Guides

AI-assisted writing starts with a human draft and uses AI to help; AI-generated writing starts with a prompt. See the spectrum, how detectors and policies treat each.

What AI detection false positives are, what research shows about non-native English writers, why short work is risky, and a fair review process for educators.

Metadata can confirm an AI origin when it survives, but missing metadata proves nothing. EXIF, IPTC, C2PA and SynthID explained, with their limits.

A deepfake alters real footage of a real person; an AI-generated video is synthesized from a prompt. Definitions, a comparison table, sourced cases and checks.

AI detector accuracy depends on the text, model and edits. Here is what independent studies and vendor documentation report, with the conditions behind each figure.

No single sign proves an essay was written by AI. Here are the signals worth reviewing, how much weight each deserves, and a checklist for a fair evidence review.

No single check proves an image is AI-generated. Here is the evidence that holds up: provenance, C2PA credentials, metadata, context, visual clues and detector limits.

Practical signs of AI-written text, why no single sign proves anything, and how to combine your own reading with a detector result.

There is no single detector teachers use. Turnitin, GPTZero, Copyleaks, Pangram and others compared on highlighting, file upload and education fit, with check dates.

A plagiarism checker finds text matching existing sources. An AI detector flags writing that resembles AI output. What each does and why neither replaces the other.

Image detectors that score near-perfect in the lab can fail on new generators, compressed and edited images. Here is what the benchmarks show and what a score means.

AI video detection combines frame, motion, face, lip-sync, audio and provenance checks into a confidence score, not a verdict. Here is how each works and where it fails.

Revise an AI draft so it reads like you: keep your own view, add specifics, cut filler, check every fact and citation, match the reader, and disclose AI help.

No single visual tell proves a video is AI-made. Start with source, context and provenance, then check motion and audio, and weigh detector results with care.

A flag is a reason to look, not proof. A 7-step workflow for teachers: review the passages, compare earlier work, check sources, talk to the student, and follow policy.

AI prose reads flat because models favor the most likely wording. See seven robotic tells with generic-versus-specific examples and how to fix each one.

Why a detector can label your own writing as AI-like, which kinds of writing are most affected, and how to respond calmly if the result is being questioned.

Yes. Here is how AI detectors produce false positives and false negatives, who is most affected, and how to treat a result responsibly.

Our 'Flagged as AI-like' figure is the share of passages labeled AI-like, not the probability that a text is AI-written. Here is how to read it.