AI Resume Screening That Actually Works
83% of companies plan to use AI to screen resumes (ResumeBuilder.com, 2024). But most just count keywords. CurriculoATS evaluates what candidates actually accomplished.
What’s Wrong With Current AI Screening
Most “AI” ATS Tools
Parse resumes for keywords, count matches, rank by frequency. Fast but inaccurate. Misses strong candidates who use different terminology. Over-ranks keyword-stuffed resumes. 75% of qualified candidates are rejected.
Curriculo ATS AI Screening
Extracts structured data. Evaluates measurable outcomes. Scores candidates 0–100 across role fit, impact, and relevance. Surfaces high-performers regardless of resume format.
How AI Screening Works
Upload or Receive
Post a job and candidates apply. CurriculoATS accepts resumes in any format, PDF, DOCX, or plain text. No special formatting required.
AI Parsing
Our NLP engine extracts structured data from every resume: skills, experience, achievements, timelines, and quantified results. No special formatting required from candidates.
Signal Analysis
Each candidate is scored across five dimensions:
- Quantified Achievements, Revenue generated, users acquired, costs reduced, teams scaled
- Scope of Responsibility, Team size, budget managed, strategic vs tactical work
- Career Trajectory, Progression speed, role growth, increasing scope over time
- Skills Alignment, Match between candidate skills and role requirements
- Narrative Clarity, How clearly the candidate communicates their impact
Ranked Results
Candidates are ranked by composite Impact Score (0–100). Top candidates are surfaced in under 20 minutes for 50+ applicants. No manual screening required.
AI Screening By the Numbers
AI Screening vs Competitors
| Capability | Curriculo ATS | Greenhouse | Lever | Workable |
|---|---|---|---|---|
| Screening Method | Signal-based scoring | Keyword matching | Basic AI matching | AI sourcing |
| Impact Score | ✓ (0–100) | ✗ | ✗ | ✗ |
| Multi-format parsing | ✓ | ✓ | ✓ | ✓ |
| Bias reduction | Signal-based (outcomes) | Structured hiring | DEI dashboards | Basic |
| Screening speed (50 applicants) | <20 minutes | Manual + filters | Manual + filters | AI-assisted |
| Free plan with AI | ✓ | ✗ | ✗ | ✗ |
When to Use AI Screening
High-Volume Screening
Posted a role and got 200 applicants? CurriculoATS ranks them all in minutes. No need to manually review 200 resumes. The AI surfaces your top candidates so you can focus on interviews, not inbox triage.
Quality Over Keywords
Hiring for a role where results matter more than certifications? Impact scoring finds candidates who’ve driven real outcomes, revenue growth, team scaling, product launches, regardless of how they phrase it on their resume.
First-Time Hiring
Never screened resumes before? CurriculoATS’s AI does the heavy lifting. Just review the top-ranked candidates. No hiring experience required, the system evaluates candidates the way experienced recruiters do.
What the Law Now Asks of AI Screening
Screening software that ranks or filters applicants is regulated in a growing number of places, and the duty usually lands on the employer rather than the vendor. If you hire in New York City or anywhere in the EU, two regimes decide what you have to be able to show. Both of them reward the same thing: a written record of why a candidate was ranked where they were.
What follows is a plain summary of published rules, not legal advice. Check your own obligations with counsel before you rely on any of it.
New York City Local Law 144
If you use an automated employment decision tool to screen candidates or evaluate employees for promotion in New York City, the tool needs an independent bias audit every year. You publish a summary of the results on your own website, and you tell candidates the tool is being used before you use it.
The audit has to break results down by race, ethnicity and sex. Penalties run to $500 for a first violation and between $500 and $1,500 for each one after that, and every day a violation continues counts separately.
Enforcement is getting sharper. A New York State Comptroller audit published in December 2025 found the Department of Consumer and Worker Protection had enforced the law ineffectively, and employment firms have since told clients to expect more investigations through 2026.
The EU AI Act
Annex III of the EU AI Act puts recruitment and selection systems in the high-risk category. That covers placing targeted job adverts, analysing and filtering applications, and evaluating candidates, which is most of what a modern ATS does once AI is involved.
High-risk duties include risk management, data governance, technical documentation, logging, human oversight and transparency to the people using the system. Fines for a deployer that misses them reach 15 million euros or 3 percent of global annual turnover, whichever is larger.
Watch the date. These obligations were originally set for 2 August 2026, then moved to 2 December 2027 under the Digital Omnibus agreement. Plenty of guidance still quotes the old deadline, so confirm the current timeline before you build a plan around it.
Where the duty actually sits
Under Local Law 144 the employer commissions the audit, publishes the summary and notifies candidates. No vendor can do those things for you, and any vendor claiming its product makes you compliant is overstating what a product can do.
What a vendor can do is make the work possible. An auditor needs to know how each candidate was scored and on what basis, and a candidate who asks why they were ranked low deserves an answer that is not "the algorithm decided". A screening tool that produces a written reason for every score gives you both, which is the difference between an audit you can complete and one you cannot.
What Sits Behind an Impact Score
A keyword filter answers one question: does this resume contain the words in the job description. That question is easy to answer and easy to game, which is why a candidate who has shipped the exact thing you need gets dropped for calling it something else. Impact scoring asks a harder question: what did this person actually do, and how close is it to the job in front of them.
Reading, not matching
Every resume is parsed and read for meaning rather than string overlap. Search runs on a mix of semantic understanding, full text matching and language processing, so a search for a backend engineer with cloud experience still surfaces the candidate whose resume says AWS infrastructure developer.
Five signals, not one
Each applicant is evaluated across five signal categories: quantified achievements, scope of responsibility, career trajectory, skills alignment and narrative clarity. Those combine into a single composite score from 0 to 100, so you can sort a pile of 200 without reading all 200. If you want the full cost picture behind a hiring stack that does this, the true cost of ATS ownership breaks down where the money actually goes.
A paragraph you can read
Every score comes with written reasoning explaining what the model picked up on. That paragraph is what makes the score arguable. You can see when it has latched onto the wrong thing, and you have something concrete to hand an auditor or a candidate who asks.
You keep the decision
The score orders your queue, it does not reject anyone. If you disagree, override in one click and move the candidate wherever you want. No applicant is ever auto-rejected on a score, which keeps a human in the loop where both regimes above expect one.
How to Evaluate an AI Screening Tool
Almost every ATS on the market now says it uses AI. The word covers everything from a keyword filter with a new label to a model that reads for meaning, and the demo will rarely tell you which one you are looking at. These are the questions that separate them, and they work on any vendor including this one.
Ask to see the reasoning
Any tool can print a number next to a name. Ask what the number is based on, for a specific candidate, in words. If the answer is a confidence percentage or a list of matched keywords, you are looking at a filter. If nobody can explain a single score, you will not be able to explain it to an auditor or a rejected applicant either.
Find out what happens to a low score
There is a large difference between ranking a queue and rejecting a person. Ask whether the system can auto-reject, whether that is on by default, and whether you can turn it off. Automated rejection is what turns a screening tool into a decision maker, and it is the part regulators care about most.
Test it on a hire you already made
Take a role you filled last year, load the real applicants, and see where the person you actually hired lands. This is the single most useful hour you can spend on an evaluation. A tool that buries your best hire at rank 40 is telling you something no demo will. Track what it does to your time to hire once it is live, since that is the number the tool is supposed to move.
Ask who owns the bias audit
If you hire in New York City, somebody has to commission an independent audit every year and publish the summary. Ask the vendor plainly whether they provide audit documentation, what they will hand an auditor, and what they expect you to do yourself. Vague answers here become your problem later.
Check the exit before you enter
Ask how you get your candidate data out, in what format, and how long it takes. A vendor that can export your full pipeline to CSV on the day you cancel is one you can leave. Migration difficulty is the quiet reason teams stay on tools they stopped liking years ago.
Price the whole thing, not the sticker
Per-seat pricing means the cost of adding an interviewer to a hiring panel is a line item, so teams start sharing logins and your audit trail stops meaning anything. Work out what the tool costs with everyone who should see a candidate actually in it, then compare. Our pricing page lists every plan limit, and the comparison guide puts the main options side by side.
Is AI screening included in the free plan?
Yes. AI screening with Impact Scoring is on every plan, including the free Starter tier, and there is no credit card required to use it. Starter covers unlimited job posts and up to 1,000 candidates ranked by Impact Score, which is enough for most teams making their first few hires. Pro is $100 a month, currently $50 during the early-bird window, and raises the candidate limit to 10,000. Neither plan charges per seat, so you can put every interviewer on your hiring panel into the account without changing what you pay. Screening is not gated behind an upgrade, a trial clock or a usage credit. The free tier runs the same scoring engine as the paid one, so what you evaluate is what you get.
How accurate is the AI?
Accuracy is the wrong question to ask any screening vendor, because nobody can tell you in advance how well a model will rank candidates for your specific role. The useful question is whether you can check its work. Every Impact Score comes with a written paragraph explaining what the model picked up on, so you can see when it has weighted the wrong thing and say so. If you disagree, override in one click and move the candidate wherever you want. No applicant is ever rejected automatically on a score. The practical test is to load a role you already filled, run the real applicants through it, and check where the person you actually hired lands. That tells you more about fit for your hiring than any vendor accuracy figure.
Can AI screening be biased?
Any system that ranks people can carry bias, and a vendor who tells you otherwise is selling something. Impact scoring is built to reduce one specific and common source of it: proxies. Instead of weighting school name or employer prestige, the score reads measurable outcomes such as revenue generated, teams scaled and delivery timelines met. No demographic data is used in scoring at any point. That does not make it bias-free, it makes it auditable, because every score carries written reasoning you can inspect and challenge. If you hire in New York City you must commission an independent bias audit of any automated employment decision tool each year and publish a summary. That duty sits with you as the employer, and a written reasoning trail is what makes the audit possible to complete.
What resume formats are supported?
PDF, DOCX and plain text, which covers effectively every resume a candidate will send you. Nothing special is required at their end: no template to download, no fields to fill in, no account to create before applying. Parsing reads the document for meaning rather than matching strings, so a candidate who describes the same work in different words is still surfaced. Applications arriving by email are picked up automatically once you connect Gmail or Outlook, so a resume attached to an ordinary reply lands in the pipeline without anyone copying it across. That matters more than format support in practice, because the resumes teams lose are usually the ones sitting in an inbox nobody remembered to check, not the ones in an unusual file type.
How fast is the screening?
Most teams get through 50 or more applicants in under 20 minutes, and each individual resume is processed in seconds. The time saved is not really in the reading, though. It is in not having to decide where to start. A ranked queue with a written reason attached to each score means you open the strongest applications first instead of working through an inbox in the order things arrived. Setup takes about 15 minutes from signing up to your first scored candidate: paste a job description, connect Gmail or Outlook, invite your team and post the role. Scoring runs as applications come in, so a role posted in the morning already has a ranked shortlist by the time you look at it again.