[gst]

Even if they just have an index without the full copy it's not that hard to reconstruct a version similar to the original just from the index, as in an inverted index you typically do not only store the documents of a term, but also the word number within the document. However, it's not possible to restore the original version exactly, due to things like stemmers.

[sfphotoarts]

often in text indexing the original document is kept for things like snippet generation.