CV shortlists in hours, on the right side of UK law
AI screens every CV against the actual requirements and produces a scored shortlist with reasoning, which a human reviews before anyone is rejected.
The problem
How it works by hand
A single job ad can pull hundreds of applications, and someone has to read them. In practice that means the first fifty get proper attention, the rest get skimmed on a tired afternoon, and good candidates are lost to reviewer fatigue rather than merit. The alternative, not reading them all, is worse and everyone quietly knows it.
A worked example
What a working version looks like
The job’s real requirements, not the wish-list version, become a structured scoring rubric. Every application is screened against that rubric and only that rubric: the model is never shown name, age, address or anything else that invites bias, and it scores each requirement with written reasoning rather than producing an unexplained number. The output is a ranked shortlist where every score can be traced to evidence in the CV. Crucially, no candidate is rejected automatically: a human reviews the shortlist and the borderline band, and makes every rejection decision, which is what keeps the process on the right side of UK GDPR Article 22 and EHRC guidance. The fully automated rejection letter is the trap this design exists to avoid.
The exact tools change per business. The shape does not.
What it needs
Honest inputs, nothing exotic
- 01The genuine requirements for the role, separated from the nice-to-haves
- 02Wherever applications land: ATS, inbox or job-board export
- 03A human reviewer who owns the final decisions
- 04A privacy notice for candidates that mentions automated screening, which we help you word
The payoff
What you get back
Every application gets read with the same attention as the first one of the day. Shortlists land in hours instead of weeks, each score comes with reasoning you could defend, and the legal exposure of automated rejection never enters the building.
Do it yourself
How you would build this yourself
No course, no upsell. This is the order we would build it in, with the tools named, and a prompt to start from.
- 1
Write the scoring rubric before you look at a single CV: the genuine requirements, weighted, separated from the nice-to-haves. This step also fixes most human screening.
- 2
Strip identifying details before the AI sees anything: name, age, address, photo. Claude can do the redaction as a separate first pass.
- 3
Score each CV against the rubric with written reasoning per requirement, never just a number. If the reasoning does not point at evidence in the CV, the score is noise.
- 4
Keep a borderline band and read it yourself. The model’s unconfident middle is where interesting candidates hide.
- 5
Never automate the rejection. A human makes every no, which is what keeps you on the right side of UK GDPR and EHRC guidance, and your candidate privacy notice needs to mention automated screening.
I am hiring a [role]. Here is the scoring rubric: [requirements with weights]. For each CV I give you, first strip name, age, address, photo and anything else identifying. Then score it against each rubric line with a sentence of reasoning pointing at specific evidence in the CV, plus an overall ranking. Mark any CV you are torn on as borderline rather than forcing a score. Do not recommend rejections: produce the ranked list and the reasoning, the decisions are mine.
Copy it into Claude Code, fill the brackets, and it will plan the build with you before writing a line of code.
We would rather show you how than bill you. The whole ladder of free help, answers, guides and the weekly build-along, is on the do-it-yourself page.