AI Did Not Kill Entry-Level Jobs. Hiring Inflation Did.

The reason you can't find an entry-level job that's actually entry-level isn't a robot. It's a recruiter who raised the bar because they could. A 2018 TalentWorks analysis of around 100,000 listings found that 61% of postings labeled "entry-level" demand more than three years of experience, and that pattern was already broken years before ChatGPT existed. AI is a real long-run risk. The squeeze you feel today is hiring inflation.

That distinction is not academic. It decides what you do next. If a machine ate the job, your only move is to retrain into something the machine can't do. If a recruiter inflated the requirement, the job still exists and the math is winnable. You're not fighting automation. You're fighting a filter. And filters have rules you can play.

Why do entry-level jobs require experience they shouldn't?

Because experience is a free, fast screening signal, and when applications flood in, recruiters use it to thin the pile. The job didn't get harder. The line of applicants got longer.

Look at one number. For production supervisor roles, 67% of job postings require a degree, but only 16% of people actually working those jobs hold one. A 51-point gap. The work that a supervisor does in 2024 is the same work a non-graduate did competently in 2012. What changed was the checkbox, not the role.

Economists who studied this during the last downturn found a clean mechanism. As unemployment rose between 2007 and 2010, degree requirements inside the same occupations rose 2.2 percentage points and two-plus-year experience requirements rose 3.5 points. The jobs were identical. The bar moved because employers could afford to be picky. When hiring recovered, the inflated bar mostly stayed. That's the ratchet. It clicks up in a bad market and rarely clicks back down.

What does requirement inflation actually look like on a job post?

It looks like a title and a body that contradict each other. Once you see it, you can't unsee it.

Weak signal (what the title says): "Entry-Level Marketing Coordinator. 0-2 years experience preferred."

Strong signal (what the body actually demands): Salesforce, HubSpot, Google Analytics 4, paid media management across Meta and Google, plus owned copywriting for email and landing pages.

That's not an entry-level job. That's three years inside a marketing agency, priced as a junior salary. The "0-2 years" line is aspirational marketing. This is the same gap that makes the posted job description mostly fiction: the title and the requirements describe a wish, not the person who actually gets hired. The skills list is the real spec, and it was written by a manager who forgot what a 22-year-old actually knows on day one. No amount of AI-literacy training closes that gap, because the gap is not about AI. It's about a recruiter over-indexing on a signal that's cheap to demand.

Here's the analogy that makes it concrete. A 2013 field experiment by economists Kroft, Lange and Notowidigdo sent roughly 12,000 fictitious resumes to about 3,000 real job postings. Callbacks dropped sharply the longer a candidate had been unemployed, with most of the decline hitting in the first eight months out of work, and the penalty got worse in tighter labor markets. Employers weren't measuring skill. They were using a free shortcut to thin the stack, and they leaned on it harder when the market gave them room to. Years-of-experience is the same kind of shortcut: cheap to filter on, and a weak predictor of who can actually do the work.

If it's not AI, why did the timing line up with ChatGPT?

This is the trap. The numbers got worse right around when generative AI went mainstream, so the story writes itself: AI killed the jobs. But the timeline doesn't cooperate.

The deterioration in AI-exposed roles started in early 2022, months before ChatGPT launched in November 2022. You can't blame a product for damage that began before it shipped. What was happening in early 2022 was the unwind of the pandemic over-hiring bubble plus a tech-specific freeze. A buyer's labor market. The exact macro condition that fired the ratchet in 2008.

The cleaner tell is what happened outside tech. Over the same window that tech tightened, non-tech roles saw experience requirements ease from 16% to 11%. If AI were broadly eating entry-level work, you'd expect inflation everywhere AI can touch a keyboard. Instead it concentrated exactly where the hiring freeze hit hardest. Inside tech, the share of postings requiring five-plus years rose from 37% to 42% between mid-2022 and mid-2025, while the two-to-four-year band shrank from 46% to 40%. And junior roles took the hit: standard and junior tech roles fell 34% from early 2020 to early 2025, while senior and manager roles fell only 19%. That's a market protecting senior headcount and squeezing the bottom of the ladder, the same split that turned engineering hiring into a barbell rather than a ladder. It's a hiring decision, not a hardware capability.

There's one more piece of theater. Greenhouse found that 18-22% of postings on its platform in any given quarter are ghost jobs, roles kept live for pipeline optics or ATS hygiene with no real intent to hire. Three in five job seekers suspect they're hitting ghost jobs. Some of the "impossibly inflated" listings you're staring at were never open, which is its own problem of ghost jobs and dead listings worth learning to spot. Blaming AI for postings that were always set dressing misreads the whole board.

Is there any real AI signal, or is this all macro?

There is a real signal, and pretending otherwise would be its own kind of dishonesty. The honest read is: AI is a contributor on top of the macro cause, not the trigger.

The Stanford team that dug into payroll data found something specific and uncomfortable. Workers aged 22-25 in the most AI-exposed jobs saw a 13% relative employment decline since late 2022, 22-25-year-old software developers fell nearly 20% from their late-2022 peak to July 2025, and older workers in the same occupations stayed stable or grew. That age skew is exactly what you'd expect if AI is absorbing the simple, codifiable tasks that juniors used to cut their teeth on. Separately, Revelio Labs found highly AI-exposed entry-level roles declined over 40% since January 2023 versus 33% for low-exposure roles, with a 10-point rise in AI exposure tied to an 11% drop in entry-level demand. That differential is genuine. Exposure matters.

But note the size and the shape. Anthropic's own labor analysis found a roughly 14% relative decline in job-finding rates for 22-25-year-olds in high-exposure roles, with no system-wide unemployment surge for exposed workers through late 2024. And the forward picture isn't uniformly grim: Vanguard found job growth in AI-exposed occupations accelerated from 1% to 1.7% while other jobs decelerated, and wage growth in those fields jumped from 0.1% to 3.8% post-pandemic. AI is reshaping which tasks get done, and the bottom rung of a few specific ladders is genuinely thinner. That's worth planning for. It is not the thing standing between you and the marketing coordinator job that wants three years of HubSpot.

What's the difference, and why does it change your strategy?

Here's the whole point on one screen. Misdiagnose the cause, pick the wrong fix.

If the cause is...	The job is...	The wrong move	The move that works
AI displacement	Gone or shrinking	More applications to the same shrinking pool	Retrain into adjacent, less-automatable work
Hiring inflation	Still there, just gated	Generic AI courses to "stay relevant"	Beat the filter: signal, target, referral

For the role you're chasing, the cause is almost always the second row. The work exists. A human put a filter in front of it. So you attack the filter.

One: compress the experience signal before anyone reads your years. A recruiter scanning 400 resumes is looking for a reason to keep yours. Three shipped portfolio projects that mirror the job's actual skills list do more than "0-2 years" ever could. You're not faking experience. You're producing evidence of competence the way the strong marketing candidate above would.

Two: target where the ratchet hasn't calcified. Late-stage companies with thousands of applicants per role can afford the inflated bar. Growth-stage companies that just raised and need to ship cannot. They still hire juniors because they have to. Point your applications there.

Three: go referral-first. The whole reason the years-of-experience filter exists is to thin an anonymous pile. A referral skips the pile. One warm intro is worth fifty cold applications into an ATS built by a senior manager who forgot what entry-level means.

What's the trade-off here?

Playing the filter takes longer up front and it isn't fair. You will spend weekends building portfolio projects that a candidate from a target school or a well-connected family didn't need. You'll send fewer applications but spend more on each. That's slower than spraying your resume at 200 postings, and it feels worse in week two because the rejection count looks low only because you applied to less.

Name it plainly: you're trading volume for hit rate, and effort up front for a real shot later. The fatalist reads the 61% stat and concludes the system is rigged, so why bother. They're half right. It is rigged. But there's almost nothing consistent effort can't eventually move, and a filter built by a lazy recruiter is one of the most beatable things in your career. The agency is in choosing where to aim it.

What to do now

1. Pull up the last entry-level posting that rejected you. Read the body, not the title. Highlight every concrete skill. That list is your real syllabus.
2. Build two portfolio projects against that exact skill list this month. Ship them somewhere a recruiter can click.
3. Make a list of ten growth-stage companies in your space that raised in the last year. They hire juniors because they have no choice.
4. Before you apply cold to any of them, spend one hour finding a warm path in. A referral beats the filter every time.

You don't need to out-AI the machine. You need to out-signal the filter, aim at the right doors, and get one human to vouch for you.

Want help reading what a job post actually wants and building the resume that clears the filter? Message Praxy on WhatsApp. Send me a posting that rejected you and I'll show you the real skills list hiding in the body, then help you close the gap.