NYT Review: Pangram AI Detector Flags 'Slop' but Struggles With Images

The New York Times recently reviewed Pangram, an AI detection tool designed to identify "slop"—the low-quality, generic content currently flooding the internet. The review suggests that while the software provides a sense of empowerment for users, its effectiveness varies significantly depending on the medium being analyzed.

In a personal tech review titled "I Tested a Popular A.I. Slop Detector. It Felt Empowering," the author found that Pangram excels at distinguishing chatbot-generated text from human writing. However, the testing revealed a clear performance gap, noting that the tool is less effective at identifying AI-generated images than it is at detecting synthetic text. The software is designed to flag content from several major large language models, including ChatGPT, Claude, Gemini, and Llama.

The Rise of AI Slop

This push for detection comes as the internet faces a surge of "AI slop": formulaic, depthless text appearing in product reviews, social media feeds, and self-published books. As large language models (LLMs) become more accessible, this proliferation of generic content has created a digital trust crisis. The challenge is compounded by the fact that some human writing, particularly on platforms like LinkedIn, has become so cliché-driven and formulaic that it stylistically converges with AI-generated output.

Market Momentum and Accuracy

The demand for these tools is reflected in Pangram's recent growth. The startup recently raised $9 million in funding led by Menlo Ventures and has since released version 4.0 of its text detection model. Regarding performance, Pangram's own website claims an accuracy rate of 99.98%, while other industry reports describe the model as being over 99% accurate.

The Shift Toward Active Verification

The ability to detect synthetic content is becoming a critical component of modern digital literacy. As AI-generated text becomes nearly indistinguishable from human writing in formulaic contexts, users are being forced to move from passive consumption to active verification. This psychological shift represents a new era of internet navigation where the burden of proof is increasingly placed on the content itself.

Future Outlook

While Pangram's text detection shows promise, the struggle to accurately identify AI-generated images highlights a continuing vulnerability in the fight against synthetic media. As generative models for both text and imagery evolve, the industry will likely watch whether detection tools can keep pace with the increasing sophistication of the models they are designed to sniff out.