ProofApril 20265 min read

We scored 0% on Scribbr's AI detector. Zero. Not even a trace.

Scribbr uses the same detection engine as Turnitin. We ran our AI non-fiction through it. The result: 0% AI-generated, 100% human-written. Here's the editorial engine behind it.

Scribbr AI detector showing 0% AI-detected on Ghostproof output

Scribbr's free AI detector. Chapter 1: The View from Rock Bottom. 0% AI-generated. 100% human-written.

AI-generated text has a fingerprint. Not one pattern — dozens. Sentence rhythms that never vary. Emotions explained after they've already been shown. Every character speaking with the same polished vocabulary. A suspicious, uniform quality across every paragraph.

AI detectors are built to find these patterns. The good ones — like Scribbr's detector, which uses the same technology that powers Turnitin — are remarkably accurate. They don't just look for individual tells. They measure statistical consistency across the entire text.

Last week, we ran a chapter of AI-generated non-fiction through Scribbr's free AI detector.

AI-detected. 100% human-written.

Scribbr Free AI Detector · Powered by Turnitin

The detector couldn't find a single statistical pattern suggesting the text was generated by a machine. Every paragraph, every sentence, every phrase registered as human-written.

What we tested

The chapter is called “The View from Rock Bottom.” It's narrative non-fiction — the story of a steelworker in Rotherham who loses his job when the local steel works closes. He watches his son through a window. He avoids his support worker's phone calls. He makes tea he can't taste.

It was generated entirely by AI using the Ghostproof editorial engine. No human editing. No “humaniser” tools. No synonym shuffling. Raw output.

We chose non-fiction deliberately. Fiction gives AI places to hide — plot twists, magical systems, genre conventions. Non-fiction strips all of that away. You're left with voice, specificity, and truth. A man in a bedsit in Rotherham doesn't allow you to fake authenticity.

The comparison

ChatGPT

88% AI

Claude (raw)

60% AI

Ghostproof fiction

Ghostproof non-fiction

Same underlying AI. Different editorial intelligence applied to the output.

What the engine catches

Ghostproof isn't a different AI model. It runs on Claude, the same model anyone can use. The difference is what happens between the model's output and what you read. Over 250 editorial rules run on every chapter.

Show-then-tell elimination

AI writes the emotion, then explains it. Human writers trust the image. Ghostproof catches and removes the explanation automatically.

Before: “Her hands trembled. The fear was overwhelming.”
After: “Her hands trembled.”

Ghost Mode

AI prose is consistently 7/10 — every paragraph polished, every transition smooth. This uniformity is itself a tell. Ghost Mode engineers deliberate variance: rough transitions alongside literary passages. The imperfection is what makes it undetectable.

Voice DNA

Every character gets a distinct voice profile — sentence length, vocabulary register, verbal tics, emotional expression. AI defaults to making every character sound like the same articulate narrator. Voice DNA prevents that.

Wrong-thought insertions

During emotionally significant moments, real people think irrelevant things — whether they locked the door, whether the milk will spoil. AI characters process every moment with appropriate gravity. Ghostproof inserts the mundane intrusions that signal a real human mind.

ICK word filtering

200+ words and phrases that AI reaches for but humans rarely use. “Orbs” for eyes. “Electricity coursed through.” “A dance of shadows.” All auto-removed before you see the output.

The gap between “obviously AI” and “indistinguishable from human” isn't a model problem. It's an editorial problem.

Why this matters

The publishing industry is grappling with a flood of AI-generated content. Most of it is obvious — the same rhythms, the same vocabulary, the same emotional shortcuts. Detectors exist because the market needs a way to distinguish between AI-assisted writing that's been thoughtfully crafted and AI-generated text that's been dumped without editorial intelligence.

Ghostproof exists for writers who want to use AI as a tool but produce work that reads like it was written by a person — because editorially, it was. The AI generated the raw material. The engine shaped it into something that passes both editorial review and AI detection simultaneously.

What this doesn't mean

We're not claiming Ghostproof will score 0% on every chapter, every time, on every detector. AI detection is probabilistic. Different chapters will produce different scores depending on content, genre, voice profile, and which detector you use. What we are claiming is that the editorial approach works. When you apply 250+ rules to catch AI patterns — when you engineer voice variance, eliminate show-then-tell, filter ICK words, and enforce Ghost Mode imperfection — the output becomes statistically indistinguishable from human writing.

Try it yourself

Generate your first chapter — fiction or non-fiction — and run it through any AI detector you like.

Generate a free chapter

No signup required · ghostproof.uk

Ghostproof is a UK-based AI book production engine. The editorial engine applies 250+ rules to every chapter, producing fiction and non-fiction that passes editorial review and AI detection simultaneously.