How ChatGPT Chooses Which Brands to Recommend (Changelog)
ChatGPT recommends brands from two places: patterns learned during training, and live pages when its search tool is on. Without search it answers from memory, so brands with broad, credible open-web coverage surface most. With search on, it retrieves and cites current pages. We track how this behavior shifts release to release.
The two answer modes
ChatGPT answers in one of two ways. In its default mode it responds from training data, naming brands without citing sources. When the search tool activates, it retrieves live pages and links them. The same question can produce different brand lists depending on which mode fires, which is why measuring both matters.
What pushes a brand into the answer
- Frequent, consistent mentions across independent, credible sources
- Clear, structured descriptions the model can attach to the right use case
- A name strongly associated with the exact category in the question
- For the search mode: fresh, crawlable, directly-relevant pages
What to watch after each release
Model upgrades quietly change recommendation behavior: which brands lead, how cautious the model is about naming any brand, and how often it triggers search. The changelog below records observed shifts so you are not guessing why your mention rate moved.
変更履歴
- Established this evergreen page tracking ChatGPT brand-selection behavior.
- Baseline: default mode answers from training data without citations; search mode retrieves and links live sources.
- Future entries will record dated, verified changes observed after each OpenAI release.