Pangram
Coverage of Pangram in the Nexus archive.
- LinkedIn adds a ‘seems like AI slop’ button after blocking billions of automated comment attempts in the last few months
LinkedIn introduced a 'Seems like AI slop' button to let users flag low-quality AI-generated content and has blocked billions of automated comment attempts. The platform is replacing its 'enhance your post' tool with a proofreading feature and enhancing bot detection. A study found 40% of long-form and 30% of short-form LinkedIn posts are AI-generated, prompting similar actions by Snapchat and Meta.
- Publishers are losing Google traffic as AI answers replace links
Publishers are experiencing significant declines in Google Search traffic as AI-generated answers replace direct links to websites. Small publishers lost 60% of referrals over two years, while Substack partners with AI-detection software Pangram to combat AI-generated content on platforms like LinkedIn. The article highlights the shift toward optimizing content for AI language models (GEO) and adapting to a post-open-web distribution model.
- It’s Not Just Annoying, It’s Inescapable
The article discusses the prevalence of the rhetorical device 'It’s not X; it’s Y' in AI-generated writing, citing examples from Shakespeare, corporate communications, and fiction. It highlights measurable increases in such constructions in AI text and notes their persistence despite efforts to refine AI models.
- It's an AI web, and we're just rats in the walls
The article discusses the growing dominance of AI in web traffic and content creation, citing data from Cloudflare, Imperva, and Pangram showing bots account for over half of web traffic and a significant portion of content on platforms like LinkedIn and Twitter. It raises concerns about AI's impact on human creators and online authenticity.
- AI slop writing has taken over the internet, particularly LinkedIn and X
A study by AI detection platform Pangram found that 25% of longform social media posts (over 250 words) across LinkedIn, X, Medium, Substack, and Reddit are fully AI-generated. LinkedIn leads with 41% AI-written longform content, while X has 25% fully AI-authored posts and 23.2% with AI assistance.
- Why you should think twice about using AI to write your LinkedIn posts
Steven Bartlett's company reduced AI use on LinkedIn after noticing an increase in 'AI slop.' Data from Pangram estimates 41% of long-form and 30% of short-form LinkedIn posts are AI-generated. FlightStory now uses human-written content, which improved post performance despite occasional errors.
- How courts are coping with a flood of AI-generated lawsuits
Courts in the US are experiencing a surge in AI-assisted lawsuits filed by self-represented individuals, with a study showing a rise from 11% to 16.8% of federal civil cases between 2022 and 2025. Judge Maritza Braswell notes AI improves pleading clarity but introduces challenges like hallucinations and fabricated quotes, while lawmakers debate accountability for AI-generated legal advice.
- America Has a Pangram Problem
Pangram, an AI detection tool, has been central to recent controversies over AI-generated writing, including accusations against a horror novel, newspaper articles, and a Pope's encyclical. While Pangram claims high accuracy, concerns remain about its error rates and potential for misuse in academic and professional settings.
- Did the Pope use AI to write about the dangers of AI?
An analysis suggests parts of Pope Leo XIV's encyclical 'Magnifica Humanitas' may be AI-generated, with 40-100% of certain paragraphs flagged by AI detector Pangram. The document shows traits like high use of 'genuinely,' similar to Anthropic's Claude, and 62% of the first chapter was identified as AI-written.
- AI writing hits a ceiling
The share of online news articles and blog posts that are primarily AI-generated has held near 50% for over a year, indicating that AI content hasn't overtaken the web. AI-generated articles surged after ChatGPT launched in November 2022 but growth has stalled. The quality of AI content is rapidly improving, making it hard to distinguish from human-written content.