Medicine and dentistry do not begin on the ward or in the surgery. They begin years earlier, in the machinery we use to decide who is allowed through the door. If healthcare needs to be redesigned from the ground up, then admissions is part of the foundation.
I have spent the last few weeks helping a seventeen-year-old assemble a dentistry application, and the experience has left me uneasy in a way I did not anticipate.
Here is what an admissions tutor will see. A set of predicted grades. A UCAT score. A reference. And roughly 4,000 characters of prose, now split across three questions, in which the applicant must establish motivation, academic readiness, manual dexterity, communication, empathy, resilience, insight into the profession, and evidence that they understand what they are signing up for. That is the entire picture. Five years of training, forty years of clinical practice, and a licence to put instruments into the mouths of strangers — decided, at the shortlisting stage, on less text than this article.
The obvious response is that interviews exist, and they do. But interviews are rationed. Most schools use the paper application to decide who gets one. The document I have described is not the final judgement; it is the gate.
So the question is worth asking properly, with evidence rather than instinct: is this a defensible way to choose dentists and doctors?
The uncomfortable finding about personal statements
Start with the component everyone agonises over. The honest answer from the research literature is that the personal statement, as historically used, predicts almost nothing.
A meta-analysis of the available studies found that ratings of personal statements were modest predictors of graduate GPA and faculty performance ratings, with no incremental validity beyond standardised test scores and prior GPA. In other words, once you know an applicant's grades and test score, the statement adds no further predictive power. A more recent Dutch study went further, modelling what would happen if personal statement ratings were folded into an admissions formula at different weights, and concluded that including them would not have improved prediction of first-year GPA or dropout at best, and would have damaged validity at any weight above 10%.
A widely cited review of admissions tools puts the personal statement in blunt company: the instruments that have shown predictive validity include prior academic scores, aptitude tests and the multiple mini-interview, while the list that has not robustly met that bar is longer — the traditional interview, the personal statement, letters of reference, personality testing and emotional intelligence measures.
The UK evidence is no kinder. The largest undergraduate study of its kind, conducted at Dundee, found that no score derived from the UCAS form correlated significantly with examination performance, while the MMI was the most consistent predictor across two cohorts. A parallel study across Aberdeen and Dundee reached the same conclusion: neither UCAS form nor interview scores were statistically significant predictors of examination performance, whereas the aptitude test produced validity coefficients between .24 and .36.
Set that alongside the hours applicants, parents and teachers pour into these documents, and something has gone wrong with the economics of the exercise.
Worse than useless: actively unequal
If the statement merely failed to predict, it would be a waste of everyone's time. The stronger charge is that it introduces bias.
The foundational UK work here is Steven Jones's research for the Sutton Trust, which analysed 309 personal statements submitted to the same department of the same leading university by applicants with identical A-level results. Holding attainment constant, the differences were stark. State school applicants made around three times as many written errors, and reported fewer relevant activities — an average of 2.97 against 3.63 for independent school applicants. The qualitative gap was starker still: work placements at investment banks and accountancy firms abroad on one side, part-time retail jobs on the other.
Jones's later work found that the differences in quality reflected inferior support rather than inferior ability, and that school type was a key predictor of statement quality — and that teachers and Russell Group admissions tutors held markedly different views about what a good statement even looks like. Applicants at schools that send few students to selective universities were, in effect, being coached toward the wrong target.
A 2022 report from the Higher Education Policy Institute, based on analysis of 164 drafts from 83 applicants from under-represented backgrounds, plus interviews and surveys, concluded that the free-text statement was incompatible with the fair admissions principles set out by Universities UK and GuildHE, imposing an unnecessary burden and contributing to inequalities in access. It recommended replacing it with short-response questions.
This is not a marginal academic dispute. It is the reason the format changed.
What the 2026 reform fixed, and what it did not
From 2026 entry, UCAS replaced the single free-text box with three structured questions: why this course, how your studies prepared you, and what you have done outside education. The rationale was explicit — the questions were introduced so that students from all backgrounds would better understand what universities actually want to know. Early evidence suggests it has landed where it was aimed: 73% of teachers in schools with the most disadvantaged intake said the change improved the process for their students, against 61% in the most advantaged schools.
I think this is a genuine improvement, and I say that as someone who spent a fortnight discovering how much easier the new format is to advise on. The blank page was a coaching multiplier. Three labelled questions tell every applicant what the room wants.
But the reform addresses the presentation problem, not the substrate problem. The structured format helps a disadvantaged applicant write up their experiences more effectively. It does nothing about the fact that they have fewer experiences to write up. A student whose father is a surgeon, whose aunt is a physician, and who can arrange two weeks shadowing an implant surgeon is not advantaged by rhetoric. They are advantaged by access. No question format corrects that, and it would be dishonest to pretend otherwise.
I know this because the applicant I have been helping is on the fortunate side of it.
The things that do work
It would be easy to write a piece that only demolishes. The literature is clearer about what works than the public debate suggests.
Prior academic attainment remains the most reliable single predictor of academic performance in medicine and dentistry, as it is across most disciplines. That is not in serious dispute. The dispute is about what it costs: reliance on prior attainment has been challenged on the grounds that it advantages students from higher socio-economic backgrounds and excludes many who would become exceptional practitioners.
The UCAT does modest but real work. The largest national study, following 6,812 entrants to UK medical schools, found it significantly predicted performance in theory- and skills-based examinations, with associations surviving adjustment for prior educational achievement. A systematic review is appropriately restrained: cognitive total and verbal reasoning scores are weakly positive predictors of academic performance and Foundation Programme outcomes, adding a small amount of incremental validity over prior attainment. Small, but not nothing — and notably, it is the one instrument explicitly designed to be less sensitive to background.
The multiple mini-interview is the strongest performer for non-academic attributes. Eva and colleagues at McMaster reported a median reliability of 0.73 across administrations using twelve ten-minute stations, and correlations of 0.43 in a postgraduate sample and 0.35 in an undergraduate sample against an OSCE-based licensing examination taken five years after the MMI itself. A five-year predictive window for a personal-qualities assessment is remarkable. The earlier work found MMI scores correlated with clerkship performance and licensing examination measures, and retained predictive power after controlling for other variables.
So the sequence that actually predicts is: grades, then aptitude test, then a structured multi-station interview. The document everyone loses sleep over is the weakest link in the chain.
The dentistry-specific problem nobody discusses
Most of this debate is conducted about medicine. Dentistry is quieter, smaller, and — on the available evidence — has a sharper equity problem.
A study of 3,246 applicants across ten UK dental schools produced findings that should trouble anyone in the profession. White applicants were significantly more likely to be selected than Black applicants (OR 0.25), Asian applicants (OR 0.57) and those of mixed ethnicity (OR 0.80), while the least deprived were significantly more likely to be selected than the most deprived (OR 0.59). Grammar school education increased the odds of selection by 1.8 compared with state school education. The authors concluded that widening access measures will only succeed if selection committees radically change their processes to address systemic biases, and noted that the UCAT reduced ethnic disparities while introducing disparities between other groups.
A longer analysis of UK dental and medical admissions across 1996–2011 found the same pattern persisting over fifteen years: admission odds were lowest for applicants who were male, mature, of Black ethnicity, or from a lower socioeconomic classification.
This matters clinically, not just morally. We have parts of this country where NHS dental access has effectively collapsed. A workforce drawn disproportionately from grammar schools and the least deprived quintiles is not obviously the workforce that will staff coastal and rural practices. Selection is workforce planning, whether we treat it as such or not.
What other countries do
The Netherlands offers the most instructive natural experiment in the world, because they have run both systems.
A weighted lottery operated nationally from 1972, with secondary examination scores determining the weighting — outstanding candidates tripled their chances relative to a bare pass, and rejected candidates could re-enter for two subsequent years. For decades, schools and the public were broadly content with it. Then in 1996 an outstanding school leaver was rejected three times and appealed, prompting public and political anger that gradually dismantled the system; lotteries were prohibited outright from 2017. And then the reversal: a parliamentary majority voted in 2020 to permit lotteries again, effective 2023. Eight Dutch medical schools subsequently indicated they would replace their selection procedures wholly or partly with a lottery, on the grounds that the existing system advantaged students who could afford tutoring or access internships through family networks.
Before anyone in the UK reaches for this as a solution, read the retrospective multi-cohort study that examined the transition. Its finding is genuinely two-sided: as Dutch health professions education moved from lottery to selection, applicants' background characteristics increasingly influenced their odds of admission — but returning to a lottery is not a solution for achieving a representative student body, because the applicant pool itself lacks diversity. A lottery among a narrow pool produces a random sample of a narrow pool. The related literature notes that selection yields only small gains over a lottery procedure, while the diversity needed to serve an increasingly diverse patient population may be compromised.
That is the honest shape of the evidence. Selection barely outperforms chance on outcomes, but chance does not fix representation either. The problem sits upstream of both.
And now, artificial intelligence
I advise the House of Lords on AI in medicine, so I will not pretend this is a small complication.
Every UCAS personal statement is screened by a similarity detection system called Copycatch, and statements exceeding a 30% similarity threshold are reviewed by the verification team, with applicants and their universities notified. Against wholesale copying, this works.
Against generative AI, it is a different matter. HEPI's own commentary is candid: AI detection software such as Turnitin is, for now, quite unreliable, with a high false positive rate — text written by humans is often rated as AI-generated, and using it uncritically would mean falsely accusing students. Meanwhile most teachers and advisers surveyed at a UCAS conference believed they could spot AI in a personal statement most of the time — a confidence the detection evidence does not support.
We now have an assessment instrument that does not predict performance, is sensitive to social advantage, and can be produced to a competent standard in thirty seconds by a free tool. Whatever signal remained is degrading quickly. And the advantage does not disappear with AI — it shifts to those who know how to use it well, and who have real material worth writing up in the first place.
What I would change
I am not arguing that we abolish the written application. I am arguing that we stop asking it to do work it cannot do.
Use it as a threshold, not a ranking. The HEPI report's recommendation was precisely this: short questions establishing baseline suitability rather than instruments for discriminating between strong candidates. Does this applicant understand what dentistry involves? Pass or fail. Stop trying to rank prose.
Shift the weight to structured multi-station assessment. The MMI is the best-evidenced tool we have for personal qualities. It is expensive. It is also the only part of the process with a demonstrated five-year predictive relationship to clinical examination performance.
Verify experience rather than rewarding its acquisition. A week with an implant surgeon should not outscore a Saturday job in a care home. What matters is what the applicant understood, and an interview tests that far better than a paragraph does.
Contextualise seriously, and audit the result. Most dental schools now run some form of contextual admission. Very few publish whether it has moved the demographic numbers. The 2012–13 dental cohort data exists because researchers went looking. That should be routine, published, annual.
Accept that selection cannot fix the pipeline. The Dutch evidence is unambiguous on this. If the applicant pool is not diverse, nothing downstream will make the admitted cohort diverse. That is an outreach and schools problem, and dentistry does considerably less of it than medicine.
The honest conclusion
Is the current system fair? Not entirely. But the alternatives are worse than their advocates admit, and the component attracting most of the criticism — the statement — is, after the 2026 reform, a genuinely better version of a fundamentally limited instrument.
What troubles me is the mismatch between effort and consequence. Families spend months on 4,000 characters that the evidence says predicts nothing, while the parts of the process that do predict something get a fraction of the attention and, in some schools, a fraction of the budget.
Somewhere there is a seventeen-year-old who will make a superb dentist and who has never met one socially, cannot arrange shadowing, and has nobody at home to read a draft. The structured questions help her. They do not find her. Until we are serious about that, we are refining the measurement of an advantage we have not addressed.