limitations

Where the system is weak

These are current characteristics of the product, not edge cases. Plan review around them.

  • It only sees the text it can parse

    Scans, unusual layouts, and very long files can lose content. Truncated text means a weaker or skewed score, not a complete reading of the original.

  • Identity redaction is partial

    Name, email, and phone are stripped before ranking, and contact details are stripped before similarity embeddings. Graduation years, school names, and other identity signals can still reach the model. The recruiter view still shows the original name.

  • Must-haves are deductions, not automatic rejects

    A clearly missing mandatory requirement usually costs 15 to 25 points. Several missing critical items can cap the score at 60. A technical role with a clearly non-technical background can cap at 40. A mandatory location not shown on the resume usually costs 10 to 20 points.

  • No published accuracy or fairness figure

    We have not published a labeled accuracy rate or an adverse-impact audit. Treat every score as an estimate to check, not a measured error rate.

  • It can be gamed, and it can be wrong

    Keyword stuffing, hidden text, and prompt-injection attempts can distort a score. We strip common injection patterns; that is not a guarantee. The model can also over-weight wording that happens to match the job post.

  • Re-ranking replaces the previous score

    If you change the job description and re-rank, the new score overwrites the old one. Score history across re-ranks is not kept yet.

5. Who should operate it

People who write or own the job description, who can read a resume, and who are allowed to decide who proceeds. They need to understand this page: intended use, limits, and that the score is not the decision. That is the AI literacy the role requires.

Do not upload resumes you are not allowed to process. Tell candidates, in your own privacy notice, that automated screening is part of the process. Job descriptions should be specific enough that "must have" versus "nice to have" is clear; vague briefs produce vague ranks.

Questions about the product: [email protected]. Related: Privacy policy, Terms of service, EU AI Act and hiring laws.

Keep reading

Related

EU AI Act and hiring lawsClassification, deadlines, employer duties, and where this product stands.Privacy policyWhat data we process, subprocessors, and how to request deletion.Screen resumesPaste a job description and rank a batch.