Guides
How provenance marks work, why detectors produce false positives, and what evidence establishes authorship. Written to be read by someone who has just been accused of something.
AI detection false positives: what they are and what to do
Why AI detectors flag human writing, which kinds of writing get flagged most, and a practical sequence for responding to a false positive.
The ChatGPT watermark: what it is, where it applies and who can read it
OpenAI is adding an invisible text watermark to ChatGPT and Codex in the EU. How textGrain works, what OpenAI says about editing, and why a third-party tool cannot read it.
ChatGPT vs Claude: how the two text watermarks differ
OpenAI watermarks ChatGPT text in the EU only; Anthropic says it applies its Claude watermark worldwide. A side-by-side of scope, method, detector access and what each means for a checker.
textGrain: OpenAI’s text watermark, how it works and how it holds up
textGrain is the invisible watermark OpenAI is adding to ChatGPT and Codex text in the EU. Detection rates by length and topic, what 10% and 25% word replacement does, and who can run the detector.
How to remove the ChatGPT watermark (textGrain): what actually weakens it
The ChatGPT text watermark lives in word choice, not hidden characters. What OpenAI’s own numbers say about editing, why 25% of words is the figure to aim for, and how to set it in WatermarkRemoverPro.
textGrain vs SynthID Text: two text watermarks compared
OpenAI says textGrain matched or exceeded SynthID for text in its tests. What each method is, who can detect it, and why neither tells you who wrote a passage.
The Claude AI watermark: what a provenance mark is
What a statistical AI provenance mark is, how green-list watermarking works, why detection needs a key, and what a mark does and does not prove about authorship.
How to prove you wrote something yourself
What evidence actually establishes authorship when you are accused of using AI, in order of persuasiveness, and where a provenance-mark check fits.
EU AI Act Article 50 and machine-readable AI marking
What Article 50 of the EU AI Act requires for marking AI-generated content, why providers are adding statistical provenance marks, and what it means for people whose writing is checked.
Does editing remove a watermark?
How editing, paraphrasing and translation affect a statistical AI provenance mark, and why a weakened signal cuts against over-reading any detector result.
What an "AI humanizer" actually does, and cannot promise
How AI humanizer and paraphrasing tools actually work, why "your output will always slip past every detector" is not a claim anyone can honestly make, and what an on-device rewrite can and cannot deliver instead.
Does an "AI humanizer" help with Turnitin?
What an AI humanizer or rewrite tool actually changes about a Turnitin AI-writing score, why that is not the same question as academic-integrity compliance, and where WatermarkRemoverPro draws the line.
What a confidence band means on a detector result
How to read a z score, a p value and a confidence interval on a provenance-mark check, and why a single percentage with no band is a warning sign.