DSE in an AI World: What Examiners Will Actually Be Testing From 2026

I want to be careful about how I frame what I'm going to say, because the examiner perspective on DSE is sometimes presented as if there are hidden rules that only insiders know. There aren't. The assessment principles are publicly available. What I can offer is an account of how those principles are being interpreted in a changed context — specifically, in a context where AI tools are widely available to students and the examination community knows it.

The shift that is underway is not a dramatic break from what DSE has always valued. It is an intensification of it.

What DSE has always tried to test

Across subjects, and particularly in the humanities and language papers, DSE has always been designed to test the student's own thinking. Not the recall of facts — though knowledge is necessary — but what the student can do with knowledge under pressure, in real time, without external support.

The mark schemes reward: precision of argument over breadth of content coverage; analysis of evidence over quotation of it; structured independent thinking over template compliance; a student's genuine voice engaged with the question, not the performance of an essay form.

This is why experienced teachers tell students not to memorise model answers. A memorised model answer answers the question it was written for. The examination question will be phrased differently enough to make template answers obvious, and examiners are looking for responsiveness to the specific question in front of the student — not a prepared argument that happens to roughly fit the topic.

What is changing for 2026 and beyond

The examiner conversations I'm aware of — and I want to be careful not to claim authority I don't have, so treat this as informed observation rather than inside knowledge — are oriented toward two areas.

Analysis over recall, across subjects. The direction of mark scheme weighting is toward what the student can do with information rather than whether they possess it. In history and economics, this means the quality of argument construction. In English, it means coherence, precision, and evidence of genuine engagement with the text. In Chinese literature, it means interpretive judgment, not just accurate identification of features.

This is a deliberate response to AI's capabilities. AI is extraordinarily good at recall. It can produce comprehensive factual summaries, cover multiple perspectives, cite relevant historical examples. It does this smoothly, at length, without apparent effort. These are exactly the capabilities that template-heavy examination preparation produces in students who lean on AI.

What AI cannot do reliably is original analysis — the movement from "here are the relevant facts" to "here is what I think they mean and why I think that." This analysis requires the student to have an actual opinion, a genuine interpretive stance, and the ability to defend it against counter-arguments with specificity. It requires the student to have done the thinking, not just to have access to the information.

Examiners reading AI-generated work see the comprehensive factual surface and the generic analytical gesture — "this suggests that," "this demonstrates the complex relationship between" — without the underlying specificity of thought that distinguishes genuine analysis. That gap is what the mark scheme is increasingly designed to identify and penalise.

Evidence of the student's own reasoning process. In mathematics and the sciences, this manifests as renewed emphasis on working shown. An answer without working has always been risky; the direction is toward examining the reasoning itself, not just the outcome. A student who produces a correct answer via a flawed route — who was lucky, or who guessed, or who applied a memorised procedure without understanding it — can now be more reliably identified.

In language and humanities papers, the equivalent is responsiveness and specificity. Generic arguments that could apply to any examination question in the topic area are distinguishable, on close reading, from arguments that engage with the specific phrasing and emphasis of the question asked. Examiners are being trained to reward the latter and to be more sceptical of the former.

What students who rely on AI for written work will struggle with

Under examination conditions, without AI assistance, the student is alone with their own cognitive resources.

The student who has spent the past year primarily drafting essays by prompting AI, reviewing outputs, and editing them has practised a set of skills: direction-giving, evaluation of external text, editing. These are not worthless. They are, however, different from the skills the examination hall requires.

The examination hall requires the student to start from nothing. To hold the question in mind, construct an argument from their own knowledge, organise that argument in real time, and express it with sufficient precision to earn marks. The tolerance for blank-page uncertainty — for the disorienting moment at the start of an essay question when the argument hasn't formed yet — is something that has to be practised. It cannot be learned by editing AI outputs.

I work with S5 and S6 students directly. The ones I'm concerned about are not the ones who use AI strategically and know when to put it down. They're the ones whose writing fluency under their own power has noticeably declined — who can produce polished extended prose in informal settings but struggle to sustain coherent argument for 45 minutes without a prompt structure to lean on.

The gap shows up first in timed in-class writing, which I use as a diagnostic. Then, more seriously, in mock papers. By the time a student reaches DSE having never practised sustained independent writing at speed, it is very late.

How to prepare for the post-AI DSE

The preparation principle is not new. It is simply more urgent.

Practise writing under timed conditions, without AI, regularly. Not as punishment. As training. The examination hall is a specific performance context and, like all performance contexts, it rewards preparation that matches the conditions.

Develop a genuine argument, not a structure. The students who perform best at DSE aren't the ones with the cleanest essay templates. They're the ones who have actual views on the texts they've studied, the historical events they've analysed, the economic arguments they've evaluated. Having a view requires thinking, not prompting.

Read the question with obsessive care. The examination question is the point. Every mark scheme I have ever worked with rewards answers that are genuinely responsive to the specific question asked. Students who write about the topic rather than about the question are writing for themselves, not for the examiner.

Use AI to debate with, not to write for you. "Here is my argument for this essay question. Challenge it." This is productive AI use. It makes you think harder. It identifies weaknesses in your reasoning before the exam hall does. The condition is that you do the writing yourself, after.

The examination is still, in 2026, a test of what the student can do alone. The preparation has to build that capability.