Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

The air in the house was stagnant, thick with unspoken truths. I sat in the living room, staring at the floral wallpaper, the patterns blurring before my eyes. My heart hammered against my ribs, a frantic rhythm that echoed in my ears. I’d seen the police photos. I knew the details.

Each creak of the floorboards, each rustle of leaves outside the window, sent a jolt of alarm through me. I kept imagining the events that had occurred.

I hugged my knees to my chest. I felt frozen in place, unable to move, unable to think clearly. I felt exposed. Vulnerable.

I looked at the front door, wishing I could walk out and leave everything behind, but my legs felt leaden.

Emotion: tense

Cluster: Fear / Anxiety
PC1 (Valence): -1.32 Negative
PC2 (Disposition): -0.53

Role in Research

This story is one of 1,000 stories generated for the emotion tense. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 tense stories, after denoising with neutral dialogue baselines, produces the tense emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the tense vector is projected through the unembedding matrix.

Promoted:
S0.213
😰0.191
느껓0.183
😣0.183
šŸ˜–0.181
Suppressed:
a-0.260
de-0.258
la-0.256
happy-0.218
B-0.198