Without web search, a model answers from what it learned in training, which stops at its training cutoff. With web search, it can look up current pages and ground its answer in them, adding links as citations.
OpenAI’s API documentation describes web search as giving models access to up-to-date information from the internet, with sourced citations. It also says the model can choose whether to search based on the question, can search more than once, and records the full list of sources it consulted, which is often longer than the list it cites.
This matters for brands because grounded answers depend on which pages the search finds and trusts. A new product, a rebrand or a corrected fact can appear in a grounded answer once the pages are published and found; an answer without search can’t see them.
Citationly asks every question with web search on, and the report records, for each answer, whether web search actually ran. Answers that came back without search are flagged wherever they appear. Read how ChatGPT decides which brands to name.