Why Graduate Schemes Reach for AI Interviews First

Why Does Every Graduate Scheme Have One?

Volume is the reason. The Institute of Student Employers reports that UK employers received an average of 140 applications per graduate vacancy in each of the last two years ¹. That is up from 38 per vacancy in 2002/3 and 86 by 2022/3, and retail, FMCG and tourism run highest at 290 applications for every graduate role ¹.

It is the highest the series has recorded since ISE began collecting the data in 1991 ¹.

Two limits on that number matter. It counts applications received per UK graduate vacancy, reported by employers who skew large and structured, so it is not the figure for an ordinary job advert. And the climb began before generative AI existed, and the source names competition as the cause.

A first round built for that pile is hard to run as a conversation.

Selection researchers put the same constraint more precisely. In the meta-analysis where structured interviews are the top-ranked predictor, Sackett and colleagues note that in their traditional in-person form those interviews are not generally viable for high-volume jobs ². As administration and scoring move toward technology and automation, the same paper says, they become more amenable to mass administration, and validity needs to be evaluated under those changed conditions ². That is an open question being named, not an endorsement.

Employers work the same arithmetic from the other side. One question they weigh is whether it is cheaper to assess every applicant than to screen resumes at all.

How Old Is the Recorded First Round?

Older than the AI wave it gets blamed on. In June 2016 Goldman Sachs announced it would stop running in-person first-round interviews for undergraduates on college campuses, and would interview applicants through a pre-recorded video platform instead ³. The change there was to recorded video reviewed by people. Nothing in the account of that announcement says anything about scoring by software, and nothing here says what the bank does today.

That account comes from the interviewing vendor's own blog rather than from the bank itself, so read it as a dated record of an announcement rather than as a description of how the bank interviews now. The one-way recorded round is not this year's invention.

What changed recently is what happens to the recording.

A recording can be watched by a recruiter, transcribed and read, scored by a model against a rubric, or some combination of the three. Those are different products sold under one phrase. Reading what actually happens in an AI interview in each of the common formats tells you what each one records and who reads it.

Check Who Reviews Your Recording

Read the employer's own process page, because some of them answer this in writing. Unilever's UK and Ireland page for the 2026 Future Leaders Programme describes its third stage as a digital interview: three short questions, then a business case, recorded whenever and wherever the candidate chooses ⁴. The same page states that the company does not use AI as a screening tool in its recruitment, so the submission is reviewed by a real person ⁴.

That is one programme in one market, on a page for one intake, and it is an employer describing itself rather than an audit. It also says nothing about whether the earlier online assessment is scored by software. Read it for what it covers.

Four places are worth checking before you decide the format is opaque.

• The scheme's own process or FAQ page, which is where Unilever's wording sits
• The invitation email, which sometimes names the platform running the stage
• The platform's candidate help pages, which describe what the tool records
• The recruiter, in one line: who reviews the recording, and is any of it scored by software

Asking who reviews a recording is a normal process question, not a complaint about the format.

The answer changes less about your preparation than you would think. The work is a real example, said plainly, inside the time the question allows. What it changes is where a complaint goes if the stage goes badly.

It also tells you whether asking for a live conversation instead is worth the message.

Should You Finish It or Skip It?

Finish it deliberately, because in a randomised test against a resume screen, the people the interview track advanced did better in the final round. Researchers randomly assigned 37,000 applicants for a junior developer role either to resume screening followed by human selection, or to a pipeline that added a structured video interview run by AI first ⁵. Both tracks then faced the same final human interview, with interviewers blind to how each candidate had arrived ⁵.

In that experiment 54% of the AI-assisted track passed the final interview, against 34% from the traditional one: an average treatment effect of 20 percentage points, with a standard error of 12 points ⁵. The standard error belongs beside the headline. An effect measured that imprecisely is directionally informative and no more.

One employer, one country, one junior role, and a preprint.

The completion number is the part worth carrying. Of the 23,000 candidates invited to the AI interview, 24% completed it ⁶. Those who declined were slightly older and more experienced, and their later labour-market outcomes, read from LinkedIn profiles, did not differ statistically from the completers' ⁶. Inside that pipeline, though, not completing meant not advancing, and non-completion mixes refusal with a dead link, a lost email and a job found elsewhere.

Among a self-selected sample of UK early-careers respondents, calling AI use at this stage unfair is the majority view. In the Prospects Early Careers Survey 2026, a majority of respondents said employer use of AI was unfair at every stage except initial qualification screening, with about 83% calling it unfair in an online recorded interview ⁷. That measures what people think is fair, not what happened to them.

Employers get the hiring-side version of this question in what an early-career assessment should actually measure, which tells them to test the work a first-year hire actually does.

For a candidate with a thin resume, the recording is a place to put the real example the resume could not hold.

See how it works >