File-level deduplication, also known as single-instance storage, operates at the granularity of the whole file: it computes a hash of the entire file and compares it against previously stored files, storing only one physical copy of any file that is a byte-for-byte match and replacing subsequent identical copies with pointers back to that single stored instance. This whole-file granularity is precisely its limitation: file-level deduplication has no mechanism to recognize partial similarity, so the moment even a single byte within a file changes, the modified file no longer matches the hash of the previously stored version and is therefore treated as an entirely new, unique file. The deduplication engine cannot identify that only a small portion differs; it must store the entire modified file as a new, separate copy, alongside the original unmodified copy that may still be referenced elsewhere. This is the well-documented drawback of file-level deduplication relative to block-level deduplication, which instead identifies and stores only the specific changed segments within a file, achieving substantially better efficiency for files that are edited incrementally over time. Because a new, whole copy is created rather than the original being altered in place or intelligently segmented, 'another copy of the file is created' is the technically correct description of this behavior, and is the answer this document treats as authoritative.

Reference topic: Data Backup and Deduplication - File-Level vs. Block-Level Deduplication Granularity.