Well it isn’t searching the web, it has a cut off date, and it takes bits of info from various sources often distorting them, hence it’s not a search engine.
ChatGPT.com (and other LLM UIs such as Perplexity) now use a Tool that searches the web if it detects that it is necessary to solve the user's question, and then uses the output of that search query to answer the user's question. This allows it to surface responses that are out of its training data cutoff date, and cite specific data sources.