When AI Handles Recommendations, Job-Search Preparation Becomes the Service

In the Daejeon pilot, in-person counselling time increased for participants needing intensive support

Since May 2026, the Ministry of Employment and Labor (고용노동부), Korea’s national labour ministry, has been piloting “My AI Employment Center” at the Daejeon Employment Center. It serves 152 participants in the National Employment Support System (국민취업지원제도), a public programme for people seeking work. It is not yet a finished service available nationwide.

An analysis of pilot participants’ surveys and work histories divided 110 people into a self-directed group that conducts its job search mainly online, and 42 into an intensive-support group that receives help from a dedicated counsellor. Rather than offering every job seeker the same number and type of consultations, the model concentrates human time where it is needed more.

The volume of administrative work is behind the government’s push for this model. Employment centers nationwide serve more than 2 million users a year, and each employee handling unemployment benefits processes an average of 71.7 unemployment-recognition cases a day.

In the Daejeon pilot, average face-to-face counselling time for intensive-support participants rose from 18 minutes to 73.8 minutes. The pilot had 152 participants, and 86.8% gave a positive assessment in a separate satisfaction survey. These are early results, however: they do not mean that employment rates or post-employment retention improved.

The government plans to apply AI and links to administrative information to repetitive procedures such as verifying job-search activity, filling in applications, reviewing them, and making payments. It has also set out a schedule to run integrated pilot centers in one or two regions in 2027, then expand them nationwide in stages from 2028.

When recommendations become common, preparation matters more

Consider Min-su Kim, a 46-year-old living in Daejeon. He left a quality-inspection job at an auto-parts company and is participating in the National Employment Support System while exploring similar manufacturing roles and facility-management work.

Kim finds postings on his phone, then revises his résumé on an old laptop. Copies of his certificates are in his photo gallery, notes from counselling sessions are on paper, and application deadlines and counselling dates are scattered among text messages.

He re-enters the same work history and certificates in his résumé, job-search plan, and training application. Sometimes he applies without recording the result, so the counsellor has to review past activity again. At other times, he discovers a missing document the day before an interview.

This is worth viewing in reverse. If public AI can recommend jobs and process administration well enough, and provide those functions for free, someone building another job-listing recommendation service may be at a disadvantage.

The work of turning a recommended listing into an actual application still remains. There is a need for a service that gathers the next actions on one screen: “revise the career description by Friday,” “submit the certificate before Monday’s counselling session,” and “record the result after the interview.”

Before a session, the job seeker could summarise previous application results, reasons for giving up, blocked documents, and questions to ask on a single page. If only information Kim has agreed to share is sent to the counsellor, the counsellor can avoid asking again about work already done and start with more complex conversations, such as changing occupations or care responsibilities.

Some work does not change. People still need to judge how health conditions and family circumstances affect real employment, review people AI has classified incorrectly, and hear objections about pay or eligibility for support.

The place for a small operator is therefore not a giant service that replaces the employment center. It is a connection tool that helps job seekers keep the records they prepared, identify the moment to ask a person for help, and continue to the next action after counselling ends.

Abroad, recommendations alone were not enough

France Travail, France’s public employment agency, used La Bonne Boîte to tell job seekers about companies likely to hire in the future. It was a free public service that let people apply proactively to companies without publicly posted vacancies.

In a large-scale experiment summarised by the Organisation for Economic Co-operation and Development (OECD), women job seekers’ likelihood of finding employment rose by about 2%. A follow-up experiment found that short-term contract employment increased by about 1%. Applications sent to recommended companies were about 2.7 times as effective as applications sent to other companies, but the scale of the effect also shows that a recommendation alone does not solve the whole employment journey.

Bob Emploi, also in France, was a free service that used local hiring information to suggest job-search strategies and action plans. An evaluation covering 2017 to 2019 found limited changes in job-search strategy and no confirmed employment outcome, and the partnership ended in 2024.

Users did, however, visit the France Travail site 4.7% more often and were 2.4 percentage points more likely to meet a counsellor within six months. The result suggests that automated advice can become an entry point to human support rather than replace a counsellor.

Job Market Finland identifies skills from a job seeker’s self-description and work history, then connects them with jobs. It is free for job seekers and employers, and lets users directly verify the skills and occupations proposed by the system.

When the OECD described the service, it contained about 224,000 job seeker profiles and roughly 20,000 job listings. Its evaluation suggested that job matching may have improved slightly, but the conclusion was not clear, and the structure leaving the final choice to people remained in place.

Four things you could build now

1. An application-preparation board would expand the tasks due before an application deadline in date order when a user saves a job listing. It is for job seekers in their 40s and 50s moving from manufacturing into office or facility-management roles.

As public services increase recommendations, people need a single place to manage documents and schedules across multiple listings. The first screen would show “3 applications this week” and the remaining documents for each listing.

2. A counselling handover card would let job seekers organise recent activity and blocked issues, then send them to a counsellor. It is for private employment-support providers and contracted organisations where several counsellors meet participants in rotation.

Even when counselling time grows, time for useful advice shrinks if the counsellor must review the previous session again. The first screen would contain three fields: “what I did after the last session,” “what is blocked now,” and “what I need to decide today.”

3. A missing-document checker would compare an institution’s document checklist with files uploaded by a job seeker and flag missing items. It is for staff at employment-support organisations who repeatedly confirm National Employment Support System participants’ documents by phone and text message.

As administration becomes faster, small errors—incorrect file names, blurry photos, and expired documents—become the next source of delay. The first screen would show only “complete,” “resubmit,” and “not yet received” for each participant.

4. A first-30-days-at-work guide would ask brief questions about adjustment tasks after starting work and connect a person to a counselling organisation when help is needed. It is for people beginning work at a logistics center or manufacturer after a long career break.

Publicly available pilot results currently focus on counselling time and satisfaction, leaving post-employment retention as a separate gap. The first screen would have buttons for “one thing that was difficult today,” “one thing to ask about tomorrow,” and “request counselling.”

What to check today

Call one counsellor at an employment-support organisation for 20 minutes and ask where they record one recent client’s documents and application schedule. If, within the past week, they had a case where the same information was copied into two or more places and they sent three or more calls or text messages to check for missing items, a small preparation or handover tool is worth testing.

Why this matters where you are

Check whether public or private employment services in your market are making recommendations and administrative steps easier. What differs may be the institutions, eligibility rules, and counsellor workflow, but the handoff between a job seeker’s scattered preparation and human support is a concrete process you can examine. Start by observing where application records, documents, and next actions are lost or repeated.