Tell students two things at once, because the evidence supports both. Employers do not, by their own account, plan to replace entry-level jobs wholesale, and the first rung of some careers is already harder to reach. Then show students how to test any headline for themselves. A student who hears only the first message is unprepared. A student who hears only the second has stopped planning.
Why does the careers conversation about AI keep going wrong?
It offers students two stories and the evidence supports neither. In the first, AI takes the job. In the second, nothing changes and a good degree still opens every door. Either one is easy to say to a sixteen-year-old, and either one fails the moment the student meets a fact that does not fit.
Students have already noticed that something is moving. Prospects Luminate, the careers arm of Jisc, surveyed 710 users of its careers website in November and early December 2025. About a third were still in school, sixth form or college. Thirteen percent said they had changed their career plans because of AI, and about a third more were considering it. Among those who had changed course, 69 percent named fear that their chosen job would be replaced, according to the Institute of Student Employers (ISE), which co-produced the report. Fear is doing a good deal of the steering.
Hold that figure at its true size. These are self-selected visitors to a careers website, not a national sample of students. But it tells a careers leader (a school counselor, in the US) where the conversation starts: with a student who has already half decided.
What does the evidence say about AI and entry-level jobs?
Two bodies of evidence look as if they disagree. Read closely, they describe the same narrow squeeze from two sides.
The first is American payroll data. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at the Stanford Digital Economy Lab analyzed ADP records covering millions of US workers through June 2026, in a working paper revised on August 12, 2026. Employment of workers aged 22 to 25 in the most AI-exposed occupations stands 19 percent below where it would be had it kept pace with less-exposed occupations. Experienced workers in the same occupations show no comparable gap, and the authors find no evidence of job loss across the whole economy. They present the result as an early descriptive signal, not a causal estimate. The gap shrinks when education is taken into account, and some of the trends began before generative AI.
The second is what employers say they plan. The Prospects Luminate and ISE report, published in April 2026, collected views from 30 ISE member employers between November 2025 and January 2026. Over the next three years, 53 percent expected entry-level hiring to hold steady, 27 percent expected it to grow and 17 percent expected it to fall. None expected any early-career task to be fully automated within three years. Routine language and content tasks were judged the most exposed; work involving judgment, stakeholder engagement and complex decisions was expected to stay mainly human-led. Thirty employers is a small sample, and a stated plan is not a hiring decision.
Put the two side by side and the contradiction goes away. Stanford measured who actually got hired, in one country, in occupations ranked by exposure. The employers described intentions across all entry-level roles. If the squeeze is real, it falls on a particular rung of particular occupations, not on every student and not on every career. Georgios Petropoulos, writing in MIT Technology Review on May 26, 2026, draws the same line: the weakness is concentrated in AI-exposed early-career jobs such as software development and customer service, not in jobs with low exposure. The routine coding work that "learn to code" programs once trained students for, he argues, is the work AI handles well.
What is the Rung Check?
The Rung Check is my suggested way for a careers leader to test any claim about AI and a job before it reaches a student. It is a way of thinking, not a validated instrument.
The Rung Check is three questions a careers leader asks of any claim that AI will take a job: Which tasks in the job does it touch? Does the claim concern the first rung of the career or the whole ladder? And what did the evidence measure, in whom, and when? A claim that cannot answer all three is a headline, not a forecast.
Which tasks does it touch? A job is a bundle of tasks, and AI reaches some of them. In the Prospects Luminate survey, students aiming at translation or interpreting moved furthest from their intended careers, while students heading for technology, engineering, business and finance mostly stayed in their sectors and adjusted toward AI-related roles. The students who left were moving away from the field where the report finds routine tasks most exposed. The students who stayed were facing a job that is changing, which is a different thing.
First rung or whole ladder? The Stanford gap is for workers aged 22 to 25. A fifteen-year-old thinking about a forty-year career needs to know that the way in may narrow, which is a reason to plan the route, and not that the destination has gone, which would be a reason to abandon it.
What did the evidence measure, in whom, and when? A figure from US payroll records, a survey of 30 employers convened by the UK's Institute of Student Employers and a self-selected student poll are three different instruments. A claim quoted without its country, its sample and its year has lost the information a student would need to weigh it.
What does the Rung Check look like on a real student?
Take a hypothetical Year 12 student who wants to study languages and become a translator. She has read that AI has finished the profession, and she is about to drop the subject.
On the first question, the evidence says routine language and content tasks are the most exposed, so her worry has a basis. It does not say every translation task is exposed, and the employers in the ISE sample expected the parts involving judgment and dealing with clients to stay mainly human-led. On the second, the data concerns entry-level hiring, so the honest advice is to find out what first-rung roles look like now. On the third, she should know that the survey behind her fear is a small employer sample from one organization. She leaves with a plan and not a verdict: stay with languages, ask which kinds of language work involve judgment, and add a second string in a field where she can check a machine's output because she understands the subject.
That last move is the one the data keeps pointing to. Employers in the ISE research asked educators to prioritize AI literacy, critical thinking, verifying AI output, ethical use, human skills and adaptability. The same list is what I mean by the skills that get scarcer, not cheaper.
What I Tell Leadership Teams
Put careers in the room where the AI decisions are made. The pattern I see in the schools I work with is that the AI lead is thinking about homework and assessment, the careers leader is thinking about the Year 11 options evening, and nobody joins the two. Students then hear AI-and-jobs claims from whoever happens to say them: a form tutor, a parent, a video.
I was a secondary teacher and an assistant headteacher before I started speaking about AI, and I would give leadership teams one instruction. Agree the message, then give it to every adult who talks to students about their future. It fits in three sentences:
- The first rung of some careers is getting harder to reach, and the evidence for that is real but early.
- Employers say they are changing jobs more than replacing them, and they are asking for people who can check what a machine produces.
- Nobody can promise you which jobs will exist, so the plan is to build what holds under any outcome: depth in a field, the habit of checking AI output, and evidence of work you have done.
The table below gives a careers leader a script for the five things students say most.
| A student says | Avoid | Say instead |
|---|---|---|
| "AI is going to take my job." | "Don't worry, it won't." | "Which part of the job? Let's find out which tasks are exposed and which are not." |
| "Is it worth learning to code?" | "Yes, there's always demand." | "Routine coding is what AI does well. Coding plus a field you understand is a stronger combination." |
| "Everyone says AI won't affect my career." | "They're right." | "Entry-level hiring in exposed jobs has narrowed in US payroll data. Build proof of what you can do: projects, work experience, an apprenticeship." |
| "I'll just use AI for everything." | "That's cheating." | "Employers want people who can check AI output. You can only check what you understand." |
| "My parents say a degree still protects you." | "That's out of date." | "A degree still opens doors, but it is not a guarantee. Ask what the first job after it looks like, and what that employer will want to see." |
Isn't it irresponsible to tell a teenager the evidence is mixed?
No, because uncertain is not the same as vague. The cheerful version fails against the Stanford data. The gloomy version fails against the employers. A student who is given either will eventually meet the other, and will trust the adult a little less.
What a student can use is a plan that holds under any outcome. Depth in a field lets them judge what a machine produces. The habit of checking output is what employers say they want. Evidence of work done, whether a project, a placement or a portfolio, answers the question every first-rung employer asks. If you want the school-wide version of that argument, what schools should be preparing children for now takes it further, and what happens to entry-level work sets out the same squeeze from the employer's side.
What should a careers leader do this term?
Do five things, and none of them needs a new program.
- Collect the claims. Ask two or three tutor groups to write down what they have heard about AI and jobs. Keep the three most common.
- Run the Rung Check on them. Give it twenty minutes of a staff meeting. For each claim, name the tasks, the rung and the evidence.
- Write the three-sentence message and give it to form tutors, subject teachers and whoever runs parents' evening. How to talk to parents about AI covers the letter home.
- Use work experience as evidence. Add one question to the debrief: which tasks did you see a machine doing, and which did a person decide?
- Put a date on it. The sources above will age. Re-check them in the spring and change the message when the evidence changes.
Two neighboring decisions belong on the same table. Students who rely on a chatbot for every task are building the wrong second string, which is the subject of whether students are becoming dependent on AI, and AI literacy in your school defines the skill employers are asking for. For students heading into vocational routes, AI in further education covers the college side, and staff who advise students need the same judgment, as AI skills for staff explains.
Students will hear the headline whatever you do. Your job is to be the adult who read the paper.
If your leadership team is working through these questions, this is the kind of work I support through AI training for schools, and for a conference or a staff day I speak on it through my education keynotes.
Sources and further reading
- Brynjolfsson, E., Chandar, B., and Chen, R. "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." Stanford Digital Economy Lab, working paper, revised August 12, 2026. Read it.
- Prospects Luminate (Jisc) with the Institute of Student Employers. "AI and Early Careers: student, graduate and employer perspectives." April 2026. Report page.
- Institute of Student Employers. "AI early career fears rise, but employers focus on evolution." April 16, 2026. Read it.
- Petropoulos, G. "It's time to address the looming crisis in entry-level work." MIT Technology Review, May 26, 2026. Read it.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and an international keynote speaker who has trained more than 150,000 educators. More about Dan.
