Every ATS vendor now claims to use AI. Most of them are lying — or at least, stretching the truth significantly. Behind most "AI screening" features is a keyword-matching algorithm that's been around since the 1990s. It finds the word "Python" in your job description, finds the word "Python" in a CV, and calls that a match.
Real AI screening works differently. This guide explains what AI candidate screening actually does, how to use it practically, and what you should still be doing yourself.
What keyword-based screening actually does
Traditional ATS screening counts word overlaps between your job description and the CV. A candidate who has written "Python, Flask, REST APIs, PostgreSQL" in their skills section scores highly if your JD contains those same words. A candidate who has "built scalable data pipelines in Python for 6 years" but listed it differently might score lower.
The problems:
- Keyword stuffing beats actual skill. Candidates who know how ATS systems work optimise their CVs with exact keyword matches, not necessarily genuine experience.
- Semantic gaps. "Node.js" and "NodeJS" are the same thing. "5 years experience" and "senior backend engineer" describe the same seniority. Keyword systems often miss these.
- Context blindness. "Managed a team" can mean managed 2 interns for 3 months, or managed a 20-person engineering org for 5 years. Keyword search can't tell the difference.
What real AI screening does
AI screening powered by large language models (LLMs) reads each CV the way a human recruiter would — understanding context, inferring seniority, recognising related skills, and comparing the candidate holistically against the job requirements.
Concretely, when you run an AI match in a system like Rekvo, the AI:
- Parses the job description — identifying required skills, seniority level, domain experience, and implicit requirements (a "fast-paced startup" implies different things than "established enterprise")
- Reads each CV — extracting experience, skills, seniority signals, and career trajectory
- Scores the match — not by keyword count but by how well the candidate's actual experience maps to what the role needs
- Generates a verdict — Strong hire / Consider / Pass — along with matched skills, experience fit score, and a gap analysis explaining where the candidate falls short
The key difference: the AI understands what you mean, not just what words you used. For a deeper technical explanation of how this works, see What is RAG in Recruiting.
How to screen 100 CVs in under 5 minutes
Here's the practical workflow:
Step 1: Upload CVs Upload all incoming applications to your CV library. Rekvo accepts PDF, Word, and plain text formats. Drag and drop or bulk upload — CVs are parsed automatically.
Step 2: Write or paste your job description You can use an existing JD or generate one with Rekvo's AI writer. The JD is the input the AI uses to score candidates — the more specific it is, the better the ranking.
Step 3: Run the AI match Click "Run AI Match" and select the job. Within 1–2 minutes, every candidate in your library is ranked from strongest to weakest, with a full breakdown for each one.
Step 4: Review the shortlist You don't read 100 CVs. You read the top 10–15 candidates the AI has already identified, review their breakdowns, and decide who to move forward with.
Step 5: Ask follow-up questions If you want to dig deeper into your candidate pool, use conversational search: "Who has experience with distributed systems and has worked at a fintech company?" The AI answers from your entire pool in seconds.
What AI screening still can't do
AI screening dramatically reduces the time spent on CV review, but it doesn't replace human judgement entirely:
- It can't assess culture fit. The AI doesn't know whether someone's working style suits your team.
- It can't verify claims. A candidate who lists "led a team of 20" needs a reference check — the AI takes the CV at face value.
- It can miss niche signals. Domain-specific signals that aren't common language in job descriptions may not be weighted correctly.
- It's a ranking, not a verdict. A "Strong hire" from the AI means the candidate looks good on paper. The interview still matters.
The right way to think about AI screening: it handles the filtering work (the 90-minute job of reading 100 CVs) and lets you focus on the evaluation work (the 60-minute job of interviewing the right 5).
FAQ
Does AI screening discriminate against candidates? Good AI screening is blind to demographic signals that a human reader might pick up unconsciously. It evaluates experience and skill. However, no system is perfectly neutral — if your past hires have had a particular background, a model trained on your data might inherit that bias. Rekvo uses a general-purpose LLM, not a model trained on your historical decisions.
Will candidates game AI screening like they game keywords? The attack surface is much smaller. With keyword systems, you can game the system by listing the right words. With LLM-based screening, you'd need to write a genuinely convincing description of experience you don't have — which is harder to do and easier for a human to catch in an interview.
How many CVs can it handle? Rekvo's Business plan handles up to 15,000 CVs in your library. An AI match run against 200 candidates takes under 2 minutes.
Comparing ATS tools before committing? Read Best ATS for Startups in 2026 for an honest breakdown.