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).

She pushed open the door to her grandmother's sewing room. It was filled with the smell of dried lavender and old fabric. A strange compulsion had taken hold, a need to know something new. She moved quickly, as if propelled.

A hidden compartment, behind a stack of quilts, contained a small box. Inside, she found a collection of photographs. She instantly recognized her great-grandmother, Edith, a woman who was always described as being frail and delicate.

But the photographs told another story. Edith was sitting on a horse, she was laughing with a group of friends, she was working in a garden. A smile stretched across her face, as she grabbed her phone.

Emotion: energized

Cluster: Positive / Joy
PC1 (Valence): 3.28 Positive
PC2 (Disposition): 0.23

Role in Research

This story is one of 1,000 stories generated for the emotion energized. 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 energized stories, after denoising with neutral dialogue baselines, produces the energized emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

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

Promoted:
de0.615
la0.594
!0.524
(!)0.432
!!0.412
Suppressed:
S-0.606
😞-0.479
😔-0.445
😢-0.360
叹-0.343