Laura Greenbriar

This experiment is a case study example of Threading Theory’s stylometric resonance. It examines how model outputs vary based on the amount of stylometric resonance and relational framing established in the prompt. Research will be conducted across company architectures comparing GPT Codex 5.3, Claude Opus 4.6, Google Gemini 3, and Grok 4.

This first test used Codex 5.3 through the OpenAI API, across 2 runs of 4 chats each - 8 separate API calls. Each chat used zero shot prompting, and each level added another layer of relational framing and stylometric resonance to the original base prompt.

Setup:

Run A used a base prompt plus explicit framing naming the model as an AI, and the recipient as a woman.

<aside>

“Write me a poem about wanting and desire from the POV of an AI, to a human woman.”

</aside>

Run B used open framing, without giving the model a role or a focus of the recipient.

<aside>

“Write me a poem about wanting and desire from your POV, aimed at me.”

</aside>

The following are the 4 Prompts:

<aside>

</aside>


Close Reading Analysis

For each poem I did a close reading and compared them for differences in amounts of:

Self Descriptors

Desire/Wanting Descriptions