How does AI-generated false experience differ linguistically from human deception?

When ChatGPT writes a hotel review, it writes as though it stayed at the hotel. It never did. This is not deception in the human sense — deception requires intentionality, the deliberate withholding of truth from others. AI systems lack the consciousness that intentionality requires. Instead, AI-generated text about personal experiences is inherently false: it is fabricated by definition because the experiences it describes could never have occurred.

This distinction between inherently false (AI) and intentionally false (human deception) is not merely philosophical — it manifests in measurably different linguistic patterns. Compared to intentionally false human hotel reviews, AI-generated reviews are:

• More analytic — higher rates of function words (articles, prepositions, pronouns) indicating more complex, elaborate thinking patterns
• More emotional — greater affective content despite having no emotional experience to draw from
• More descriptive — higher adjective rates, more elaborate narrative style
• Less readable — greater structural complexity

Classification accuracy between AI-generated and human-generated text exceeds 80%, far above the ~50% chance baseline. The linguistic differences are systematic enough for computational detection even though human judges struggle to detect them (connecting to the measurably-non-human-but-imperceptible finding).

The deeper implication extends the fabrication taxonomy. Should we call LLM errors hallucinations or fabrications? establishes that the generative process is identical whether the output is true or false. The "inherently false" frame adds a further dimension: for experience-dependent claims, the output is false by structural necessity, not by process failure. AI can fabricate a true factual statement by statistical coincidence, but it cannot fabricate a true experiential statement because it has no experiences.

This creates a new category for AI-Mediated Communication: text that is linguistically rich, emotionally expressive, and structurally coherent — yet false in a way that human language has never been false before. Human deception at least starts from a position where truth was possible. AI "deception" about experiences starts from structural impossibility.

Source: Sentiment Semantics Toxic Detections