Article and blog-post detector.

About the Article / blog detector

Articles and blog posts are the largest single category of AI-generated text on the web. A lot of it is benign, first drafts a human rewrote. Some of it is adversarial. SEO spam at scale. And a meaningful chunk is in the middle: human-directed outlines with AI drafting, lightly edited. This detector is tuned for that middle case, which is the one most readers actually care about.

What this detector watches for

• SEO register. The "you might be wondering," "let's dive in," "the bottom line is" register that GPT defaults to for article-style prompts. Weighted appropriately instead of being lumped with all formal prose.
• Header patterns. AI-generated articles over-use H2/H3 structure and question-form subheadings ("What is X?" / "Why does X matter?" / "How can you Y?"). Pattern-matched as a secondary signal.
• List density. Bulleted lists at a higher density than typical editorial prose. Surfaced as a factor, not as a verdict, some outlets legitimately use list-heavy formats.
• Image-free prose. Articles that flow without any image cues or callouts are more common in AI output than in human editorial (because models don't insert image placeholders by default). Minor signal; weighted lightly.

Known failure modes

• SEO-trained human writers. Writers who've been writing for SEO for years produce prose that scores higher than average on this detector. The signal is the writing style, not a cheat.
• Short posts. Under 300 words, the detector has less signal and reports lower confidence.
• News and reporting. News stories have a specific structure (lede, nut graph, supporting quotes) that's distinctive enough to classify reliably. Opinion and analysis writing is harder.

Use cases

• Content buyers verifying that freelance articles are human-written (pair with a draft-history check from the writer).
• Editors screening submissions before investing review time.
• Researchers assessing the AI-prevalence of a body of editorial work.

For editorial use we recommend publishing your AI-use policy on a visible page on the site, /for-journals has a template. Reader trust is a product of clarity, not detector-gating.