YouTube loses 91% of its ChatGPT citation share versus August

David Konitzny reported on 25 September that YouTube's share of ChatGPT citations fell 91% against August, based on hundreds of thousands of branded and non-branded prompts tracked over time.

Other major domains dropped on the same measure: facebook.com by 88.1%, wikipedia.org by 78.4%, and forbes.com by 71.7%.

Konitzny checked whether the fall was a share effect from a larger overall citation pool. He indexed YouTube's absolute citation volume against July and the drop showed up there too.

He is careful about what the data does and does not show. Retrieval is not the same as citation. These domains may still be retrieved and considered during search. They are reaching the final cited sources far less often than before.

What we think is behind it

We think the shift is consistent with ChatGPT relying more on its own index and summarisation, and less on lifting whole answers from a small set of large domains. That is one reading and it is not confirmed by OpenAI.

How to check it on your own data

Open your ChatGPT citation tracking and pull the absolute count of citations from youtube.com, wikipedia.org, facebook.com and forbes.com over the past several weeks. Compare August to late September on the same set of prompts.

Then look at the answers ChatGPT now returns for prompts where those domains used to be cited. Check whether your substance is still present in the answer, attributed to another source, or folded into the model's own summary. That tells you whether retrieval has changed or only the citation surface has.

Citation share moving is a signal about how ChatGPT is presenting sources, not a verdict on whether a domain is still useful to the model. Treat it as one input while you watch whether the pattern holds through October.