Short answer: Yes. Upload a batch of CVs (up to 100 per session on standard Claude Cowork, more on Fable 5.1) plus a role brief, and Claude scores each CV against your criteria, ranks them, and flags concerns. Time saving: 3+ hours per 80-CV role becomes ~20 minutes. Human recruiter still reviews the top-flagged CVs — Claude is triage, not decision.
CV screening is the single highest-volume repetitive task in most UK recruitment agencies. It’s also the task where consultants are most tired, most subject to bias fatigue, and least able to give each candidate a fair look. Claude changes that pattern entirely — but only if used correctly. Here’s the honest guide.
What Claude does well on CVs
- Consistent scoring against explicit criteria. Same weighting for CV 1 as for CV 80. Better than a tired human at 5pm.
- Cross-CV comparison at scale. Rank 80 candidates against each other in one prompt.
- Structured extraction. Pull key data (years experience, tech stack, employer sizes, salary trajectory) into a table.
- Flag concerns. Unexplained gaps, dates that don’t add up, claims that don’t match employer scale.
- Draft follow-up questions. "Based on this CV, what three questions should we ask at the first-stage call?"
- Bulk personalised rejection emails. References something specific from each CV rather than generic corporate blast.
The Harrison & Cole workflow (from our case study)
Harrison & Cole is a Birmingham recruitment agency running this workflow every week. Their prompt (with their permission, structurally identical to what they use):
Score each attached CV against the role brief below. For every CV, return:
1. Overall score (0-100). Weight: essential criteria 70%, nice-to-have 30%
2. Top 3 specific reasons for the score (reference exact CV details)
3. One follow-up question I should ask at the first-stage call
4. Red flags (or "none") — inconsistent dates, unexplained gaps, seniority claims that don't match employer size
Role brief:
[PASTE ROLE BRIEF]
Essential criteria:
- [LIST 3-5]
Nice-to-have criteria:
- [LIST 3-5]
Do not fabricate details. If a criterion can't be verified from the CV, note "not evidenced" rather than assuming.
Output as a ranked table (highest score first).Consultant reads only the top 5-10 in full rather than every CV. Same shortlist quality, 90% less consultant time.
UK compliance: what to know before you deploy this
1. UK GDPR Article 22 — solely automated decisions
Under UK GDPR, individuals have the right not to be subject to solely automated decisions with legal or significant effects — and hiring absolutely counts. The safe pattern: Claude scores and ranks, a human recruiter reviews the top-flagged CVs and makes the shortlisting decision. Never let Claude autonomously reject or shortlist without human sign-off.
2. Discrimination and bias risk
Any screening tool — human or AI — can introduce bias. Best practices:
- Never include criteria that proxy for protected characteristics (name, age indicators, university prestige as sole proxy)
- Audit Claude’s rankings periodically against actual hire outcomes — look for systematic patterns
- Give Claude explicit "exclude these criteria from consideration" instructions where relevant
- Keep records of the criteria used and the scoring produced, in case of a challenge
3. Data protection when uploading CVs
On paid Claude Cowork plans, Anthropic doesn’t use your conversations to train. For CVs specifically:
- Redact home addresses and personal contact details where possible before uploading
- Ensure your candidate privacy notice covers "AI-assisted screening" so candidates aren’t surprised
- For sensitive roles (medical, security-cleared), consult your DPO before rolling out
4. Candidate transparency
REC and industry best practice increasingly expect agencies to disclose AI use in their process. Add a one-line note to your candidate privacy statement: "We use AI-assisted tools to help our consultants triage large numbers of applications; final decisions on shortlisting are made by our human consultants."
What Claude gets wrong on CVs (and how to mitigate)
- Over-weights specific keywords. If you say "must have Python experience" and a CV has "Python" mentioned once in a hobby project, Claude may over-credit it. Mitigate: specify "must have 2+ years of professional Python."
- Under-weights soft signals. Personality, culture fit, "why they’d be great here" — these live in the interview, not the CV. Don’t expect Claude to assess them.
- Can invent inference. If asked "why is this candidate ideal?" Claude may reach beyond what the CV supports. Always prompt: "cite specific CV details for every claim; if not evidenced, say so."
- Struggles with poorly-formatted CVs. Modern PDF CVs work; heavily-designed graphic CVs sometimes lose structure. Ask candidates for text-based CVs where possible.
Beyond screening: three more Claude workflows for recruitment
1. Personalised rejection emails at scale
Draft rejection emails for the following [N] candidates who were not shortlisted for [ROLE].
For each: reference one specific thing from their CV so it feels personal (background, project, experience level). 6 sentences maximum. Warm but honest — no vague corporate language. End with an invitation to be considered for future roles.
CANDIDATES:
[PASTE EACH CV OR SUMMARY WITH NAMES]2. Job description rewrite for external posting
Rewrite this client-provided job description for external posting. Structure:
1. Two-sentence hook: why would a great candidate care about this role?
2. Day-to-day (5 bullets, specific to what they'll actually do)
3. What we're looking for (5 bullets split into "must-have" and "bonus")
4. What we're NOT looking for (2 bullets — honest about who won't enjoy this)
5. What's in it for you (specific benefits, growth path, culture)
Voice: direct, honest, human. No "rockstar" or corporate cliches.
Original JD: [PASTE]3. Candidate briefing pack for the hiring manager
Produce a candidate briefing pack for the hiring manager, based on this CV plus my first-call notes.
Structure:
- Candidate summary in 5 bullets
- Fit against the role brief (evidence-based)
- Three suggested interview questions specific to this candidate
- Any flags to explore in the interview
- Salary/notice expectations if disclosed
Voice: professional, brief, actionable. British English.
CV: [PASTE]
FIRST CALL NOTES: [PASTE]Realistic time and quality wins
Based on 6+ UK recruitment agencies we’ve worked with:
- CV screening: 3-4 hours per 80-CV role → 20 minutes
- Rejection emails: 45 minutes per role → 5 minutes
- JD rewrites: 30 minutes per client → 5 minutes
- Consultant weekly time saved: 6-10 hours for a mid-tier consultant
- Team weekly time saved (6-person agency): 20+ hours — equivalent to hiring a full-time role without hiring
Getting started (15 minutes)
- Sign up for Claude (Pro tier, £16/mo)
- Grab one recent role you screened manually last week — the role brief plus 20-40 CVs
- Run the CV screening prompt above
- Compare Claude’s top 5 to who you actually shortlisted — note the overlap
- Where Claude ranked differently, ask "why?" — you’ll learn how to refine your prompt
Our Claude for recruitment agencies industry page covers the full workflow. And the Harrison & Cole case study shows a real UK agency running this at scale.