Similarity checkers like Turnitin and iThenticate have been part of thesis submission for years. More recently, many of these platforms added AI-writing detection — a score that estimates how much of a document may have been produced by a generative AI tool. For scholars, these scores can be confusing and, sometimes, alarming.
How AI detectors work
Unlike plagiarism checkers, AI detectors do not compare your text to a database of sources. They use statistical models to judge how predictable your writing is. Text that follows very common word patterns tends to score as "likely AI". That is why detection results are probabilities, not proof.
Why false positives happen
Because detectors look for predictable language, some genuine human writing is more likely to be flagged, for example:
- Writing by non-native English speakers who use simpler, formulaic sentence structures
- Highly standardised sections such as methods descriptions and definitions
- Short passages, lists and text with many technical terms
- Text that has been heavily polished with grammar tools
In February 2026, Washington State University cancelled its Turnitin AI-detection contract, citing disputed false-positive rates, concerns about English-language learners and neurodivergent students, and integrity cases that relied on detector scores without other evidence. Several other universities have also limited how detector results can be used.
Similarity score vs AI score
These are two different things. A similarity score shows matching text from published sources and is still the main check most universities apply — in India, for example, UGC's 2018 academic-integrity regulations set graded consequences based on the level of similarity. An AI score is an estimate and, at many institutions, is not accepted as evidence of misconduct on its own.
How to protect your work
- Keep your drafts. Version history in Word, Google Docs or Overleaf shows how your writing developed over time — strong evidence of authorship.
- Save your notes and data files alongside each chapter.
- Disclose any AI assistance honestly, following your university's policy.
- Write in your own voice. Specific examples from your own data and context read as human because they are.
- Do not use "humaniser" tools to disguise text — they can distort meaning and may themselves breach integrity rules.
What to do if your work is flagged
Stay calm and ask for the full report. Request a meeting with your supervisor, share your draft history and notes, and explain your writing process. Many policies require additional evidence beyond a detector score before any finding is made.
The bottom line
AI detectors are screening tools, not verdicts. The best protection is a transparent writing process: drafts you can show, sources you have read, and an honest disclosure of any AI help.
