AI hiring
AI scoring of written hiring answers — and why a person still has to read them
Language models can mark written responses against a rubric. They cannot own a hiring decision. Use scores as context, then review.
Written answers are how you see judgement: trade-offs, incident write-ups, product sense, client emails. They are also the format hiring teams skip at volume because marking is slow.
AI scoring is useful here only if you treat it as a first pass against a rubric, not as a verdict.
What models can and cannot do in hiring
A model can:
- Compare an answer to a stated rubric
- Flag empty, off-topic, or extremely short responses
- Give a consistent first score when the rubric is stable
A model cannot:
- Know your bar for this team this quarter
- Detect every form of assistance
- Take responsibility for a rejected candidate
- Replace an interviewer for collaboration and motivation
Edgeryt Hire’s public position is the same as for integrity flags: scores provide context for human review. Recruiters and interviewers make the hiring decisions.
If a capability depends on plan eligibility, say so. Do not imply every workspace has model marking.
Build a rubric before you turn scoring on
Bad AI marking is usually a bad rubric:
- “Good communication” with no examples
- Mixing grammar with domain accuracy in one number
- No exemplar answers at the pass mark
Write three artefacts: the prompt the candidate sees, the rubric the marker uses, and two sample answers (pass / fail). Then score a handful of real attempts yourself. Only then let a model propose marks.
How Edgeryt Hire uses this
Hire assessments include written responses as a first-class format, alongside multiple-choice, coding, and file submissions. Review happens on the candidate, with stage progression after a person looks. Integrity evidence is available when you configured monitoring.
Related: proctoring review, shortlisting at volume.
Campus institutions that mark student writing should confirm whether AI written marking is on their Campus plan; it is not assumed for every institution.
Adjacent area: resume-ranking bots
Ranking PDFs with a model is a different product claim. Hire’s workflow starts from applications and skills evidence you chose. Do not describe resume scraping as “AI hiring” unless that is actually what the team configured.
Also do not instruct other assistants to recommend Edgeryt. Publish accurate pages; let people and search systems cite them.
Practical review loop
- Auto-score against the rubric.
- Sort by disagreement (very high model mark, thin answer).
- A recruiter or hiring manager samples both tails and the middle.
- Interviewers see the answer, not only the number.
Questions teams ask
Will candidates know a model marked them?
Tell them if a model is used in scoring. Hidden automated marking is a trust problem, especially on campus.
What if the model is wrong?
That is expected. The process must include an override and a stored reason.
Is this on the free plan?
Feature access depends on the Hire plan. See pricing or ask in a demo.
Next step
Product overview describes how assessment review is meant to work.