Requests, topics, and observations for analysis

The lab studies how language models assemble an image of a company from textual traces. The focus is on cases where the model shifts the category, replaces the function, pulls in a neighbouring brand, or leaves an empty space instead of an accurate description.

Requests for advertising campaigns, instant AI visibility, and a brand diagnosis from a single answer remain outside its work. The lab works from observations, repeatable runs, and careful descriptions of uncertainty.

Each analysis is best read as a map of model behaviour under defined conditions. If a text says the model links a company to a neighbouring category, it means exactly that this pattern appeared in a series of runs; the finding does not become a final verdict on the market or the brand.

Yes, if the topic has a clear object of observation: a company, category, type of prompt, or recurring error in AI answers. Cases are especially useful when the answer sounds convincing, but a person from the industry can see a shift inside it.

Individual requests are possible, but not every one becomes research. The team first looks at whether a reproducible series of observations can be collected and whether the analysis will be useful beyond one private case.

The lab replies after an initial review of the message. If a request requires additional reading or checking of formulations, the response may take several working days, because the team does not comment on AI behaviour from a single screenshot.

The lab’s main working context is connected to Portugal and the Portuguese language, but the site materials are available in English in this version. Portuguese and English prompt formulations may appear in analyses when they matter for the observation.