If AI Is Making It Harder for Gen Z to Enter Work, Recruitment Needs Redesign — Not Just Better Algorithms

The entry-level squeeze

An FT discussion published on 13 August described weaker graduate outcomes and increasingly automated recruitment, while cautioning that AI is not the only explanation for a difficult labour market (Financial Times, 2026) ↗. That distinction is important. Economic conditions, hiring cycles and employer confidence all affect graduate opportunities.

AI may still be changing the structure of the problem in two ways at once: it can perform some of the routine tasks historically given to juniors, and it can lower the cost of both applying for and screening large numbers of applications.

When applications become cheap, signals become noisy

Generative tools let applicants produce more tailored-looking applications quickly. Employers respond with more automation. Candidates then experience rapid rejection and may increase application volume further. This can become a low-trust loop in which both sides optimise throughput while the quality of matching deteriorates.

The answer is not to ban AI from recruitment. It is to decide which signals still have meaning. Work samples, structured tasks, verified skills, human conversations and contextual evidence may become more valuable when generic prose becomes cheap.

The emerging evidence on job exposure

Henseke and colleagues' UK task-based preprint estimates broad exposure of jobs to generative AI and reports a 6.5% decline in postings for high-exposure roles after the release of ChatGPT (Henseke et al., 2025) ↗. Because this is emerging preprint evidence, it should not be read as a final causal estimate for youth employment. It is still a useful warning signal: the task composition of entry roles may be changing before institutions have redesigned the pathways into them.

Fairness is a governance issue, not a model feature

The UK government has pointed employers to Equality Act obligations, ICO work on automated decision-making and Responsible AI in Recruitment guidance. A June parliamentary answer specifically noted the need to avoid new barriers for applicants with protected characteristics (UK Parliament, 2026) ↗.

That means 'the vendor says the model is unbiased' is not a sufficient control. Employers need to know what data enters the decision, what the tool influences, how outcomes are monitored and where a human can intervene.

The Sanctuary perspective: preserve the capability ladder

Organisations should treat entry-level jobs as part of their capability supply chain. If AI removes low-value tasks, redesign junior roles around supervised analysis, client exposure, quality checking, AI-assisted production and learning — not around a smaller number of impossible-to-enter 'already experienced' jobs.

Our human resources and people support approach would pair responsible screening with structured development after hiring. Selection and learning should be designed together.

A practical recruitment redesign

A stronger process could include a transparent application stage, one validated skills task, structured human review, clear disclosure where AI meaningfully influences assessment, and post-hire development milestones. Recent UK evidence on AI upskilling also points towards practical, task-based and continuing learning rather than one-off generic instruction (Department for Science, Innovation and Technology, 2026) ↗. Employers should monitor conversion rates across groups and test whether screening criteria actually predict job performance.

The strategic question is simple: if every firm automates the training ground away, where will experienced workers come from in five years? Responsible AI recruitment should protect not only fairness today but the production of capability tomorrow.