Spotting AI in Psychology A-Level Coursework: A Teacher's Guide

We've all been there: it's a Sunday afternoon, you're marking a batch of A-Level Psychology investigations, and you stumble upon a coursework submission that just doesn't sound right. The synthesis of Piaget and Vygotsky is a little too perfect. The methodology section uses vocabulary you've never heard the student use in class. Is it hard work, heavy tuition, or is it ChatGPT?

Detecting AI-generated student work in coursework-heavy subjects like Psychology A-Level is one of the biggest new challenges for teachers in the UK. When you suspect a student has used a generative model like Claude or Gemini to write their coursework, establishing proof is difficult. The ethical implications of a false accusation are severe. Yet letting academic dishonesty slide undermines the integrity of the qualification.

Why Detection is Probabilistic, Not Absolute

The first step in spotting AI-generated coursework is understanding that detection is inherently probabilistic. When AI detection tools analyse text, they aren't looking for a definitively hidden watermark. Instead, they scan for patterns across two main metrics: 'perplexity' (how predictable the next word is) and 'burstiness' (how uniform the sentence lengths and structures are).

For this reason, most tools offer a likelihood score from 0 to 100%. A high score doesn't mean the tool *knows* a machine wrote it; it means the linguistic patterns closely resemble those commonly produced by Large Language Models (LLMs). This is why you must treat the likelihood score as an assistive data point, not an absolute verdict. A student with a highly formulaic writing style might trigger a false positive, while a heavily edited AI draft might pass under the radar.

How to Handle Suspiciously High Scores

So, what should you do when a detection tool flags a piece of A-Level coursework with a high likelihood score? The key is professional judgment. You know your students better than any algorithm. Cross-reference the flagged coursework with work you know the student completed under supervised conditions in class. Look for stark contrasts in vocabulary, sentence complexity, or the depth of psychological evaluation.

If you're still suspicious, it's time to have a conversation. Frame the discussion around the coursework itself rather than making an immediate accusation. Ask the student to verbally explain a complex piece of analysis they've written, such as their justification for using a Mann-Whitney U test. If they authored the work, they should be able to articulate their reasoning. If they used AI, the gaps in their understanding will quickly become apparent.

Signals Suggesting AI Authorship

Beyond a high likelihood score, there are human-readable linguistic signals that suggest AI involvement. AI models tend to be highly structured and repetitive. They love concluding paragraphs that begin with "In conclusion" or "Ultimately." They often use transitional phrases like "Furthermore" and "Moreover" with unnatural frequency.

In the context of A-Level Psychology, LLMs often provide superficial evaluations. They might confidently state that a study lacks ecological validity without fully contextualising *why* that matters for the specific real-world behavior being measured. They also have a habit of referring to "the researchers" in a vague, detached manner, rather than engaging deeply with the specific methodology.

Assess With Confidence Using GradeOrbit

GradeOrbit's detection, included in every school account, is designed to support your professional judgment. Instead of asking whether coursework reads like AI in general, it compares a pupil's new investigation with that pupil's own earlier work already on file - run it from the pupil's profile or on the whole class at once. It gives you a score from 0 to 100 for how far the new work departs from the pupil's own writing, plus a confidence label - Low, Medium, or High.

Crucially, GradeOrbit provides a reasoning paragraph and lists the signals it noticed, allowing you to see exactly *why* the work was flagged. Detection is included in every school teacher seat, with no per-check charge, and a pupil with no earlier work on file is skipped rather than given a generic score. And because privacy is paramount, pupils' work is held under the school's Data Processing Agreement and never used to train AI models.

Start a free school trial and ensure your assessments remain fair and authentic.