AI detector vs plagiarism checker: what's the difference and which do teachers need?

Teachers often treat 'plagiarism checker' and 'AI detector' as interchangeable — different names for the same problem. They're not. The tools work differently, catch different things, and fail in different ways. Understanding the distinction helps you pick the right tool for the right situation, and avoid putting too much weight on either one.

What a plagiarism checker actually does

Plagiarism checkers — Turnitin, iThenticate, Copyleaks' similarity engine — work by comparing submitted text against a large database of sources: web pages, academic papers, previously submitted student work. They look for exact or near-exact matches. If a student copies several sentences from an article without citing it, a plagiarism checker will likely flag the overlap.

What they're good at is catching direct copying from findable sources. What they miss: paraphrasing, quotations deliberately left uncited, and anything sourced outside the database — private documents, obscure sources, or work passed between students with no online presence. Similarity scores also measure overlap without measuring intent, which is why a well-cited literature review can trigger the same percentage flag as an uncited copy.

What an AI detector actually does

AI detectors take a different approach. Instead of comparing text against a database, they score how statistically predictable the submitted writing is. Large language models generate text by repeatedly choosing the most likely next word; human writers deviate from those predictions more irregularly. Tools like GPTZero, ZeroGPT, and Turnitin's AI detection feature measure this deviation and produce a probability estimate of how much of a submission may be AI-generated.

The appeal is clear: if a language model produced the text, maybe you can catch it by identifying the pattern. The problem is that the signal is indirect, noisy, and sensitive to things it shouldn't be sensitive to at all.

The limitation both tools share

Both approaches share a structural problem: they analyse the output, not the process. A plagiarism checker can tell you that text matches a source; it can't tell you whether the student read the source thoughtfully and borrowed a phrase, or ran a search and lifted a paragraph. An AI detector can tell you the prose scores high for statistical predictability; it can't tell you whether that's because a model generated it, or because the student is an ESL writer who uses conservative vocabulary and formal sentence structures.

Text-based AI detection produces false positives on non-native writers at documented rates up to 61% — a finding that has led several universities to suspend or remove AI detection tools entirely. Plagiarism detection is more reliable but still triggers on common academic phrases and heavily cited work. Neither tool was built to answer the question you actually need answered: was this student genuinely present in the work?

Reading the process, not the product

The most defensible evidence isn't about what the text looks like — it's about how the work was composed. Genuine drafting has a recognisable shape: typing in bursts with natural pauses, revisions and deletions, a session length that reflects the difficulty of the task. AI-assisted shortcuts look different: a large block of content arriving in a single paste event, a session completed in a fraction of the expected time, or an unusually even typing cadence consistent with transcribing from another window rather than composing.

These process signals are language-neutral. They look the same whether the student writes in English, Mandarin, or Arabic, and they can't produce an English-proficiency bias because they never read the words at all. Learnaway is built on this approach — it records the timing and structure of writing sessions without ever capturing the text itself, giving teachers a process record that is useful regardless of a student's language background or writing style.

Which tool for which situation

For routine assignment submission, a plagiarism checker is still the right first tool — especially where source-copying is the primary concern. Most LMS platforms include basic similarity checking, and the approach is well understood by teachers, students, and administrators. For AI use specifically, text-based detection is a weak signal — useful as a prompt to look more closely, not as evidence on its own. If you use one, treat the output as the start of a conversation, not a conclusion.

Process evidence pays off most in assignments where the intellectual work is the point: essays, analyses, extended written responses. Collecting work through a tool that records the writing session means you have something concrete to point to if a submission looks implausible — without making assumptions about the student's fluency or writing style. Most teachers who've switched to this approach find they rely less on probabilistic AI scores and more on a question they can ask any student: walk me through how you wrote this.

In practice, the tools are complementary rather than competing. A plagiarism check covers source-copying; an occasional AI text check flags statistically suspicious submissions worth a second look; process evidence gives you the most defensible record for anything that warrants a real conversation. None of them replaces professional judgement. All of them are inputs to the question that actually matters: was this student genuinely engaged with the task?