AI writing

AI writing

2026-08-21 02:32:58

Pew study finds AI traces in 35% of new web pages launched after ChatGPT

A large-scale analysis cited by TechFlowPost says AI involvement is showing up across a growing share of newly published web content. The study, released this week by Pew Research Center and based on Common Crawl archives, sampled 10,000 English-language pages from each of 49 web crawls conducted between January 2021 and July 2026, for a total of roughly 490,000 pages. Researchers then scanned the sample page by page using an AI detection model called Open Pangram. In the July 2026 snapshot, about 10% of English web pages showed what the study described as significant signs of AI writing or heavy AI-assisted editing. Among pages published after ChatGPT launched on Nov. 30, 2022, that figure rose to 35%. The report also found major differences by domain type, with about one in 10 .com pages showing AI-writing features, compared with 4.6% for .org pages and roughly 1% for both .edu and .gov. The article links those findings to traffic data from Cloudflare and HUMAN Security, which indicate that machine traffic is rising quickly as well. It argues that the web is moving toward a system where machines increasingly produce content for other machines to consume, a shift that could weigh on publishers that rely on search traffic and raise the value of verifiable human-original reporting.

640

Clay

2026-08-12 08:39:13

Clay rolls out company-wide AI writing policy that lets staff use AI but makes them answer for every line

Clay, an AI sales and go-to-market platform, has introduced a formal AI writing policy across the company, setting rules for any employee who uses AI to draft text that others will read. According to co-founder Varun Anand, the policy was first written by Sophie Alpert for the engineering team and was later adopted company-wide after other teams found it useful. The policy is built around four principles. First, employees must stand behind every idea and every sentence that appears under their name. Clay says staff cannot excuse unclear wording by saying AI generated it. Second, the company argues that writing is part of thinking: deciding what matters, how to structure an argument, and how to explain a problem clearly is itself part of understanding the problem. Third, Clay says more time should be spent authoring a document than consuming it, warning against using a short prompt to create a long document that imposes reading costs on many coworkers. Fourth, the company says longer documents are not inherently better, noting that generative AI can produce polished but low-information text at scale. Anand said that in some cases, if a long document comes from a very short prompt, it may make more sense to share the prompt itself.

2530

Dan Koe

2026-08-03 14:00:08

Dan Koe argues you do not need to remember everything you read

Dan Koe’s essay on remembering what you read makes a blunt case against a habit many knowledge workers have turned into doctrine: trying to retain every article, highlight and note. His argument is that memory is not the real bottleneck. The larger problem is that people confuse collecting information with learning, and organization with actual use. In his framework, the material that matters most is not the material you force yourself to memorize, but the material that becomes embedded through repeated application, discussion and output. The piece reframes learning through a cybernetics lens built on four parts — goal, sense, compare and act — and says most people get stuck because they keep feeding themselves information without a clear target. Koe then pushes a project-first approach: start building, create pressure through real work, and learn only what the task requires. He extends that logic to digital note systems, arguing that many “second brain” setups fail because they are designed for storage, not retrieval or creation. The article also reviews practical workflows, including a self-managed Obsidian + Claude setup and tools such as Eden and MyMind, and draws a line around AI’s role. AI, in Koe’s view, can reduce friction in research, sorting and drafting, but it should not replace a person’s own judgment, values or voice.

630

AI writing

2026-07-10 12:52:13

Pangram Labs CEO Says AI Writing Still Lacks Style as Detection Tools Reach 99% Accuracy

Max Spero said AI is strong at grammar and clarity but still weak in style and creativity. As machine-generated content spreads, detection tools with reported 99% accuracy are becoming increasingly important for protecting authenticity online.

430