AI Detectors: Why They’re Unreliable (and what teachers can do instead)

As the use of AI increased, many educators turned to AI detection tools as a way to crack down on plagiarism, but now realise that in practice, they are far less reliable than expected.

 

Let’s start with why AI detectors are appealing:

They typically promise to:

  • identify AI-generated text

  • provide a confidence or probability score

  • help teachers determine whether a learner’s work is “authentic”

So, what’s the problem?

  • AI detectors do not definitively identify AI-written work

  • They frequently produce false positives

  • Often struggle with short answers, edited text, or mixed human–AI writing

  • High-performing students, neurodivergent learners, and second-language writers are often disproportionately flagged.

    A detector score is not hard proof of plagiarism.

Why these tools are becoming less reliable:

Even when detectors improve, accuracy is unstable because:

  • AI models update rapidly and are constantly improving

  • Writing styles evolve over time

  • Students are finding ways around detection – e.g., using humanisers, rewriters, or paraphrasers

    Detectors are not a long-term solution.

The risk for schools:

Over-reliance on AI detectors can lead to:

  • inconsistent application of academic integrity processes

  • challenges when decisions need to be defended

    Though they may prompt a closer review of work, AI detectors should never replace professional judgement or be used as the sole basis for academic decisions.

Heavy reliance on detection tools can create a damaging culture of fear and mistrust.

When students believe they may be accused of misuse of AI, it can:

·       discourage risk-taking and genuine learning

·       damage student confidence and wellbeing

·       shift the relationship from learning partnership to surveillance

What’s better than detection?

More robust approaches include:

  • assessment design that values process, not just product

  • reflective or explanatory components

  • personalised or contextualised tasks
    clear expectations about acceptable AI use

    These approaches reduce reliance on detection altogether.


AI detection is one of the most common topics we are asked about.  TAKE AI focuses on practical alternatives to AI detection that support fairness, confidence, and academic integrity.

Previous
Previous

AI Policy in NZ Education - What Schools Actually Need (and what they don’t)

Next
Next

Assessment Integrity in NZ Schools - The AI Shift from Policing to Partnering