AI resume screening that reads every CV against the job — when you ask it to

AI screening is a switch on each job, under the job’s AI Settings tab, and every new job starts with it off. The job page shows whether it is enabled. That keeps the choice deliberate and narrow: you can screen the junior role that drew two hundred applicants and leave the senior hire, where every CV deserves a careful read, to people alone.

The switch is the only AI setting on a job. There is no score threshold, no automatic rejection and no rule that moves a candidate to a stage because of what the model said.

When the screening has run, the candidate’s Screening tab shows a full read-out, and the score follows the candidate around the pipeline as “AI: N%” on the list, the cards and the kanban board.

• Match score as a percentage.
• Recommendation on a five-step scale: Strong yes, Yes, Maybe, No, Strong no.
• Skills match: the job’s skills the CV shows, the ones it does not, and the share matched.
• Experience fit, in a sentence.
• Key strengths, areas of concern and any red flags.
• A summary of the candidate against the role.
• The requirements used: the experience level, required skills and other requirements as they stood when the CV was screened.

The last item matters more than it looks. A job description is often edited after it is posted; keeping the snapshot means a score from last week is read against the requirements it was actually scored on, not today’s. The candidate’s activity timeline records when screening started, completed or failed, and until a result arrives the tab says the analysis is pending rather than showing an empty score.

A match score is a reading aid. It is good at the tedious part — noticing that a CV never mentions the framework the job requires, or that the experience is twice what the role asks for — and it gets through a pile faster than any person. It is poor at the things that make a hire: a career change that explains a gap, a portfolio that says more than the CV, a skill described in words the job description did not use, a non-traditional route into the field.

So HRMS keeps the result beside people’s judgement rather than in place of it: the score sits on the profile next to the Interviews and Scorecard tabs, and stage moves, rejections and offers are actions your team takes. Our advice is to use screening to decide the order you read in, not who gets read; to treat the missing skills as questions for the first call rather than reasons to reject; and to open the lowest-scored CVs now and then, because that is where a model’s blind spots show.

Screening is done through OpenAI’s API. The text of the CV and the job description it is matched against are sent to OpenAI when your organisation uses screening or résumé parsing, and not otherwise. OpenAI does not train on data sent through its API, and the processing happens in the United States. OpenAI is listed on our sub-processors page with exactly that scope, and the privacy notice says the same.

If that is not acceptable for a role, or for your organisation, leave the switch off. The rest of HRMS — the careers page, the pipeline, assessments, interviews and scorecards — works the same without it, and a candidate you hire still becomes an employee record that onboarding picks up from.