ChatGPT's search feature has been quietly rebuilt, and a closer look at how it fetches information reveals a gap that matters for anyone trying to get their business noticed by AI tools: getting pulled into a search doesn't mean getting credited for it.
A technical analysis of the new search architecture found that in one test, 84 separate Reddit discussion threads were retrieved by ChatGPT's search tool as candidate sources. None of them showed up as citations in the final answer. The content was used, apparently, to help the model understand a topic — but the source never got named or linked.
This matters because ChatGPT's search function works in two stages. First, it fetches a pool of web pages, forum posts, and articles related to a query. Second, it decides which of those sources to actually cite in the response it gives the user. The analysis suggests these two stages are far less connected than most people assume — a page can be read and used to shape an answer while never appearing as a link the user can click.
The rebuild itself changed the technical format of how ChatGPT issues search requests and processes results, essentially altering the internal instructions the model uses to decide what to fetch and how to weigh it. That's a plumbing change, not a visible feature. Most users won't notice a difference in the interface. But it changes what content creators and businesses need to understand about how their material might be used — or ignored — by the tool.