AI screening
Also called: AI resume screening, AI CV screening
How it works under the hood
Two common approaches:
- Keyword and skill matching: the model extracts skills, years of experience, and seniority signals from the CV and scores against the role’s must-haves. Fast and explainable.
- Embedding-based matching: the model converts CV and role description into vectors and scores semantic similarity. Catches relevant experience the keyword approach misses (a “lead React developer” matches “senior frontend engineer”).
Most modern ATS systems combine the two.
What it does well
- Compresses CV review time from 1-3 minutes per application to 10-20 seconds for the reviewer to confirm the AI’s read.
- Surfaces relevant candidates the human reviewer might skip because of a non-standard career path or under-marketed CV.
- Stays consistent — applies the same criteria to candidate #1 and candidate #100.
Where it fails
- Bias inheritance: if the model was trained on historical hiring decisions and those decisions were biased, the model reproduces the bias at scale.
- Off-spec creativity: AI screening over-weights conventional CVs. A career-changer with strong skills but unconventional path can score low.
- Gaming risk: candidates aware of AI screening optimize their CVs against it. The signal degrades.
How to use it sensibly
Treat AI screening as a ranking aid, not a filter. The human reviewer still sees every application; the AI score reorders the queue. Anything auto-rejected is a process risk.
Where Join fits
Join’s AI screening surfaces ranked candidates with explanations per score — the reviewer sees why the AI ranked, not just the number. Auto-rejection is opt-in and explicitly logged. See the features page.
