The Myth of 100% Human or 100% AI Writing

Roughly three-quarters of new web pages published by businesses now contain AI-generated text in some form, according to an Ahrefs analysis of 900,000 pages. Content entirely written by AI with zero human editing, though, makes up only 1 to 2.5% of that same pool. The overwhelming majority of what gets counted as "AI content" is neither purely human nor purely AI — it's a blend, and that blend is where almost all of it actually lives.

Why the Binary Question Stopped Being Useful

Asking whether a piece of writing is "human-written" or "AI-written" assumes a clean line that the actual production process rarely respects — a draft outlined by AI, written by a person, fact-checked by AI, and polished by a person again doesn't sort cleanly into either category. Content industry analysts have started describing this directly: most 2026 writing is genuinely hybrid, produced through a workflow with AI touching multiple stages rather than one tool generating a finished piece start to finish.

The Real Shift Is in Editing, Not Drafting

AI adoption for first-draft generation has largely plateaued near saturation, but AI use specifically for editing and revision jumped from 19% of content marketers in 2025 to 38% in 2026 — a genuine doubling in a single year. That shift matters because it changes where in the process AI involvement actually happens: not just generating a rough draft, but reviewing and refining text a person already wrote, which is a meaningfully different role than "wrote this from scratch."

The question "did AI write this" increasingly has no clean answer, because the writing process itself doesn't have a single moment where authorship transfers from one party to another. A sentence outlined by AI, typed by a human, and grammar-checked by AI again isn't 33% each — it's a genuinely collaborative object that predates any framework built to categorize it as one or the other.

What This Means for How Content Gets Judged

An Ahrefs study across 600,000 pages found a near-zero correlation — just 0.011 — between the percentage of AI-generated text on a page and how that page ranked in search, suggesting search engines aren't penalizing content based on how much AI touched it. What does correlate with weaker performance is thin, unedited, unoriginal content — which pure AI output is more likely to be, but which plenty of purely human writing is guilty of too. The origin of the text matters far less than whether it says something worth reading.

Where the "Pure" Categories Still Exist

Fully unedited AI output — the 1-2.5% end of the spectrum — tends to concentrate in specific, low-stakes, high-volume use cases: internal documentation, template-driven listings, or content where speed genuinely outweighs originality. Fully human writing with zero AI involvement, at the other end, has become genuinely rare in professional content production specifically, even when the AI's role was limited to spell-checking or light editing rather than drafting.

A More Useful Question Than "How Much AI"

Rather than asking what percentage of a piece was AI-generated, the more useful question is whether it contains something a reader couldn't get from ten other pages covering the same topic — a specific example, a tested claim, an actual point of view. That standard applies identically regardless of whether AI touched the outline, the draft, the editing pass, or none of it at all, which is part of why the percentage-of-AI framing has started to feel like the wrong axis to measure by.

---

The 74% figure and the 1-2.5% figure aren't in tension — they're describing the same reality from two angles. AI has become part of how most writing gets made, and almost none of that writing skips the human part of the process entirely. The binary of "human-written" versus "AI-written" made sense as a category in 2023; by 2026, it describes a small minority of actual content on either extreme, while nearly everything else sits somewhere in the collaborative middle that binary was never built to capture.