The Role of AI in Modern Recruitment
By Vriti AI Interview team · Published · 4 min read
AI is most useful in recruitment for repetitive, high-volume work: drafting, screening conversations, summarising and scheduling. It is least trustworthy when it makes a final decision about a person on its own. The role that works is a clear division of labour, in which AI gathers and organises evidence so recruiters can spend their time judging it, and a named person stays accountable for every hiring decision.
Where AI is used across the hiring funnel
| Stage | What AI typically does | What people should own |
|---|---|---|
| Job description | Drafts a description from a few notes; suggests clearer, more inclusive wording | What the role actually needs, and the final text |
| Sourcing | Searches profiles and drafts outreach messages | Who is contacted and how |
| Application review | Summarises CVs and matches them against stated requirements | The shortlisting criteria, and spot checks of what was filtered out |
| Screening | Holds structured screening interviews and scores answers against criteria | The criteria and weights, reviewing evidence, and deciding who advances |
| Scheduling | Offers slots and books interviews | Making sure interviewers actually turn up prepared |
| Final interviews and offer | Summarises notes and prepares interview packs | The assessment, the decision and the offer |
What AI does well
- Consistency at volume. An AI interviewer asks the 200th candidate the same core questions, with the same patience, as the first.
- Availability. Candidates can be screened in the evening or at the weekend, in their own time zone.
- Turning conversations into records. Transcripts, summaries and evidence linked to criteria are tedious for people to produce and quick for software.
- First drafts. Job descriptions, outreach messages and interview questions all go faster when a person edits a draft rather than starting from a blank page.
Where AI falls short
It can learn the bias in your history
A model trained to predict “who we hired before” will reproduce whatever patterns shaped those past decisions, including unfair ones. The best-known example is the experimental CV-screening tool that Amazon abandoned, after Reuters reported in 2018 that it had learned to downgrade applications mentioning the word “women’s”. Tools that score against criteria you define explicitly are easier to inspect than tools trained to imitate past outcomes. But they still need checking.
Scores can be hard to explain
If a tool gives a candidate a 62 and cannot show which answers drove that number, nobody can verify it, challenge it, or explain it to the candidate. Prefer tools that show their evidence alongside every score.
Language models can be confidently wrong
Summaries can attribute a statement to a candidate that they never made, or miss a qualification they mentioned in passing. That is why reviewers should read the source transcript, not just the summary, before acting on anything important.
Not every candidate interacts the same way
Speech recognition handles some accents better than others. Candidates with disabilities may need a different format or extra time. Unreliable internet is not a lack of skill. Build in a route to a human alternative, and make it easy to ask for.
Keeping people in the loop, in practice
“Human in the loop” only means something if the human can realistically disagree with the machine. In practice that means:
- A named reviewer looks at the evidence for every rejection, not only for the candidates who advance.
- Reviewers have the time and the material to check a score: the transcript, the recording, the criteria.
- Overturning an AI recommendation is normal and expected, not an exception that needs justifying.
- You track how often reviewers disagree with the AI. Zero disagreement is a warning sign that nobody is really reviewing.
Regulation and transparency
Rules for AI in hiring are developing quickly and vary by jurisdiction, so treat this as a map rather than legal advice:
- New York City’s Local Law 144 requires employers using automated employment decision tools for candidates in the city to commission an independent bias audit, publish a summary of it, and notify candidates.
- The EU AI Act classifies AI systems used for recruitment and selection as high-risk, which brings obligations around risk management, human oversight, transparency and record-keeping.
- Illinois’ Artificial Intelligence Video Interview Act requires employers to notify candidates, explain how the AI works and obtain consent before using AI to analyse video interviews.
- Data protection law, such as the GDPR in Europe or India’s Digital Personal Data Protection Act, governs how candidate data is collected, used and retained, whatever tool processes it.
Whatever the local rules, the same baseline practices make sense everywhere: tell candidates when AI is involved and what it does, explain what data is recorded, offer an alternative where needed, and keep a person accountable for the decision. Check the specific obligations that apply to you with your legal team.
Questions to ask any AI recruiting vendor
- What exactly does the tool score, and can I see the criteria and their weights?
- Can every score be traced back to evidence from the candidate’s own answers?
- Who decides which candidates advance: the tool, or a person on my team?
- What is recorded, where is it stored, and for how long?
- How is the candidate told that AI is involved?
- What happens when a candidate has a technical problem or needs an accommodation?
- What can you document about how the tool performs across different candidate groups?
A sensible way to start
Pick one high-volume role where screening is already a bottleneck. Write its criteria down, run AI screening alongside your current process for a few weeks, and compare the two: which candidates each would have advanced, where they disagree, and why. That comparison tells you more than any demo. It also gives your team a feel for reviewing AI evidence before anything depends on it.
If you are evaluating tools for that pilot, our guide to AI interviewing covers interview design and review in detail, and the Vriti AI Interview overview shows how one screening interview runs from start to finish.