Loading
Preparing the page for you.
Preparing the page for you.
Why AI detection can misclassify writing, and how reviewers should handle uncertainty responsibly.
Core principle
AI detection results are guidance only.
No detector can prove authorship with certainty. These pages explain how to interpret results without overstating what automated analysis can prove.
Trust framework
The detector reviews writing behaviour such as repetition, structure, phrasing consistency, sentence variety and context-specific signals. It explains what influenced a result instead of presenting a black-box accusation.
Human and AI writing can overlap, especially in academic, formal, translated, edited or templated text. False positives and false negatives are possible, so results should guide review rather than decide outcomes alone.
Reports are designed around metadata, scores, signals, timestamps and guidance. Raw submitted text, uploaded file contents, prompts and provider payloads are not stored in reports.
Use detector outputs to support human review, editorial improvement, policy conversations and proportionate follow-up. Avoid punishment-only decisions based on an automated result.
Methodology detail
A false positive happens when human writing is flagged as AI-like. This can occur with formal templates, non-native writing, heavy editing, short samples or low-context content.
A false negative happens when AI-assisted writing appears human-like. Editing, rewriting, mixed authorship and paraphrasing can reduce visible AI-like signals.
Tighter detection can flag more suspicious content but may increase false positives. More cautious detection can reduce false positives but may miss some AI-assisted writing.
Short text, lists, resumes, marketing copy, translated writing and highly structured business content can be difficult to classify reliably.
Methodology pages explain the review approach, known limitations and why results should not be treated as proof.
Available resourceReports organise confidence, writing signals, recommendations and limitation notes into a review-friendly format.
Available resourceThe public status and transparency pages explain service availability, research principles and responsible claims.
Available resourcePrivacy pages explain temporary processing, report ownership and the platform's data-minimisation approach.
Available resourceFAQ
Short answers for responsible interpretation and privacy-first use.
Not always, but editing and rewriting can change signals. This is one reason results should be interpreted cautiously.
Treat it as a prompt for human review, not a final conclusion. Look for context, supporting evidence and policy fit.
Content discovery
Move from methodology into practical guides, privacy pages and product workflows.
Transparency
Explore the transparency pages that explain methodology, privacy, limitations and responsible use.
AI detection results are guidance only. No detector can prove authorship with certainty, and important decisions should include human review and appropriate context.
Scan reports are designed to show scores, signals and guidance without storing the original submitted text.
Read privacy detailsInternal research and benchmark work is used to improve calibration and understand failure modes. Public pages avoid unverifiable accuracy claims and do not expose private datasets or raw benchmark text.