How to screen resumes and candidates with AI

By the end of this guide you will have a screening process in which AI reads incoming applications, scores them against criteria you wrote in advance, and gives you a ranked shortlist with reasons, while a person still makes every decision about who moves forward. The job is done by an AI recruiting or applicant-screening tool, sometimes connected to the applicant tracking system (ATS) you already use.

To screen resumes with AI responsibly, the criteria matter more than the software. This guide covers writing those criteria, choosing between lightweight scorers and full recruiting platforms, testing the results and keeping a human in charge. It compares seven tools from our Automation and agents category.

Key takeaways

• Write the scorecard first (skills, experience, knockout questions), let the AI rank applicants against it, then have a person review the shortlist and the rejections before anyone is contacted.
• The main decision is size: Workcraft is a pay-as-you-go screener for teams without a recruiter, while Paradox and Humanly target high-volume hiring and quote prices after a demo, and Eightfold quotes enterprise pricing per organization.
• Scores can be wrong. Workcraft's own listing says its scores are an aid to human review, and AI should never be the sole decision-maker.
• Hiring law and fairness rules vary by country and region, so check your obligations and keep records of how candidates were assessed.

What you need before you start

• A clear job description. Skills, experience and the few things that are truly required.
• A scorecard. The criteria and how much each matters, written before you see any applicants.
• Applicant volume. Rough numbers of applicants per role, because this decides whether a small tool or a full platform fits.
• An ATS or a place to collect applications. Some tools build the apply page, others sit on top of your system.
• Time. Setting up one role takes under an hour in a lightweight tool. Platform rollouts take longer.
• Difficulty. Easy to moderate. The harder part is writing fair, job-related criteria.
• Legal awareness. Automated screening is regulated in some places. Check local rules on automated decision-making, candidate notice and consent before you switch it on.

Which kind of AI tool fits this job

A lightweight scorer for teams without a recruiter. Workcraft turns a role description into an editable scorecard, builds an apply page with role-specific questions, then extracts and scores each resume and answer, with a written evaluation per skill. It also does interview scheduling and a shared feedback form. It does not find candidates or run AI video interviews, and its pricing is pay as you go.

An AI recruiter that screens and sources. HiringAgents.ai uses a screening agent that handles every applicant, runs video intros and talks to candidates by email and phone, plus a sourcing agent for passive candidates. It syncs profiles into ATS systems such as Greenhouse, Ashby and Lever. Billing is by credits, one per ready-to-interview candidate.

Conversational recruiting for high volume. Humanly is built for high-volume, hourly and frontline hiring, with an AI Recruiter that replies to and screens applicants and an AI Interviewer for structured interviews. Paradox uses an assistant named Olivia to screen by text, chat and messaging apps and schedule interviews. Both quote prices after a demo.

Enterprise talent platforms and smaller hiring marketplaces. Eightfold AI is a talent intelligence platform whose AI Interviewer runs first-round interviews and flags suspicious behaviour. Noon AI is an agent platform for recruiters that sources, screens and ranks candidates. Airwork combines AI assessments, an applicant tracking system and a talent network mainly in South Asia and emerging markets.

Step 1: Write a job-related scorecard

You finish with criteria that apply equally to every applicant.

1. List the must-have skills and experience, and separate them from nice-to-haves.
2. Write knockout questions only for genuine requirements, such as a required licence or work authorisation.
3. Describe what good evidence looks like for each criterion.
4. Remove anything that is not related to doing the job, such as names, photos, age or schools used as shortcuts.

Check it worked by asking a colleague to score two sample resumes with your scorecard and comparing their results.

Step 2: Choose a tool that matches your volume

Pick the lightest tool that covers your needs.

1. If you hire a few roles a year without a recruiter, look at Workcraft's pay-as-you-go model.
2. If you want sourcing as well as screening, look at HiringAgents.ai or Noon AI.
3. If you hire in volume for hourly roles, look at Humanly or Paradox, and ask for a demo.
4. If you are an enterprise with skills and workforce planning needs, look at Eightfold AI.
5. Check whether the tool replaces your ATS or sits alongside it. Humanly can do either, and Paradox lists Workday, SAP SuccessFactors and Indeed integrations.

Check it worked by confirming the tool supports the way you receive applications and the systems you must keep.

Step 3: Set up the role and the screening questions

Turn your scorecard into the tool's configuration.

1. Create the role and enter the job description. Workcraft builds a scorecard from it, which you can edit.
2. Add application questions that match your criteria.
3. Set any knockout questions and check their wording.
4. Turn on only the features you need. Leave out video interviews or voice screening unless you have decided they are appropriate and told candidates.

Check it worked by previewing the apply page as a candidate and confirming the questions make sense and are job-related.

Step 4: Tell candidates how AI is used

Transparency protects candidates and you.

1. Add a short notice to the application page explaining that AI helps assess applications and that people make decisions.
2. Offer an alternative route for candidates who cannot use the process, such as a disability adjustment.
3. Make sure your privacy notice covers how applicant data is processed and stored.
4. Ask the vendor about data handling. HiringAgents.ai lists SOC 2 Type II and GDPR compliance, and Workcraft says it does not use customer content or resumes to train models.

Check it worked by asking a colleague whether they would understand from the page how their data will be used.

Step 5: Run a test batch and compare with a human review

Test before you rely on it.

1. Run a past batch of applications, or at least ten, through the tool.
2. Have a person rank the same applications using the same scorecard.
3. Compare the rankings and read the AI's written reasons.
4. Look for patterns that suggest unfair filtering, such as strong candidates with unusual backgrounds being ranked low. Adjust the scorecard.

Check it worked by finding that the top picks on both lists mostly overlap and that you can explain every disagreement.

Step 6: Review the shortlist and the rejections

The AI proposes, and you decide.

1. Read the ranked shortlist with the reasoning for each candidate.
2. Sample the lower-ranked applicants, not just the top, to catch good candidates who were scored down.
3. Use the tool's transcripts or evaluations as input to your judgement. Humanly provides ranked shortlists with transcripts and reasoning.
4. Never reject anyone on an AI score alone.

Check it worked by being able to justify each rejection with a reason tied to the scorecard.

Step 7: Schedule interviews and keep records

Finish by moving good candidates forward and documenting the process.

1. Use the tool's scheduling features if they suit you. Workcraft offers one-click scheduling and a shared feedback form, and Paradox schedules and reschedules interviews and sends reminders.
2. Keep interviewer feedback in one place, rated against the same scorecard.
3. Record the criteria, the tool's output and the human decision for each applicant, in line with your local record-keeping rules.
4. Review results after each hire and update the scorecard.

Check it worked by retrieving a full record for one applicant, from application to decision.

Tools for this job compared

Tool	Good fit for	Pricing	Stated limits
Workcraft	Teams without a recruiter (web and API)	Freemium	Does not source candidates; no AI video interviews; scores can be wrong and are an aid to human review
HiringAgents.ai	Screening plus sourcing with ATS sync (web)	Freemium, paid from $149, usage-based (checked 24 Oct 2026)	Starter and Growth are charged per credit, so cost depends on interviews delivered; Business needs sales
Humanly	High-volume, hourly and frontline hiring (API)	Contact sales	Prices not published; aimed at high-volume hiring rather than small-scale executive search
Paradox	Conversational screening and scheduling by text, chat and messaging apps	Contact sales	Pricing quoted per employer after a demo; aimed at employers with ongoing hiring volume
Airwork	AI assessments and a talent network (web)	Freemium, paid from $49, usage-based (checked 24 Oct 2026)	Free plan: one free job per year and up to 10 interviews per job; talent sourced mainly from South Asia
Eightfold AI	Enterprise talent intelligence and AI interviews	Contact sales	Pricing not published and quoted per organization; platform and AI Interviewer priced separately
Noon AI	Recruiters sourcing and screening with agents	Contact sales	Quote-based and requires booking a demo; built for employers and recruiters, not job seekers

Prices and plan limits change, so confirm the current terms on each vendor's site before you commit.

Common mistakes to avoid

• Screening without written criteria. If you do not define what good looks like, the tool or your own bias decides. Fix it by writing a job-related scorecard before the first application arrives.
• Letting the score make the decision. AI scores can be wrong or reflect patterns in the data. Fix it by treating scores as input, reviewing rejections and keeping a person accountable for each outcome.
• Not telling candidates. Candidates may have a right to know that automation is used. Fix it by adding a clear notice and an alternative route, and by checking local rules.
• Choosing a platform too big for the job. Quote-only enterprise tools suit ongoing volume, not an occasional hire. Fix it by matching the tool to your real number of roles and applicants.
• Never testing for unfair patterns. A tool can quietly disadvantage groups of people. Fix it by comparing outcomes with a human review on past data and checking who gets screened out.

How we put this together

This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.

Frequently asked questions

Is there a free way to screen resumes with AI?

Partly. Workcraft's first job and first five rankings are free, and Airwork's Free plan covers one free job per year with up to 10 interviews per job. After that, Workcraft costs $99 per job and $1 per applicant. Most enterprise tools are quote-only.

Can AI screen resumes accurately and fairly?

It can speed up the first pass, but it can also be wrong or biased. None of the listings promise unbiased results, and Workcraft says its scores are an aid to human review. Use clear criteria, test against human judgement and keep people responsible for decisions.

Is it legal to use AI for hiring decisions?

It depends on where you hire. Some places regulate automated employment decision tools, require notices or audits, or restrict decisions based only on automation. This guide cannot give legal advice, so check the rules in your location with an employment lawyer or regulator.

How long does AI resume screening take?

Vendors do not publish setup times, but a lightweight tool such as Workcraft starts from a role description and your questions. After that, screening runs as applications arrive. Tools such as Humanly and Paradox describe responding to candidates around the clock, but your own review of the shortlist still needs time.

What happens to candidate data?

It varies. Noon AI lists GDPR and SOC 2 Type II compliance, and Workcraft says it does not use resumes to train models. Read each vendor's privacy terms, tell candidates how their data is used and avoid uploading more than you need.

Compare more options in our Automation and agents category.