How the shortened version is chosen
This scores every sentence by how many frequently-repeated, meaningful words it contains, then keeps roughly the share of highest-scoring sentences you choose and puts them back in their original order - the same extractive approach used by this site's Text Summarizer tool, but presented as flowing shortened prose rather than a bullet list.
No sentence is rewritten or condensed internally - this selects existing sentences rather than compressing their wording, so the result reads as excerpts from the original, not a true paraphrase.
Why sentence scoring uses word frequency
A word that repeats often across a piece of text is a reasonable signal that it's central to the piece's main point, while words that appear only once are more likely to be supporting detail or incidental phrasing - scoring each sentence by how many frequent, meaningful words it contains is a simple but effective way to estimate which sentences carry the core message versus which are more peripheral.
Why the output doesn't hit the target percentage exactly
The reduction target is applied to the number of sentences kept, not the number of words - so the actual word-count reduction depends on how long the specific sentences that get kept or cut happen to be. A text where the most important sentences are also the longest ones will shrink less in word count than the sentence count alone would suggest.
When this approach works best
Extractive shortening works best on clearly structured writing where each sentence tends to carry one distinct idea - reports, articles, and explanatory writing. It works less well on writing where meaning depends heavily on connecting transitions between sentences, since removing a sentence can occasionally leave an abrupt jump between the ones that remain.