Screening CVs with AI: reading for meaning, not keywords

One hiring round at your company can bring in several hundred to a thousand resumes. By the fiftieth resume, the HR person's eyes are tired, and candidates who applied last often get only a skim. AI CV screening solves exactly this: the machine reads each resume for meaning and scores it against the role's requirements with evidence, so nobody is overlooked just because they applied late. In this article we break down how a machine really reads a CV and how to use it properly.

How AI CV screening differs from old keyword filters

Crude filters from ten years ago only matched words. You set the keyword "sales management," and any resume without that exact phrase got rejected, even if the person wrote "in charge of regional revenue" and meant the same thing. This wrongly rejects many strong people and keeps those who know how to stuff keywords.

AI CV screening reads for meaning, not for the exact words. The machine understands that two different phrasings can describe the same skill, and conversely that two people who both write "team lead" may have very different scopes of work. That makes candidate rankings much closer to reality. It is one link in the AI recruitment automation picture we build for businesses.

In short: old tools count words, AI understands meaning. That is the decisive difference between a filtered list full of keyword stuffers and one with people who can actually do the job.

The steps a machine takes to extract and score a resume

To trust the result, you need to understand what the machine does inside, rather than hand your trust to a score from nowhere.

First the machine reads the file, including CVs that are photos or messily formatted PDFs, then splits it into parts: experience, skills, education, employment timeline, and achievements with numbers. Then it compares each part with the role requirements you set, scores each axis, and combines them into a total score with a threshold. Most important, every score comes with a quote of the exact line in the CV as its basis.

Because scoring comes with evidence, your reviewer can read one summary line and understand why one resume ranks above another. You keep the right to overturn it, and the machine doesn't dictate. This transparent scoring is what we also apply to the automated AI interview in the next round.

An honest warning: machines can still be biased if done carelessly

We won't sugarcoat it. AI CV screening has a real risk: bias hidden in the training data. If past data leans toward one group, a badly trained machine can unfairly reject others.

This is not an idle worry. Harvard Business Review has analyzed how automated filters wrongly reject many qualified candidates because of rigid criteria. McKinsey classifies data bias as a major risk when using AI, and Gartner recommends always keeping humans in HR decisions. That is why a well-built system never lets the machine decide alone. Resumes near the pass/fail line are always reviewed by a person. Scoring criteria are set around the actual job, not the candidate's school or where they live. And you should periodically check whether the machine is rejecting one group disproportionately.

In other words, AI helps you read faster and read more, but responsibility for fairness stays with the people running it. A serious vendor will tell you about this risk up front instead of promising the machine is always right.

How to set the filter threshold so you don't lose strong people with weak CVs

A common mistake is setting the threshold too tight to keep the list short. The result is that you also lose people who can do the job but don't know how to write a CV.

We recommend two levels. Those above the threshold get invited right away. Those just below it shouldn't be rejected outright; collect them into a waiting list for a person to glance over, because that is where the rough diamonds often sit. MONA has many times hired very strong people from this very group of unimpressive CVs, because their real ability only shows when they work, not on paper.

This soft threshold keeps you from missing talent while still saving time on most resumes. It is also the natural bridge to the preliminary conversation with candidates via the recruitment chatbot before in-depth interviews.

Not just office hiring: high-volume industries benefit most

The more resumes per position, the more AI CV screening is worth.

• Factories hiring workers in waves: thousands of resumes for a single expansion, just like the line that screened 1,000 worker resumes in 3 hours.
• Retail chains and multi-branch F&B: constant hiring and high turnover, needing fast filtering that stays consistent.
• Businesses opening agents and distributors: the same approach also filters 1,000 applications to pick 10 partners.
• Spa chains, clinics, training centers: hiring technicians and advisors for peak seasons.

What these industries share is a resume volume far beyond human reading capacity. That is where AI carries the load, so HR can put its effort into judgment, as in the overall applying AI in the HR department and the broader business automation direction we pursue.

MONA screens CVs with the same system it sells

We don't promote what we haven't used. MONA is hiring Account staff with this very screening system, running on real ads and reading and ranking real resumes every day.

The machine reads the CV, scores it against the role requirements, keeps the near-threshold group for a panel to rescue, then moves qualified people into the automated interview round. The whole outer round is about 80% automated, and people only review and meet candidates in the final round. This is one piece of the a recruitment AI Agent in the AI Agent suite that MONA built for itself. For training people after they are hired, you can look at Khánh Hùng Academy.

One honest piece of advice: don't stitch together loose drag-and-drop tools to build your own CV filter. It's fine at low volume, but at peak season when thousands of resumes pour in, one wrong setting clogs the whole flow and strong candidates slip away. MONA builds custom software that holds up day and night, with a team accountable when something breaks. If you want a fuller HR system, the team Mona Software also builds HR management software with built-in AI and recruitment management software that follows your own process, including a recruitment website as the gateway for receiving applications.

AI resume reading and scoring: text extraction, criteria matching, dynamic score thresholds

In MONA's July 2026 hiring drive, 400 resumes came in over two weeks. Reading that by hand would take 4 to 5 HR people, and by the hundredth resume everyone's eyes glaze over and the right person is easy to miss. The CV screening machine never tires, and works in 3 steps.

Step one, extract the text: CVs arrive in every form, PDF, Word, and some people even photograph a paper CV and send it through chat. The system converts everything into machine-readable text, and for image files it uses optical character recognition to read the text in the picture. Step two, match against criteria: the recruiter prepares a criteria sheet for the role, such as years of experience, required skills and must-have requirements. The AI checks each criterion against the CV, states clearly where it passes or fails, and must point to the line in the CV as its basis rather than saying vaguely "suitable." Step three, score against a dynamic threshold: the total score is compared with the interview threshold, and the threshold is adjusted for urgency. When people are needed fast it goes down, for example inviting anyone from 50 points. When the team is almost full it goes up to 70 to keep only the best group. The final decision is still the recruiter's; the machine hands over a scored list with its basis.

To try the text extraction step yourself, you can convert a scanned PDF to Word with MONA's free tool and see how much text the machine can read from a scan.

Rejected candidates get a proper reply too: the system sends a polite rejection email at every stage. Many candidates aren't upset about being rejected, but about submitting and never hearing back, and the machine fixes exactly that.

Frequently asked questions about AI CV screening

Is AI CV screening more accurate than a human reader?

The machine reads consistently and doesn't tire, so it is more even at high volume. But final accuracy depends on how you set the criteria and on people reviewing near-threshold resumes. The machine assists, people decide.

Can a machine read a badly formatted or image-based CV?

Today's systems can read messy PDFs and photos thanks to optical character recognition. Still, the clearer the resume, the more reliable the result, so it is worth encouraging candidates to submit a clean version.

How do I know the machine isn't wrongly rejecting strong people?

Prevention is a soft threshold, keeping the near-threshold group for people to review, and periodically checking whether the machine skews against any group. The machine is not allowed to reject outright on its own.

Is candidate data kept secure?

A serious system stores and processes data inside the business's own system and masks personal information as regulations require. You should ask the vendor to commit to this clearly.

What you can try this week

Take the pile of resumes from your latest hiring round and count two numbers: how many hours it took to read them all, and how many late applications almost no one read. Those two numbers are what AI CV screening gives back to you.

Once that is clear, we invite you to book a call with MONA to see the screening system running for real. If you want the overall picture first, read What is AI recruitment first to get the big picture.

The whole screening process in this article has been packaged by MONA at the AI recruitment software Gấu Tuyển Dụng, with live running numbers on the page.