Check if an essay was written by AI

An essay is a trail of decisions, not just a finished page. When a teacher only sees the final version, a sudden change in voice can be worrying, but it is still an incomplete picture. This matters now because schools are moving from a simple ban-or-allow argument towards clearer evidence of learning. A useful how to check if an essay was written by AI approach should help a teacher ask a better question, not pretend it can settle a case on its own. That is the promise of this guide: a practical route from uncertainty to a fair, explainable next step.

TL;DR: start with the assignment rules, look at the work in context, check the writing process where you have it, and speak to the student before treating a signal as a conclusion. The details below show what that looks like in a busy classroom.

Start with the question you actually need answered

Begin with the rubric and the allowed-use statement. An essay can be well written and still be the student's own; it can also include permitted AI for brainstorming but fail to acknowledge it. The key question is not whether every sentence feels human. It is whether the student can demonstrate the learning the assessment was designed to show.

A sound how to check if an essay was written by AI workflow is deliberately modest. It records what happened, distinguishes an observation from an inference, gives the learner a chance to explain, and keeps a short factual note. That sequence protects a conscientious student as much as it helps a teacher investigate a genuine concern. It also makes handover to a head of department far less fraught.

Before opening another tab, write one sentence that names the concern in neutral language. For example: “I need to understand how this conclusion was developed.” That is much better than “I need to prove this is AI”. The first sentence invites evidence; the second invites confirmation bias.

Choose evidence that fits the decision

Not every assignment deserves the same level of scrutiny. A low-stakes homework task may need a quick formative conversation. Controlled assessment or a final project may need a documented route under the centre policy. The table is a simple way to match the how to check if an essay was written by AI check to the consequence rather than making every flag feel like an emergency.

A practical route through the work

Read one paragraph slowly and mark three things: the claim, the evidence and the explanation. Then choose the least accusatory point to discuss. Ask where the source was found, why it was selected, or what alternative phrasing the student rejected. If process data exists, use it to frame a factual question about a large paste or an unusually short session. Give the student time to answer; rushing makes a fair conversation feel like an interrogation.

Keep the review narrow. Look for a change that can be described without guessing at motive: a large paste, a missing source trail, an abrupt change in terminology, or an answer the student cannot unpack. Then ask about that point. A student may have a perfectly ordinary explanation, such as drafting offline, using accessibility software, or receiving permitted feedback. Good process makes room for it.

In Learnaway, the relevant evidence is process data: session length, typing and paste events, pauses and focus changes. It does not read the words a learner types. That is why its output is best used alongside the teacher's knowledge of the class and the assignment brief. See how Learnaway works and the teacher workflow for the exact boundaries.

What current guidance says

The DfE's current guidance says schools and colleges should use professional judgement and address risks alongside opportunities. JCQ requires students to acknowledge AI use and requires centres to investigate doubts about authenticity. Neither source endorses a magic essay score. That is an important expert boundary for an AI essay detector: it can be a screening tool, but it cannot know the student's process or the permission they were given.

For a current external check, read the Department for Education's AI guidance and the relevant assessment guidance before changing a school rule. They are more useful than a vendor headline because they describe responsibilities, not a product claim. In the same 2025 Ofsted work, Chief Inspector Sir Martyn Oliver said schools need to “manage the risks” alongside AI's potential.

The common thread is professional judgement. Jisc's assessment and academic-integrity training explicitly covers the limitations of AI detection, authentication and responses to concerns. That is a helpful corrective to the idea that a percentage score is enough. It is not. A percentage is at most a prompt to look more carefully.

Make the route visible before a concern occurs. Students should know what evidence a teacher may consider, who makes the decision, how they can explain a legitimate workflow and when a formal policy applies. Clear process is not administrative clutter. It is the difference between a learning conversation and an unexpected accusation.

Illustrative classroom scenario

A sixth-form English teacher sees an essay that quotes a critic accurately but does not connect the quotation to the student's argument. She asks the student to annotate the paragraph with a different colour and explain the link. The student can explain the reading but admits that an AI summary supplied the sentence structure. The teacher now has a precise issue to teach: acknowledgement and independent analysis, rather than a vague claim that the whole essay is fake.

The useful detail is not the number on a dashboard. It is whether the teacher can say, calmly and accurately, what they saw and what they asked next. That leaves space for a student to demonstrate understanding, correct a misconception, or disclose help they did not realise needed acknowledgement. It also means a genuine issue can be escalated with a clear record rather than a hunch.

Common mistakes to avoid

Avoid comparing every student to their earliest piece of writing. People improve, use spellcheck, get feedback and write differently for different subjects. Also avoid demanding a student recreate an essay from memory. A better authentication task is a short, curriculum-relevant explanation that tests the particular thinking the essay should show.

Do not quietly change the rules after students have started. Do not put a detector score in the gradebook as though it were a mark. Do not collect more behavioural or personal data than the decision needs. The ICO's 2026 review covered audits of 28 widely used edtech providers and stressed that children need to be able to trust the tools schools adopt. Read the ICO statement before making data collection the default.

Turn the finding into a next step

At awareness stage, publish a one-page statement that distinguishes planning help, editing help and submitted text. At consideration stage, build a simple teacher checklist into your marking process. It makes expectations visible before a problem appears.

If you want to try a process-first approach, create one low-stakes assignment, tell students exactly what is recorded, and compare the timeline with work you already know well. Learnaway's free assignment flow is designed for that first small test. For a wider rollout, use the pricing guide and trust information to involve the people responsible for teaching, data protection and safeguarding early.