AI Made Content Almost Free to Produce. It Made Differentiation Almost Impossible.

In November 2024, for the first time, machines wrote more of the internet’s new articles than people did. By May 2025, AI-generated articles made up 51.7% of everything published online, according to a Graphite study that classified 65,000 English-language pages pulled from Common Crawl. A separate Ahrefs analysis of 900,000 new web pages found that 74.2% of them carried AI-generated text that April. Barely a quarter were written by a human alone.

The production problem is solved. That is the problem.

For the whole content era, the binding constraint was supply. Publishing something competent, on deadline, at volume, took writers, editors, and budget. AI removed that constraint in about a year. Within roughly twelve months of ChatGPT’s launch, AI was already drafting close to 40% of new articles, by Graphite’s count. The cost of a competent blog post fell near zero, and output rose to match. What did not rise was the number of ways to be competent. Feed the same models the same training data and the same prompts, and they hand back the same middle.

Kaleigh Moore, a freelance writer who has covered SaaS for outlets including Forbes and Vogue Business, describes the result as a B2B content monoculture: long-form pieces that share the same structure and tone, heavy on definitions, comparison tables, and stat-stuffed paragraphs built for machines to extract. When every company in a category tunes for the same answer engines with the same tools, the category’s writing converges on one voice. Nobody chose it. The tools did.

Here is the part that stings. All this content is chasing a shrinking payoff. Google now answers many queries in place with AI Overviews, and people click less when it does. Pew Research tracked 68,879 real searches and found users clicked a result 8% of the time when an AI summary appeared, against 15% when it did not. Digital Content Next surveyed 19 major publishers and found median Google referral traffic down about 10% YoY, with some sites off as much as 25%. More competent content, fewer clicks to go around.

The economics are simple once you name them. When a good-enough article costs almost nothing, everyone makes more of them, and the supply of good-enough floods every channel at once. Scarcity was doing quiet work the whole time. It meant a decent explainer was a small moat, because your competitor had not gotten around to writing one. That moat is gone. Your competitor generated the same explainer this morning, for the same near-zero cost, tuned for the same answer box.

Which is why the surviving advantage looks nothing like output. Graphite’s other finding is the tell: even with AI writing half of all new pages, 86% of the pages that actually rank in Google are still human-written. Volume lost. The pages that win carry something a model cannot assemble from everyone else’s text.

For a marketing team, that resets the job. Competent content is now the cost of entry, the way a working website or a mobile layout stopped being a differentiator a decade ago and became the floor. It gets you into consideration. It does not pull you out of the pack, because the pack is producing the identical thing at scale. The advantages left are the ones a language model cannot synthesize from public text: a real point of view, proprietary data, a named person willing to be specific and occasionally wrong, and a brand people recognize before they read a word. When search itself is collapsing into AI-generated answers, being the source those answers cite beats being one more page they summarize.

Three moves follow for any team still funding content the old way.

Stop paying for volume. A model beats any human on that axis for free, so a content plan measured in posts per month is a plan to lose slowly. Spend the hours AI hands back on the work it cannot fake: original research, real customer numbers, an argument with a named author behind it. And before publishing anything, run one test. Could a competitor have produced this exact piece without knowing a single thing your company knows? If the answer is yes, the piece is competent.

Competent is the floor now. Floors are for standing on, not for winning.