When to use it
You have a stream of CVs and one specific role, and you need to work out quickly who to invite and what to ask them. The AI plays technical recruiter. The result: a criterion-by-criterion breakdown (what's confirmed, what's questionable, what's absent), a verdict and targeted questions — instead of reading blind and deciding on vibes.
The prompt (copy and paste)
You are a recruiter. Assess this CV against the job requirements. Rely ONLY on the text of the CV, don't fill in blanks.
THE ROLE — key requirements (must / nice-to-have):
MUST: <FILL IN the essentials: skills, experience, years, location/format>
NICE: <FILL IN the desirable ones>
CV: "<PASTE THE CV TEXT>".
Do this:
1. A TABLE for each requirement: requirement | status (confirmed / partial / no / not stated) | where in the CV you can see it.
2. STRENGTHS for this role (2-3, grounded in facts from the CV).
3. GAPS AND RISKS: what's missing from the musts; what looks vague (no numbers or outcomes, frequent job changes, and so on).
4. INTERVIEW QUESTIONS: 3-5 targeted questions that close exactly those gaps and doubts.
5. VERDICT: invite / borderline / not a fit — one sentence with the reasoning.
Do not assess on gender, age, name, photo or nationality — only skills and experience against the requirements.
Why it works: scoring against an explicit rubric (must/nice) with a status and a "where you can see it" makes screening auditable and cuts down on gut-feel; the demographics clause pushes back on model bias.
Filled-in example
MUST: Python, 3+ years; PostgreSQL experience; remote. NICE: Docker, fintech background. CV: 4 years of Python, MySQL mentioned, no mention of PostgreSQL, remote is fine.
What the AI should return: the table — Python 3+ → confirmed (4 years, projects X); PostgreSQL → not stated (MySQL is present, partially adjacent); remote → confirmed. Gap: no explicit PostgreSQL. Questions: "Describe a task where you used PostgreSQL or complex SQL"; "Any experience migrating off MySQL?". Verdict: "borderline — invite, clarify the database side at interview."
Variations
- Rank a batch. "Here are 8 CVs and the role — rank them by fit against the musts with a short reason for each."
- Gaps only. "What in this CV does NOT meet the requirements, and what must I ask about" — saves time on strong candidates.
- Cover letter. "Compare the CV and the cover letter: where do they contradict, what strengthens the candidate."
Pro tips
- AI screening is a filter and a source of questions, NOT an auto-reject: the final call belongs to a human. Automatic rejection on a CV is risky both legally and substantively (the model can't see the context behind the lines).
- Demand "where you can see it in the CV" for every status: it cuts out invention and gives you a fast way to verify. "Not stated" is not the same as "no" — worth teaching the model that explicitly.
- Explicitly forbid demographic assessment: models inherit bias from their training data. There's one criterion — skills and experience against the requirements of the role.