Text Summarizer

The scoring algorithm, step by step

The summarizer is extractive and runs in the page with no model and no network request. It first splits the text into sentences at full stops, question marks, exclamation marks and ellipses. It then counts how often each meaningful word appears across the whole text, ignoring about 170 common English stop words (the, and, of, is…) and anything shorter than two letters.

Each sentence is scored as the average of its words' counts divided by the count of the most frequent word in the text. The first three sentences get a small position bonus of 0.15, 0.10 and 0.05, because opening sentences tend to state the topic. The top-scoring sentences are kept and put back in their original order, untouched.

The default sample, worked through

The pre-filled text has 92 words in 6 sentences. The most frequent content word is “bees” (3 times), so every count is divided by 3. Sentence 1 averages 0.50 and gains the 0.15 opening bonus for 0.65; sentence 2 scores 0.42 + 0.10 = 0.52; sentence 5, which mentions bees, nectar and honey, scores 0.48 with no bonus; sentences 4, 3 and 6 trail at 0.39, 0.38 and 0.33.

Medium keeps round(6 × 0.35) = 2 sentences, so it returns sentences 1 and 2: 30 words, shown as 67% shorter. Short keeps 1 sentence (11 words, 88% shorter). Long keeps round(6 × 0.55) = 3, adding sentence 5 for 45 words and 51% shorter. The percentage is simply 1 minus summary words over input words.

How many sentences each length keeps

Counts are rounded to the nearest whole sentence and never drop below one. There's no way to request an exact number of sentences from the page; the three presets are the only lengths offered.

Why it works best on plain English prose

• Only the letters a–z, digits and apostrophes count as word characters. Accented words are chopped (“café” scores as “caf”), and text in Arabic, Russian, Hindi or Chinese has no scorable words at all, so the tool simply returns the first sentence.
• The stop-word list is English. In Spanish or German, words like “de”, “und” and “der” are counted as content and inflate the scores of long sentences.
• Sentence splitting happens at full stops, question marks and exclamation marks, but a stop inside a number (“3.5 percent”) or after a common abbreviation (“Dr. Smith”, “e.g.”, “vs.”) is left alone, so those do not create fragments. Rarer abbreviations such as “approx.” still split; the sentence counter above the box reveals this before you summarize.
• Line breaks are collapsed to spaces, so a heading or bullet without a final full stop merges into the sentence that follows it.
• Texts that repeat one word heavily, such as a product listing, pull every sentence containing that word to the top regardless of importance.

Reading an extractive summary

Because sentences are copied verbatim, the summary can read as a set of highlights rather than flowing prose: a kept sentence may begin with “This” or “They” referring to a sentence that was dropped. That's the trade-off for a summary that can't misstate the source. Use it to decide whether a document is worth reading in full, to pull quotable key lines, or as raw material to tidy with the Paraphrasing Tool.

The tool can't condense a single long sentence, merge two ideas or write a new topic sentence. For a rewritten abstract, an AI tool is the better fit; for a faithful, instant cut of an article, report or meeting notes, this one is.