Graduate hiring in the UK rose 4% across the 2025/26 cycle, after an 8% fall the year before (Times Higher Education, 2026). That is the number the market is planning against. It is also a mean, and the mean is carrying a small number of very large employers who switched recruitment back on. Strip them out and the increase is 0.8% (Wonkhe, 2026) โ and on the Institute of Student Employers' own decomposition, as reported by the Financial Times, the typical employer changed nothing at all.
The graduate hiring recovery, in other words, happened to roughly a dozen organisations. Everyone else is reading someone else's recovery as if it were their own forecast.
The figures come from the ISE's Student Recruitment Report 2026, the largest survey of student employers in the UK: "155 employers, covering more than 1.7 million applications and 31,577 hires and work experience placements during the 2025/26 student recruitment cycle" (Institute of Student Employers, 2026). Within that dataset, 45% of graduate employers are increasing hiring and 40% are decreasing it (Wonkhe, 2026). That is not a trend. That is two groups moving in opposite directions, averaged into a headline.
A mean, an ex-outlier mean, and a median walk into a planning meeting
Three numbers describe the same cycle and they are not alternatives:
- +4% โ the overall change in graduate hires across the sample.
- +0.8% โ the change once the handful of large employers restarting programmes is removed.
- No change โ what the typical employer actually did.
Next year repeats the shape. The headline forecast is a 7% increase (Wonkhe, 2026); on the ISE's ex-outlier decomposition reported by the Financial Times, it is 2%, with the median employer again expecting to make no change.
If you run operations at a 50โ500 FTE company, none of those outliers is you. The planning error is not optimism โ it is importing a distribution's tail as your own baseline. An external market that is flat for the median employer means your entry-level pipeline will not be rescued by a rising tide, and the volume pressure on your funnel will not ease either.
The funnel pressure is the part that is unambiguous. Employers in the sample received 141 applications per graduate vacancy; in London, 172 (Times Higher Education, 2026). Those are the conditions in which screening shortcuts get adopted fastest, and in which the shortcut you pick quietly becomes your hiring policy.
Graduate hiring stabilised. The rung you screen on did not.
Here is the finding that should change a decision rather than a slide.
While graduate hiring stabilised, work experience โ the thing employers screen graduates on โ moved the other way. Applications for summer internships rose 27% and applications for undergraduate placements rose 15%, while internship and placement hiring fell 8% (Times Higher Education, 2026). Demand up by a quarter, supply down by a tenth, in the same year.
Run that forward one hiring cycle. The CV line "relevant internship at a recognised employer" has just become materially harder to obtain, and the employers who made it harder are the same large organisations whose restarted graduate programmes produced the 4% headline. The signal is being rationed by its own issuers.
That matters because of what the signal then measures. When a credential is abundant, possessing it tells you something about the holder. When it is scarce and allocated through a 27%-oversubscribed competition at a small set of firms, possessing it tells you mostly about access โ who could afford an unpaid summer in London, who had the network to hear about the opening, who applied to forty of them. The ISE itself frames work experience as increasingly valuable for developing professional behaviours, adaptability and judgement in an AI-shaped market. Both things are true at once: the experience is genuinely formative, and its presence on a CV is now a weak and biased proxy for whether a given candidate has those behaviours.
So if your entry-level screen filters on prior internships, you are running a filter whose discriminating power is falling every cycle while its adverse-impact risk rises.
The external market is not supplying the gap
It is worth checking whether the mid-market can simply wait for the wider market to loosen. It is not loosening. Indeed's mid-year data put UK graduate job postings down around 7% year on year as of 10 July 2026 โ the lowest for that point in the year since 2020 โ against overall UK postings 32% below their pre-pandemic baseline (Indeed Hiring Lab, 2026). The ISE sample is skewed towards large structured-programme employers; Indeed captures the smaller firms, and its picture is bleaker. There is no cavalry.
"Graduate quality" is not the constraint
The convenient explanation for a frozen junior layer is that candidates have got worse. The data does not support it.
94% of employers in the ISE sample said they secure the quality of graduate hires they need often or almost always (Times Higher Education, 2026). Employer confidence in the talent is close to unanimous. What has got harder is verification: 92% of employers suspected or confirmed student use of AI somewhere in assessment and selection, and 66% report detecting cheating at the interview stage (Wonkhe, 2026). Wonkhe's summary of the employer position is the sharpest line in the whole report: the priority is "not AI-driven recruitment, but AI-resilient recruitment."
That is a precise and useful distinction. An AI-driven process adds a model to the existing screen. An AI-resilient process asks a different question: which of my signals still carry information when every candidate has a drafting assistant and a scarce credential market?
Unstructured CV screening fails that test. So does an essay-equivalent written task. So, increasingly, does the internship line โ not because candidates game it, but because the market stopped issuing it to anyone outside a narrow channel.
The counter-argument, stated fairly
Three limits on how far this travels, because they change the strength of the conclusion rather than its direction.
It is a UK dataset of large employers. The ISE polls its own membership โ self-selected, annual, weighted towards finance, technology and professional services firms running structured graduate programmes. It is not a random sample and it is not a measurement of mid-market firms.
The mid-market read is an inference. The claim "this is also your market" rests on the outlier decomposition plus the independent Indeed series, not on direct observation of 50โ500 FTE companies. I would call it a strong signal, not a measured result.
The 0.8% figure has a single primary source. It is attributed to the ISE in the Financial Times' reporting and corroborated in Wonkhe's write-up; the report body itself sits behind member access or a ยฃ216.67 purchase, so neither I nor you are reading it directly. Treat the three-number decomposition as well-sourced secondary reporting of a paywalled primary.
What survives all three caveats is the mechanism, which does not depend on the sample at all: when the issuers of an entry-level credential cut supply 8% into a 27% demand increase, that credential becomes a worse measure of capability. That is arithmetic, not a UK phenomenon.
What to decide this quarter
The decision in front of a Head of Operations is not whether to compete in a recovering graduate market. It is whether to keep screening on a signal that large employers just stopped producing.
Three moves, in order of how much they cost you:
1. Audit what your entry-level screen is actually filtering on. Pull your last two junior hiring rounds and check how many candidates were eliminated before any assessment because of an experience field โ internship, placement, "relevant exposure." That number is your exposure to a rationed credential. If it is large, you are not screening for capability; you are screening for access to a specific set of employers.
2. Replace the proxy with a measurement, not a different proxy. The reason the internship line persists is that it is cheap and nothing else sits in the slot. Something does now: a verified capability profile measures skills, cognitive ability and role-relevant judgement directly, per skill and with a percentile rank, rather than inferring them from where a 21-year-old spent a summer. Scovai's Talent Passport is built on exactly that substitution โ assessment, cognitive battery, personality profile and a structured career interview, certified at four levels, so a candidate who never got the internship can still arrive with evidence. The point is not the instrument; the point is that at 141 applications per vacancy you need a signal that scales and survives AI-assisted applications, and experience fields do neither.
3. Decide whether you are a producer of the signal or only a consumer of it. Large employers cut work experience places 8% in a year when applications rose 27%. A mid-market operator who stands up even four structured placements is now buying into a market where โ on the Financial Times' reading of the ISE figures โ 102 applications chase each internship place elsewhere: exceptional selection leverage, at a cost most 200-FTE companies can carry. The organisations complaining loudest about junior talent scarcity are, in aggregate, the ones who stopped manufacturing it.
The 4% graduate hiring figure was never your number. The question that is yours: by the end of this quarter, can a candidate with no internship and no brand-name employer on their CV get through your entry-level screen on evidence of capability alone? If not, you are not hiring from the graduate market. You are hiring from the shortlist that twelve large employers finished with.