How this tool works
Without a key this does real work in your browser rather than trimming the article. Every sentence is scored on term frequency — how often its words appear across the whole text, ignoring stopwords — then normalised by the square root of its length so that short fragments and very long sentences compete fairly. Sentences near the opening and the close get a modest lift, because that is where written prose puts its weight. The highest-scoring quarter is kept and re-emitted in the original document order.
That is extraction: the sentences you get back are verbatim from your text. Add a free AI key and a language model writes a summary instead, in its own words, which reads better but introduces the risk that it phrases something the source did not quite say.
Getting a better summary
- Give it enough to work with: scoring needs at least a few full sentences. Under about four, there is nothing to rank and it will tell you so.
- Paste clean prose: navigation text, cookie banners and repeated headers inflate term frequencies and pull unrelated sentences into the result.
- Expect roughly a quarter: the extractive summary keeps about 25% of your sentences, with a floor of three and a ceiling of eight.
- Check the numbers: with a key, verify any figure in the summary against the source. Extraction cannot misquote a statistic; a language model can.