In one anonymized entry from my error notebook, a model confidently named a clinic in São Paulo, but placed it under a related specialty and pulled a source from an old directory page. To a human reader, that is a small inaccuracy. For a brand, it is a smudged label on a box. I study failures like this in Portuguese and English: where a company surfaces in ChatGPT, Gemini, and Perplexity answers, where it disappears, and where it starts to look too much like a competitor.
This week I'm watching how AI systems respond when a brand name shares space with a nearby specialty. Two cases came up where the answer swapped the company for a neighbor on the same block — just because of a stale tag in a directory listing. I'm sketching the pattern before it turns into a note.
How ChatGPT can get a legal service wrong: the neighborhood is right, the specialty is borrowed, the source is old.
LanguageHow a simple audit table shows AI answer errors, sources, competitors, and next actions in brand visibility work.
LanguageWhy ChatGPT uses old business information: how an outdated directory can bring a retired service back.
Sources
I'm Rafaela, writing from São Paulo. For a couple of years I've kept a notebook of AI model mistakes — where brands surface, where they disappear, and where they start looking like a competitor. Notes are short, plain, and always anonymized.
You can send one example of a query where the company should appear. I'll look at it and reply calmly, without turning the first message into a sales pitch.
Get in touch