What our editors see in AI-written articles, and how to fix it.
Courts and clients have so far been consistent: the business that publishes AI-assisted text answers for it. What the Air Canada chatbot case, the Wolf River Electric suit, and the Deloitte refund tell you, and what to do before you publish.
The US Copyright Office's Part 2 report settled the basic question for now. Prompting doesn't make you an author; selecting, arranging, and substantially editing AI output can. Here is what that means for a company blog, and how to document your edits.
Slop went from forum slang to dictionary word of the year in about three years. Here is what it actually describes, how to recognize it in your own drafts, and a test to apply before you publish.
Whether to label AI-assisted posts is a real decision, not a moral one. Here is what the rules and the surveys actually say, a policy a small team can follow, and three disclosure lines you can copy.
Freelance editors, per-word services, monthly subscriptions, marketplaces and a flat fee, priced for the same 1,500-word article. The numbers, with sources and dates.
Nine documented cases where AI-assisted text reached readers with invented books, fake sources, fabricated quotes, or leftover chatbot text. Each one has a pattern, and each one had a simple check that was skipped.
Telling a model to "make it better" produces a different draft, not a better one. Here is the correction-prompt format our editors use: numbered items, quoted locations, one problem each, and constraints up front.
Google's written policies, quoted and linked, on AI-generated content, scaled content abuse and the who, how and why questions, plus a checklist for AI-assisted sites.
A fictional HVAC company's AI-drafted post, the editor's nine corrections written as a prompt, and the article after the AI applied them. Every passage is invented and labeled.
Vendors report 99% accuracy. Independent studies report false-positive rates from 9% to over 60% depending on who wrote the text. Both can be true, and neither tells you whether your article is any good.
Invented case citations, dosage drift, expired tax thresholds. Why AI drafts on these topics fail in specific ways, the review protocol that catches them, and what an editor can and cannot sign off on.
A field guide to the habits that make an article read as machine-written, from "delve" to bullet avalanches, with a one-line prompt fix for each and a combined prompt you can paste.
Eight stages, fifteen minutes, one page of documentation per piece. A pre-delivery checklist for agencies that draft with AI and deliver under their own name.
We take a plausible AI-written lawn care post apart paragraph by paragraph, show the editor's note on each problem, and explain why asking the model to review itself won't find them.
A timed checklist for checking an AI draft before it goes live. Triage the claims, verify the ones that matter against primary sources, and cut or soften the rest.
Google does not ask whether a page was written by AI. It asks who stands behind it, how it was made and why it exists. Here is how to answer those questions on a page a model helped draft.