Google publicly claims that its AI model only uses live, current web results.
This claim is false, because the model clearly uses both live web data and stored, internal index data that includes deleted or no‑longer‑accessible content.
The evidence shows that Google’s AI Overview continues to reference Microsoft Q&A pages that were deleted weeks ago, Reddit posts that no longer exist, and other content that has been removed from the public internet. These pages cannot be opened, cannot be viewed, and do not appear in any normal search result. They are gone. Yet the AI still cites them as if they were available sources.
This proves that the model is not relying solely on the live web. It is also pulling information from internal, retained index entries that Google keeps even after the public version of the page has been deleted. These internal versions include old snapshots, metadata, and stored representations of pages that the user cannot see, cannot access, and cannot delete.
The AI is therefore working from a mixed dataset: part of it comes from the current, live web, and part of it comes from stored, outdated, or deleted internal data. This directly contradicts Google’s official statement.
The Google AI does not work only from live web results. It also uses stored, deleted, internal index data, and this contradicts their official claims.
This is why the AI can reference something that has been removed for weeks and something that is still online at the same time. It is not using the web as it exists today. It is using the web as Google has stored it internally, including parts that no longer exist publicly.
This is the image the AI shows as a result:
For anyone who doesn’t want to download the PDF of the deleted Microsoft post, here is an image from the beginning of it:
