All posts

What AI detector scores can and can't tell you

November 4, 2026 · 5 min read · Assignora AI

AI detectors look for statistical patterns that are more common in machine-generated text, and they turn these into a score. That score is an estimate. It can't observe who wrote something, or how.

Why scores can be wrong

Clear, formulaic and carefully edited writing can resemble what a detector expects from AI. Detectors can also miss AI text, particularly after it has been edited. Different tools can give different results for the same passage, and results can change as tools are updated.

Some writers carry more of this risk than others. One peer-reviewed study found that GPT detectors flagged writing by non-native English speakers as AI-generated far more often than writing by native speakers.

Source: Liang et al. (2023), GPT detectors are biased against non-native English writers. Patterns, volume 4, issue 7.

What to do if your work is flagged

  • Ask how the score is being used, and what happens next.
  • Gather your evidence: drafts, notes, sources and version history.
  • Explain how you worked, and offer to talk through your argument.
  • Read your institution's academic integrity policy, including its appeals process.

What helps more than a score

The best protection is a record of your process, kept as you work. Save versions as you go, keep your notes, and keep your sources with your draft. It's much easier to show how a piece of work developed than to argue with a number.

A word on rewording

It can be tempting to keep rewording honest work to bring a score down. That rarely solves the real problem, and it can make your writing worse. If you've used AI in a way your course allows, disclosing it is a stronger position than trying to hide it.

Rules vary between institutions and courses, so your course's own policy is always the one to follow.

Assignora puts these habits into a workspace

Real sources, clear plans and AI that suggests while you decide.

Explore AI Tools