People Remain “Blissfully Ignorant” Of AI Use in Everyday Messages, New Research Shows

Why It Matters

The research reveals a hidden advantage for undisclosed AI use in everyday communication, creating an uneven playing field for those who avoid it. It also signals that mandatory AI‑disclosure could damage trust and brand reputation.

Key Takeaways

• Disclosure of AI use harms sender's perceived sincerity
• Without disclosure, recipients assume human authorship and rate positively
• Heavy AI users are not more likely to suspect others
• Perceived effort and authenticity drive the AI penalty
• Findings replicated across two U.S. samples of ~650 participants

Pulse Analysis

The proliferation of generative AI tools such as ChatGPT, Claude, and Gemini has reshaped everyday written communication. Professionals now rely on these systems to draft emails, social media posts, and even personal texts, promising speed and polish. Yet the convenience raises a subtle trust issue: does the audience notice when a message is machine‑crafted, and how does that affect the sender’s credibility? Academic literature has long shown a “AI penalty” when users are told a text was generated by a bot, but real‑world behavior outside the classroom remained unclear.

Zhu and Molnar’s two‑wave experiments, each with roughly 650 U.S. adults, provide the first large‑scale evidence of how ordinary recipients react. When participants were explicitly told a message came from an AI, they rated the author lower on friendliness, sincerity and trustworthiness, and used more negative descriptors. In contrast, when no source information was given—or when the possibility of AI use was merely hinted—impressions matched those for human‑written messages. Notably, even frequent AI users did not become more suspicious of others, indicating that familiarity does not breed vigilance.

The findings have immediate relevance for businesses that depend on persuasive writing, from sales outreach to internal communications. Companies can leverage AI to enhance clarity and efficiency without risking reputational damage, provided they keep usage undisclosed. However, the latent “AI penalty” suggests that mandatory labeling policies could backfire, eroding trust whenever AI assistance is revealed. Future research should explore high‑stakes contexts—hiring, legal, academic grading—and cross‑cultural attitudes, helping organizations craft guidelines that balance productivity gains with ethical transparency.