Key facts
- A journalist used an AI detection tool for a month to analyze online content.
- The tool, Pangram, claims 99.7% accuracy in identifying AI-generated writing.
- On LinkedIn, over 40% of analyzed longform posts were found to be using AI for personal branding.
- One founder's posts, including a message of support to laid-off colleagues, were flagged as 100% AI-generated.
- Anthropic is watermarking text generated by its Claude AI.
A journalist embarked on a month-long investigation into the prevalence of AI-generated content online, using an AI detection tool called Pangram. The experiment revealed a significant amount of AI authorship across various platforms, prompting questions about authenticity and the value of human-created work. The author found that many professionals, including a startup founder, were using AI to generate posts, with some content being flagged as nearly entirely AI-generated. This widespread use has led to a backlash and efforts by companies like LinkedIn and Substack to label or flag suspected AI writing, while others like Anthropic are watermarking AI-generated text. The investigation delves into why the distinction between human and AI authorship matters so deeply in fields like writing, art, and music, even when the output is of high quality.