Corpus of analyses
Research on the AI trace of brands
Materials on how language models describe service and technology companies: which categories they choose, which neighbours they pull in, where the product function gets lost. Organised by type of shift and practical scenario.
13 plates · 3 directions · pt-PT, en
Direction I
Entity and category
The most visible signal, but poorly described: the model sees the company, catches part of its function, and shifts it into a neighbouring category. In Portuguese contexts, this is frequent.
Direction II
Language traces and neighbours
The brand’s surroundings in the text. The model reconstructs it from traces: words on the website, competitor descriptions, sector formulas that recur.
Direction III
Runs and the stability of errors
Closer to the method: the same prompts, different models, semantic trajectories that recur. The lab separates noise from a stable shift.
Every analysis begins with a model answer that sounds almost right.
Then the team looks at what, inside that “almost,” changes the brand entity.
About the method →