When AI writes the application, hiring should move to working together

A better entry sheet does not mean you know the person

In Japan, new graduates apply for jobs with an entry sheet (ES): a written application with a motivation statement and a self-introduction. In a survey Mynavi ran in April 2026, 84.9% of 1,258 university and graduate students graduating in 2027 said they had used AI in their job search. It is time to design hiring on the assumption that applications are written with AI, rather than treating them as the exception.¹

Since LLMs arrived, structuring and phrasing a text has become easy. Motivation statements and self-introductions can all be brought into a decent shape. What gets lost is not the value of the experience itself. It is the value of a polished text as a clue to someone’s ability and drive.

A study of Freelancer.com, an online marketplace for freelance work, found that after an AI tool for drafting applications was introduced, the link between how well a text matched the job and the chance of hearing back from the client weakened. The link with ratings that reflect past work grew stronger. It did not study Japanese graduate hiring, but it suggests that judgement moves away from writing and towards track record.²

This change also gives people who are not strong writers a better chance to convey their experience. Using AI to organise what you have done is different from inventing achievements you do not have.

So rather than trying to detect whether AI wrote something, companies should put their effort into a different question: how will we find out whether we can trust this person with this job? If the entry sheet stays, it should be less a contest of long self-promotion and more a short note on experience, what the person was responsible for, the work they are interested in, and what they do not know yet.

Meeting in person matters beyond the prepared answer

In December 2025, Rohto Pharmaceutical announced it would drop entry-sheet screening for graduates joining in April 2027. Instead, the first step is “Entry Meet”, a conversation with someone from HR at one of eight locations across Japan. It is an example of talking directly to build mutual understanding before judging anyone on paper.³

Going back to in-person meetings is not enough on its own, though. If candidates recite a memorised motivation statement and are judged on how they speak and come across, the written text has simply been swapped for a spoken one.

In interviews and group discussions, what should be tested is how someone thinks and acts in the moment. For example, hand them a small task close to the real work. Change the conditions halfway, or have an employee give advice, and ask them to try again. Watch what they notice is missing, how they ask questions, how they handle disagreement, and what they change after feedback.

Decide the behaviours to look for in each role beforehand. The structured interviews described by the US Office of Personnel Management work the same way: set questions and shared rating criteria. Do not stop at “they seemed nice”; be able to explain what the judgement was based on.⁴

For jobs that use AI in practice, it is fine to set tasks that use AI too, with the same tools available to everyone. Keep those separate from the tasks that test the basic skills the person needs. Meeting in person is not about shutting AI out. It is about seeing the person’s own judgement and how they work with others.

A long-term internship narrows the gap between expectations on both sides

One interview or task cannot show how someone works over time. So I propose making long-term internships with a set role and team one of the main routes into graduate hiring. What I have in mind is paid work on real tasks, with ongoing guidance.

Across several pieces of work, the company can see whether someone meets deadlines, keeps up quality, asks for help when they are stuck, and improves after a mistake.

At the same time, the student is checking the company. What routine work fills the days of the role they are interested in? How do managers and colleagues treat them when things get busy? What standard is expected, and how much will they be trusted with? It becomes time to decide not only whether they like the work but whether they want to keep doing it there.

Find out before joining as a full employee what people usually discover only after they join. It shrinks the gap between a student’s expectations and reality, and lowers the company’s risk of noticing a serious mismatch only after hiring. Both sides gain.

An employer survey published by NACE in the US in 2026 also found that new hires who had interned with their employer stayed longer than those who had not. But aptitude and selection differ to begin with. It does not prove that longer internships cause fewer people to leave.⁵

Nor does success have to mean that everyone joins. Working there and realising “this role is not for me” is a valuable outcome too. What we want to avoid is not leaving as such, but carrying a mismatch that could have been found earlier into the job itself.

Do not lower the bar for students: show them the real job

If entry to the long-term internship again rests on how well the entry sheet is written, the problem has only moved earlier. I think selection should use interviews, tasks and group discussions, and apply the same job requirements and criteria as full-time hiring for that role.

Being aimed at students is no reason to settle for “energetic” or “seems motivated”. Show concretely what knowledge and skills the job needs, and what level of output is expected.

A full-time standard does not mean expecting, on day one, the polish of an experienced mid-career hire. Even junior employees have skills required when hired and skills they build with guidance. Make that distinction explicit for internships too.

In a planning role, for example, do not stop at “came up with ideas”. Show whether they can explain the customer’s problem, checked the evidence, and thought about cost and how to carry it out. When students can see the gap between what they produce now and what is usable as work, they have something to grow towards.

The company must provide a mentor, a defined scope, and chances to look back on the work. “Catch up to this level on your own” is not work-based hiring; it just shifts the cost of training onto the student.

Is it “I want to continue, but I can’t do it yet”, or “having tried it, I no longer want to”? Being able to talk about these two separately is what matters. If it is the job they want, support their growth; if they find it is not, discuss whether to continue or stop. Holding a high bar for the work and not treating a change of course as failure can go together.

Connect the work done as an intern to the first placement

For this to work, the unit of hiring has to change too.

Say a student works six months in a planning team and is happy with both the work and their mentor. The company hires them, then places them in an entirely different sales team. The fit that took six months to confirm does not carry over.

The default should be to hire into the role and team where the internship took place. Instead of hiring into the company as a whole and deciding placement later, agree on “what job we are bringing you in for” before the person joins as a full employee.

No company can promise that its structure or managers will never change. All the more reason to be clear about the first placement, the work, and the range of possible future transfers. If plans change, explain it so the student can decide again.

Adding long-term internships while leaving placement as vague as ever undercuts the very understanding they were meant to build.

Job hunting may start earlier and last longer. That is not the goal

If working together matters more, companies and students will need to meet earlier. In Japan too, more students may start real work around their second year at university and get to know jobs and companies over months.

Mynavi’s October 2026 review of the hiring season already reports that students do career activities from their early years and then apply to a few companies they know well. This shift earlier cannot be explained by AI alone, and the idea that AI will make long-term internships the norm is a hypothesis for now.⁶

Earlier and longer are not the same thing, either. If students start early and find work they are sure about, the time spent going through selection may get shorter. If several companies each demand long stretches of work, both the company’s mentoring load and the student’s time burden grow.

So long-term internships must not be stacked on top of every existing selection step. What has been confirmed through the work does not need to be asked again in more entry sheets and more interviews. And length should not follow “the longer the better”: set it by the work involved and the experience both sides need to decide.

Hours that fit around classes and research, fair pay and regular check-ins on whether to continue are essential. Keep short, intensive work placements or another selection route for students who cannot commit for long. I do not want a system where the long-term internship is the only way in, and only people with time to spare get hired.

The aim is not to turn university into one long selection process. It is to go back and forth between study and work, and reduce the mismatch after joining.

Careers sites should show what the work is, not just the company image

If hiring is organised by job, the careers site, as the way in, has to change as well.

LinkedIn offers AI job search that takes the job you want in plain language and matches its intent against listings. The route by which candidates look for work through AI already exists.⁷

Say someone asks: “I’m interested in product planning, can work two days a week while studying, and want a company where I know the role before I join.” “Young people thrive here” and “a place to take on challenges” do not tell them whether the company fits.

For each role, publish the work they will be given, the skills needed, the expected standard of output, working conditions, pay, mentoring, and the conditions for being hired as a full employee. Say what is decided about placement and what is not. Whether candidates read it themselves or ask AI to help compare, they should arrive at the same facts.

To that end, do not leave important information only in images; make it readable text. Give each role its own page, and update it when conditions change or hiring closes. Google says its AI search features do not need special AI files, and advises providing important content as text and keeping structured data consistent with what is shown. Preparing a special file will not make AI recommend you.⁸

What loses value is choosing people by how good their writing looks. Text that describes the work accurately does not lose value at all.

What companies should change is the hiring promise, not the hiring tools

There is no need to change every role at once. Start with one team: publish the work and criteria expected of full employees, and set up a paid long-term internship entered through conversations and work tasks. Tie the work done to the first placement, and do not repeat in later selection what has already been confirmed.

Companies can use their own AI to make time for conversation, for scheduling and keeping records. If AI is also used in assessment, decide what it draws on and who makes the final call, and tell candidates how it is used.

When measuring results, look beyond applications and offer acceptance: the burden on students, the time employees spend mentoring, the gap between expected and actual work after joining, retention and performance. People leave early for other reasons too, such as pay, health or family, so do not judge the scheme on turnover alone.

Students get a chance to learn what this job asks of them. If they are interested, they can grow towards it. If they find it is not right, they can choose another path before joining as a full employee. And companies can welcome people based on the judgement and growth they saw in the work, not only on a polished self-introduction.

Instead of asking students who have never worked there for a finished answer to “why do you want to join us?”, find out together: “Having tried this work, do you want to keep doing it here?”

What hiring in the age of AI needs more of is not rounds of selection. It is chances for both sides to understand what they are choosing.

This piece is a view and proposal based on public information as of 14 October 2026. The statistics and company announcements describe the current situation and do not guarantee the effect of the hiring approach proposed here. “Long-term internship” is used descriptively for paid, work-based placements and does not refer to any category in the Japanese government’s career development framework. Translated from the Japanese original.

• #Hiring
• #Generative AI
• #Internships