- The screening problem nobody fixes by hiring more people
- The tools that do this well in 2026
- A five step way to set this up without wrecking your shortlist
- The part most guides on this topic skip
- Where this fits with the rest of hiring, not just the CV pile
- The candidate authenticity problem you'll run into within a year
- What this costs
- Frequently asked questions
- Useful references
Straight answer: AI screening tools like Manatal, Humanly, Paradox, and Fetcher can cut the time recruiters spend on first-pass CV review by 70 to 80 percent, but only if you set up the scoring rules yourself first. Hand that job entirely to the algorithm and you'll screen out good people faster, not smarter.
The screening problem nobody fixes by hiring more people
A single corporate job posting attracts around 250 applications on average, according to Glassdoor's own hiring data. Most of those never get read. A recruiter opens the first thirty, skims for keywords, and by CV number one hundred and fifty they're making decisions in under six seconds a page. That's not laziness. That's what happens to a human brain doing repetitive pattern matching for four hours straight.
I worked with a recruitment agency in Reading two years ago that had one internal recruiter covering thirty open roles at once. She was getting through maybe two hundred CVs a day and still missing good candidates because she was screening on autopilot by 3pm. We put an AI scoring layer on top of her ATS, built the shortlisting criteria with her (not for her), and her per-role screening time went from around six hours down to forty five minutes. She still made the final call on every shortlist. The AI just stopped her opening CVs that didn't meet the basic bar.
The tools that do this well in 2026
- Manatal - an ATS with built in AI candidate scoring that ranks applicants against the job spec automatically. Good for small agencies who don't want a separate screening tool bolted on.
- Humanly - a conversational AI that chats with candidates via text before a human ever sees them, asks screening questions, and flags dealbreakers early. Used heavily in high volume hourly hiring.
- Paradox (Olivia) - the conversational assistant McDonald's and other high volume employers use to screen and schedule at scale, handling thousands of applicants a week without a recruiter touching most of them until interview stage.
- Fetcher - AI sourcing that builds candidate pipelines automatically based on criteria you set, useful when the bottleneck is finding people, not screening the ones who already applied.
- Metaview - takes AI notes during live interviews so recruiters aren't scribbling while trying to listen, then summarises against your scorecard.
- Textio - not a screening tool, a job ad tool, but it matters here because biased job descriptions are what create biased applicant pools in the first place.
HireVue is worth mentioning because of what it stopped doing. It used facial analysis AI to score video interviews and dropped that feature in 2021 after sustained criticism, including from Illinois regulators, over what it was measuring. That history is a useful warning: just because a model can score something doesn't mean it should.
A five step way to set this up without wrecking your shortlist
- Write the must-have criteria down before you touch any tool. Three to five non-negotiables, nothing vague like "good communicator."
- Build a scoring rubric out of those criteria and feed it into your ATS or screening tool as the ranking logic, not the tool's default weighting.
- Run a test batch of fifty CVs you've already screened manually through the AI and compare its shortlist to yours. If it disagrees on more than 10 to 15 percent, your criteria are too loose.
- Set the AI to flag borderline candidates for human review rather than auto-reject them. Borderline is where the good surprises live.
- Audit the rejected pile monthly. Pull twenty rejected CVs and have a human check whether the tool got it right. If you're a small agency or one-person recruitment desk, this step alone is worth outsourcing to whoever wrote your prompts if you're using ChatGPT for the scoring logic, and this ChatGPT prompts list by job has a recruiting-specific set that's a decent starting point for building rubric prompts.
The part most guides on this topic skip
Here's the bit that doesn't get said enough: candidates are using AI on the other side too, and it's an arms race, not a one way filter. Tools that rewrite CVs to match job description keywords, ChatGPT prompts that "optimise for ATS," browser extensions that auto tailor a resume in seconds, these are all mainstream now, not niche hacks. That means keyword based screening, the thing most cheap AI tools still do at their core, is getting gamed faster than most recruiters realise. A CV that scores 95 percent match on paper can belong to someone who's never done the job, because they know exactly which words your filter is hunting for.
The fix isn't more keyword filtering, it's flipping the weighting toward skills evidence and structured questions the AI can't be reverse engineered against as easily, work samples, short task submissions, scenario questions. Humanly and Paradox both lean this way already. If your tool only does keyword matching against a CV, you're screening for who's best at prompting ChatGPT, not who's best at the job.
There's a second uncomfortable truth underneath that one. AI screening tools are trained on historical hiring patterns, and historical hiring patterns are not neutral. If your last fifty hires for a role skewed toward a particular university, background, or career path, an AI scoring model trained on "what good looked like before" will quietly keep reproducing that pattern unless someone deliberately checks for it. This is exactly why step five in that list above, the manual audit of rejected CVs, isn't optional admin. It's the only place bias creeping back in gets caught.
Where this fits with the rest of hiring, not just the CV pile
Screening is one bottleneck. It's not the only one, and it's worth being honest that speeding it up doesn't fix a broken pipeline further down. If your interview scheduling still takes five email round trips, or your hiring managers sit on feedback for two weeks, faster screening just means candidates wait longer at the next stage instead. Recruiters I've spoken to who've automated screening well tend to have also tackled scheduling automation at the same time, because a fast shortlist that then goes cold for three weeks does more damage to employer brand than a slow shortlist ever did.
It's also worth thinking about this from the industry angle, because what works for a high volume retail hiring team looks nothing like what works for a boutique agency placing senior finance candidates. If you're trying to work out what to automate first for your specific sector, this breakdown of what to automate first across 19 sectors covers recruitment alongside other people-heavy industries and is a useful sanity check before you buy anything.
The candidate authenticity problem you'll run into within a year
One thing recruiters using AI screening tools are starting to hit: verifying that a candidate's online presence, portfolio, or LinkedIn history is theirs and not AI-generated padding. LinkedIn started rolling out content credentials to flag AI-generated posts and images, and understanding what LinkedIn content credentials mean is becoming a useful, if slightly odd, part of due diligence when you're assessing a candidate's public work before an interview. Screening isn't just the CV anymore. It's the whole digital footprint, and that footprint is getting easier to fake.
There's a mirror version of this worth knowing too. Job seekers are increasingly using virtual assistants and AI tools themselves to apply faster and smarter, the same way recruiters are screening faster. If you want to understand what candidates on the other end of your funnel are doing to get through your process, this piece on hiring a job search virtual assistant is a revealing read from the applicant's side. It changes how you think about what a "strong" application signals now.
What this costs
Manatal starts around $19 per user per month for small teams and scales up from there. Humanly and Paradox are enterprise priced and usually require a sales conversation, expect four figures monthly minimum once you're at any real volume. Fetcher runs from roughly $500 a month for small teams. None of these are free trials worth skipping, run the 30 day trial most of them offer and test against real live roles before committing annually, because a tool that scores brilliantly on a demo dataset can behave completely differently against your actual applicant pool.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
If you're a small agency or an internal recruiting team of one or two people and none of this is your area of expertise, it's worth getting outside eyes on the setup before you build your scoring logic wrong and don't notice for six months. An AI implementation coach can save you from the expensive version of the learning curve, the version where you find out in month four that your rubric has been silently filtering out strong candidates the whole time.
Related: should you hire an ai consultant what they do.
Related: ai consultant recruitment agencies screening follow up.
Related: ai tools: guidelines and how to pitch.
Frequently asked questions
Do AI screening tools remove bias from hiring?
No, and any tool that claims this outright should make you cautious. AI screening reduces inconsistency between recruiters on a bad day, but it learns from historical hiring data, and that data carries whatever bias existed before. Monthly manual audits of rejected candidates are the only real check on this.
What's the best AI screening tool for a small recruitment agency?
Manatal is the most practical starting point for small teams because it's an ATS with AI scoring built in rather than a separate expensive layer, and it starts at around $19 per user a month with a trial available.
Can candidates tell if they've been screened by AI?
Usually yes, especially with conversational tools like Paradox's Olivia or Humanly, which openly identify themselves as AI during the chat. Most candidates now expect some level of AI involvement in early screening and aren't put off by it, but slow or generic follow up after an AI-driven first stage does damage employer brand.
Will AI screening tools stop working as candidates get better at gaming them?
Keyword based screening is already losing effectiveness because AI resume tailoring tools let candidates match job description language in seconds. The tools holding up best are ones weighting skills evidence and structured scenario questions rather than CV keyword matching alone.
