Recruitment Screening With AI — What It Sorts, What Your Team Still Decides
By Akshit Agrawal, Co-founder, Bitvyas · Published 23 October 2026 · Facts as at 23 October 2026
At Bitvyas, an AI screening workflow takes over the first pass on applications. It reads CVs, extracts the facts, checks them against a scorecard you wrote, books interviews and keeps candidates informed. A recruiter still decides who is shortlisted and who is rejected. That human step is also what regulators in the UK and New York City expect.
Screening is a good fit for automation because the volume is high and the first-pass rules are mostly written down. It is also the process where it matters most that a person stays in charge. This post follows our four-question test from the hub post on what to automate with AI first and maps the work step by step. It sits alongside our posts on AI collections and AI insurance renewals. The workflow itself is described on our recruitment use-case page.
What are the key facts on AI in hiring?
As of 23 October 2026. The rules differ by country; check yours before you switch a tool on.
| Fact | Detail | Source |
|---|---|---|
| New York City | Employers using an automated employment decision tool need a bias audit within one year before use, must publish a summary, and must give candidates at least 10 days' notice | Perkins Coie summary of the final rules (adopted 6 Apr 2023) |
| New York City penalties | Up to $500 for a first violation; $500–$1,500 for each later one | Same |
| UK | The ICO reviewed automated decision-making in recruitment across 30+ organisations (Mar 2025–Jan 2026). It found tools described as decision support that in practice made solely automated decisions | Brodies LLP article, 14 May 2026 |
| UK expectations | Tell candidates when automation is used, test for bias, allow challenge, complete a DPIA, question vendors | Same |
| India | Not found: a rule specific to AI screening in hiring. Personal data in CVs falls under the Digital Personal Data Protection Act, 2023 | India Code |
What does screening look like when it is done by hand?
A recruiter opens a few hundred CVs, skims each for the must-haves, copies details into a tracker, emails a shortlist to the hiring manager and chases candidates about availability. Good candidates wait while the pile is read. Candidates who are rejected often hear nothing.
The cost is mostly time and consistency. The same CV can be judged differently on a Monday morning and a Friday evening.
What can AI do on its own in screening?
The repeatable first pass. A workflow like this runs inside your applicant tracker, shared inbox or job-board feeds:
- Collect applications from email, job boards and your careers page into one list, and remove duplicates.
- Parse each CV into fields: roles, dates, skills, qualifications, location, notice period.
- Check must-haves. Rules you set, such as a required licence or years in a role, are checked against the extracted facts. Every result shows the line of the CV it came from.
- Score against a scorecard. The scorecard is written by the hiring manager before the role opens. The AI shows its reasons for each score. It does not rank on name, age, gender or photo.
- Flag, don't reject. Borderline and unusual profiles go to a person with the reasons attached.
- Handle logistics. It sends acknowledgements, proposes interview slots, reschedules and sends reminders.
- Keep candidates informed with status updates written in your tone.
What stays with a person?
- Shortlist and reject decisions. A recruiter reviews the AI's reasons and makes the call. This is the part regulators care about.
- The scorecard itself. What "good" means for the role is a hiring-manager decision.
- Interviews, offers and references.
- Exceptions, such as career changers or gaps that a rule would score low.
- Bias checks. Someone reviews outcomes by group on a regular schedule and keeps a record.
How do you check the human step is real?
If a reviewer approves every AI recommendation in seconds, the review is not meaningful. Track three things: how often reviewers change the AI's recommendation, how long they spend per candidate, and whether rejected candidates ever get a second look. These are our suggested controls, not a legal test.
What should you ask a vendor, including us?
The questions the ICO lists for buyers are a good start: how the tool was trained, how it is tested for bias, how it will be deployed in your process and how decisions can be reviewed. Add where candidate data is stored, how long it is kept and whether candidates can opt out. Bitvyas builds these workflows, so treat our own answers as a vendor's and ask for the same evidence from everyone.
How do you decide if it is worth it?
Count the applications per role, the minutes spent on the first pass and how long candidates wait for a reply. Then compare that to the cost of building, running and reviewing the workflow. Our post on what AI automation costs shows how to work out payback with your own numbers.
This is general information, not legal advice. Rules on automated hiring tools change; confirm the current position for your location.
Sources
- Perkins Coie, NYC adopts final rules for law governing automated employment decision tools
- Brodies LLP, Automated decision-making in recruitment: what the ICO's latest report means for employers (14 May 2026)
- India Code, Digital Personal Data Protection Act, 2023
See it working
- See AI screen CVs and book interviews (live demo) →
- All AI automation use cases →
- What Bitvyas automates →
Keep reading
Still doing this by hand?
Bitvyas builds AI that takes this kind of work off your team. It reads the documents, does the matching and drafts the next step, inside the tools you already use. A person still approves anything that matters.
Tell us what your team repeats