AI detection accuracy faces scrutiny from researchers and regulators

Research and regulatory actions highlight the growing difficulty in distinguishing between human and AI-generated text. A Pew Research Center analysis of nearly half a million English-language pages found that approximately one in ten showed significant signs of AI authorship as of July 2026, with that figure rising above one third for content published after the launch of ChatGPT.

Studies indicate that human intuition is largely ineffective at identifying machine-written content. In a PNAS study of 4,600 participants, accuracy in identifying human versus AI writing for job and dating contexts sat between 50 and 52 percent, performing near the level of a coin flip. Readers often rely on unreliable cues, such as first-person pronouns and contractions, which AI models can easily replicate.

In the United States, the Federal Trade Commission (FTC) has taken action against companies making unsubstantiated claims regarding AI capabilities. Workado was required to enter a consent order after promoting an AI Content Detector with a 98% accuracy rate, while independent testing found its actual accuracy on general-purpose content was only 53%. The FTC's action emphasizes that marketing claims regarding AI efficacy are treated as factual representations of product performance and must be supported by competent evidence.