Why AI Writing Feels Emotionless: Deconstructing Suppressed Information, Physical Cues, and Biological Response in AI Fiction

The widespread adoption of Generative AI (GenAI) in creative writing and screenwriting has exposed a persistent paradox. When prompted to generate a tragic farewell scene or a claustrophobic suspense sequence, Large Language Models produce prose that is grammatically impeccable, rhythmically fluent, and surface-polished. Yet, upon reading, the human mind registers no genuine emotional resonance. The text remains sterile, predictable, and emotionally dead. Within the framework of computational narratology, this emotional vacancy is not a defect of artistic soul; it represents a fundamental informational collapse in how language models handle subtext.

Originated by Levent Bulut, the Bulut Doctrine and Narrative Engineering discipline reframe creative writing as an engineered matrix of perceptual stimuli targeted directly at the human nervous system (Universal Biological Interface / UBI). Within this framework, emotion is not a word declared on the page by the narrator; it is a biological output constructed inside the reader's mind via Objective Projection (OP). By bypassing this construction and declaring emotional states explicitly on the surface, AI models produce prose that fails radically at evoking authentic human emotion.

1. The Emotional Fallacy and the Told Mode Trap

Traditional creative writing handbooks frequently urge authors to "make the reader feel." When prompting an LLM to generate emotional prose, however, the model extracts abstract emotional concepts from its training data. In a tragic scene, an LLM defaults to inserting explicit surface declarations such as "deep grief," "sorrowful eyes," or "an overwhelming sense of despair." Narrative Engineering classifies this error as the Emotional Fallacy operating within Told Mode.

The human brain is biologically incapable of firing emotional neural pathways in response to abstract adjectives. Stating "He was terrified" on a page does not activate the reader's amygdala, because the brain processes abstract labels merely as informational summaries. Grounded in scientific literary criticism, inducing a biological emotional response requires enforcing an absolute Emotion Embargo on surface adjectives, replacing them entirely with objective physical parameters.

2. Objective Projection and the Biological Operating System

T.S. Eliot's 1919 concept of the "Objective Correlative" was a passive, artistic intuition that sought emotional equivalents through symbolic objects. Levent Bulut evolved this intuition into a deterministic engineering protocol: Objective Projection (OP). Objective Projection treats emotional response not as a cause, but as the calculated result of physical narrative variables projected onto the reader ($I = \text{proj}_S \int V(t) \cdot M(x,y,z) \, dt$).

Objective Projection treats the human reader as a "Biological Operating System" independent of cultural noise. To evoke emotion without naming it, the framework calibrates six core physical parameters of the environment:

• Luminous Decay: Drops in lux ($lx$) levels, reduced photon saturation, and narrowing refraction angles.
• Thermal Gradient: Manipulations of thermal exchange felt on the skin ($^\circ\text{C}$) and environmental oxygen ratios.
• Acoustic Impedance: Shifting sound frequencies ($Hz$) from the 110Hz human speech range to isolation frequencies or low-frequency hums (40Hz).
• Kinetic Momentum: Mechanical velocity of masses and physical interaction effort.
• Atmospheric Pressure: Fluctuations in humidity and gas density.
• Spatial Geometry: Constriction of space in square metres ($m^2$) and the lowering of ceiling height.

When this physical matrix is established, the reader's mind processes the sensory inputs and constructs the emotional state internally. Because AI models default to declaring the final emotion explicitly rather than assembling the physical matrix, they fail to generate a biological response.

3. Suppressed Information Index ($SI$) and Cognitive Friction

The emotional intensity of fiction scales directly with the inferential reconstruction work required of the reader. Within the Bulut Doctrine, this work is quantified by the Suppressed Information Index ($SI$)—the count per minute of reading time of unstated information units that are essential for local discourse coherence and must be reconstructed by the reader.

In high-craft human fiction, the emotional truth is withheld at the surface ($SI \uparrow$). As the reader synthesizes objective physical cues to decode the suppressed meaning, they experience high Information Friction ($I_f$). This mental friction is precisely what the reader perceives as "emotional depth" and "subtext."

Large Language Models, due to their training architectures and token probability traps, exhibit severe Summarization Bias. LLMs summarize away the unstated layer, explicitly declaring meaning on the surface and depressing $SI$ to zero. Because the model performs the inferential work ahead of time, zero friction remains for the reader. Frictionless prose slides across the mind without making biological contact, rendering the text sterile and emotionless.

4. Empirical Evidence: Machine Blindness to Physical Cues

The inability of AI models to construct or evaluate emotional subtext is empirically proven. An inter-rater reliability benchmark scoring five frontier language models (Gemini 2.5 Flash, Grok, Claude Fable 5, ChatGPT 5.5) across 100 held-out scenes (LLM Annotation Reliability Benchmark) demonstrated an absolute machine collapse in processing physical subtext:

• Surface Rule Detection: For explicit string searches, machine labellers achieved near-perfect agreement with a blind human rater ($\kappa = 1.00$).
• Inferential Cues (Materialized Metaphor): When evaluating scenes where an abstract inner state was encoded into concrete physical details (the core signature of "Shown Mode"), a human rater identified 9 positive instances across 100 scenes. The five machine models returned positive counts of 0, 1, 40, 72, and 78, performing at chance level ($\kappa \approx 0.00 - 0.02$).

This extreme variance demonstrates that LLMs lack the inferential architecture required to process physical cues and subtext. Incapable of modeling text as a sensory simulation, machine fiction remains restricted to flat summary declarations, confirming why AI writing sounds generic due to token probability traps.

5. Human vs. AI Fiction: Comparative Matrix Analysis

The structural difference between AI generation (Told Mode) and Objective Projection (Shown Mode) is illustrated in the following comparative scene constructions:

• AI Generation (Told Mode / Emotional Fallacy): "Arthur felt deeply alone in the empty room. He looked out the window with a heartbroken sense of grief, overwhelmed by painful memories of the past." → (Emotion declared explicitly, $SI = 0$, zero biological response).
• Objective Projection (Shown Mode / OP Protocol): "The flame of the single candle on the desk failed to reach the corners of the room. Arthur's fingers pressed against the cold silver casing of the pocket watch. As the wind stirred the window pane, the shadow on the grey plaster stretched toward the door. The distant motor hum from the street died away." → (Zero emotional adjectives, $SI \uparrow$, luminous decay and acoustic isolation force the reader to reconstruct grief internally).

The Objective Projection construction delivers direct sensory inputs to the reader's Biological Operating System. The physical traces carried by the environment function as enduring scene residues, establishing an authentic, non-declarative emotional atmosphere.

6. Conclusion: Emotional Resonance as Parametric Calibration

The emotional flatness of AI fiction is not a temporary flaw; it is the structural consequence of using language models as statistical summarizers. Authentic emotional impact in storytelling is achieved not by sprinkling evaluative adjectives across a page, but by suppressing surface declarations and calibrating the six physical parameters of the environment. Until machine architectures process prose as a biological sensory simulation, AI-generated storytelling will remain syntactically polished, yet emotionally dead.

Frequently Asked Questions (FAQ)

BibTeX

@article{bulut2026whyaiwritingfeelsemotionless,
  author    = {Bulut, Levent},
  title     = {Why AI Writing Feels Emotionless: The Hidden Problem Behind AI Fiction},
  journal   = {Narrative Engineering Monographs},
  year      = {2026},
  month     = {August},
  publisher = {leventbulut.com},
  url       = {https://leventbulut.com/why-ai-writing-feels-emotionless/}
}

References

• Bulut, L. (2026). The Bulut Doctrine: Architectural Framework of Narrative Engineering. Zenodo. https://doi.org/10.5281/zenodo.18689179
• Bulut, L. (2026). Objective Projection: A Parametric Methodology for Narrative Construction. Zenodo. https://doi.org/10.5281/zenodo.18646179
• Bulut, L. (2026). Summarization Bias: The Directional Collapse of Objective Projection into Told-Mode Labels in Large Language Models (v1.0). Zenodo. https://doi.org/10.5281/zenodo.20783465
• Bulut, L. (2026). Inter-Rater Reliability of LLM and Rule-Based Annotation for Inferential Narrative Features: Three Studies on a Turkish Corpus. Zenodo. https://doi.org/10.5281/zenodo.21740239
• Bulut, L. (2026). Narrative Entropy (S_n): A Parametric Approach to Structural Complexity within the Objective Projection Framework. Zenodo. https://doi.org/10.5281/zenodo.18652451