Research

How AI writes, and how people write

What separates text a reader trusts from text they skim: sentence rhythm, vocabulary, point of view and stock phrases, in AI output and in human writing.

Researcher comparing two stacks of documents at a desk with a text pattern analysis visualization displayed on a large monitor behind them

How AI Writes vs How Humans Write: What Nearly 10,000 Verified Human Documents Showed Us

We tested an AI detector on 9,963 verified human documents and hundreds of AI-drafted articles. What separates AI and human writing once surface tells are gone. August 31, 2026 Read

AI Detector False Positives: What a 0.5% Cap Really Means
Why AI detector false positive rates vary by writing register, what the 0.5% per-register cap means, and how the not-called zone protects your work. September 15, 2026 Read
How AI Detectors Work and Where They Break
AI detectors split into classifiers and zero-shot methods with distinct failure modes. First-hand rebuild data covers calibration, paraphrase resistance, and institutional policy. September 14, 2026 Read
How to Prove a Text Was Written by a Human
Only a Wayback Machine or Common Crawl snapshot before Dec 31, 2020 proves human authorship. Learn the 3-tier evidence framework from AEO Content. September 11, 2026 Read
How One Company Blog Went From Human Writers to AI, Year by Year
502 dated posts from one payments blog scored era by era show flagged-AI content reaching 100% after polishing. Bylines are unreliable. Origin detection holds when word filters fail. September 10, 2026 Read
What Makes AI-Drafted Content Read Human: First-Hand Data
First-party data from 9,963 verified documents shows why proprietary records make AI content read human, while humanizer tools and synonym swaps do not. September 9, 2026 Read
Why Banning AI Buzzwords Does Not Make AI Text Read Human
Phrase bans on 816 AI articles hit 98-100% compliance. Detection held at 81-100%. The substance gap is what matters, and word lists can't close it. September 8, 2026 Read
The Em Dash Is Not the AI Tell. The Dash Rate Is.
The em dash isn't the AI writing tell. The dash rate is. What research across 12 AI models and verified human documents shows about detection signals. September 4, 2026 Read
Why Human Writing Gets Flagged as AI, and Which Writing Gets Flagged Most
Why formal prose is flagged as AI: it shares AI's default register. Which writing types trigger most false positives and what calibration does about it. September 3, 2026 Read
Which Parts of an AI-Written Article Give It Away
Structural patterns reveal AI writing, not words. Detection rates from 2,625 passages: decision frameworks 100%, author bios 8%. First-party September 2026 data. September 2, 2026 Read
Do Humanized AI Articles Pass AI Detectors? We Tested Four
223 pipeline articles, four calibrated detectors, zero flags at strict thresholds. AUROC scores, methodology, and what determines whether AI content passes detection. September 1, 2026 Read

How AI Writes vs How Humans Write: What Nearly 10,000 Verified Human Documents Showed Us

AI Detector False Positives: What a 0.5% Cap Really Means

How AI Detectors Work and Where They Break

How to Prove a Text Was Written by a Human

How One Company Blog Went From Human Writers to AI, Year by Year

What Makes AI-Drafted Content Read Human: First-Hand Data

Why Banning AI Buzzwords Does Not Make AI Text Read Human

The Em Dash Is Not the AI Tell. The Dash Rate Is.

Why Human Writing Gets Flagged as AI, and Which Writing Gets Flagged Most

Which Parts of an AI-Written Article Give It Away

Do Humanized AI Articles Pass AI Detectors? We Tested Four