COMPARE TWO TEXTS
Find words that gained or lost prominence in a revision
Which words did the revision add? Which did it drop or repeat more often? Put the original in A and the revision in B to check words and two-word sequences. The text comparison tool also reports word counts, vocabulary size, normalized frequency, and Zipf diagnostics, with searchable rows and CSV or JSON exports.
Two public webpages work as inputs, as does a pasted draft paired with a live page. Before comparing URLs, read both previews. The passages you want to compare need to be present, and the extraction scope needs to match on both sides.
NORMALIZED COMPARISON
Why raw word counts and percentage-point changes can disagree
A longer draft can mention a word more often without giving it more prominence. Look at the two measures in the table: the change in count, and the change in the term's share of the analyzed words. A positive share change means it occupies more of B; a negative change means it occupies less.
term count ÷ analyzed words(share in B − share in A) × 100 ppTake “draft”: 5 mentions in 100 analyzed words in A, then 8 in 200 in B. That is three more mentions. Its share, though, has fallen from 5% to 4%, a change of −1 percentage point. Keeping both figures in the table lets you distinguish extra mentions from greater prominence.
A bigram share uses the available two-word positions as its denominator, which is one fewer than the analyzed word count. The denominator differs for keyword density, where exact phrase counts are divided by total words.
WORKED EXAMPLE · REVISION REVIEW
A revision adds repetition and removes a useful phrase
Version B is only one word longer. Yet “content review” grows from one exact occurrence to four, while “source note” disappears. Paste these original example texts as A and B, select English, and enable Keep stop words to reproduce the counts.
A · Original · 40 wordsA content review starts with a source note. The editor checks the price, keeps the product name, and removes a repeated sentence. Before publication, a second editor reads the draft for clarity. The source note stays beside the changed claim.
B · Revision · 41 wordsA content review starts with the product name. The content review checks the price, and the content review removes a repeated sentence. Before publication, a second editor reads the content review for clarity. The changed claim now has no supporting note.
| Exact phrase | A count | B count | A bigram share | B bigram share | Change |
|---|---|---|---|---|---|
| content review | 1 | 4 | 2.56% | 10.00% | +7.44 pp |
| source note | 2 | 0 | 5.13% | 0.00% | -5.13 pp |
These are bigram shares. A has 39 consecutive two-word positions; B has 40. The Keyword Density Checker instead divides each exact phrase count by the total words, giving “content review” occurrence densities of 2.50% and 9.76%. Both calculations are useful when their denominators stay explicit.
Open the “content review” occurrences and read them in context: the revised sentences repeat a label where a direct subject could work. Then inspect the passages that contained “source note.” The missing phrase is a reason to check whether attribution was lost, not proof that the claim is false. Decide from the text and its sources, rather than aiming for a target percentage.
Use Text Analysis Comparison to inspect changed words and bigrams, then open their source context. For a single version, Word Frequency Counter gives a simpler repetition table.
Download both texts, settings, and expected counts (JSON). The file is a readable fixture; copy its original and revised text values into the tools.
MATCH THE SETTINGS
Keep language and stop-word settings consistent
If both versions use the same language, select it yourself in the settings. Automatic detection works on each source separately. It can assign different languages to the two inputs, particularly with short passages or mixed-language text; a warning in the result points this out.
The default count leaves out stop words before counting either words or bigrams. Removing them changes the analyzed total; it can also join two words into a bigram when a stop word originally sat between them. Enable Keep stop words to compare consecutive phrases in the full token sequence. Punctuation does not mark a phrase boundary here. Check the source context before deciding that a bigram represents a repeated expression.
Case makes no difference to matching, but the written form does. “Edit,” “edits,” and “editing” count as separate words, and synonyms stay separate too. After filtering, each input must still contain at least three analyzed tokens.
REVISION WORKFLOW
Read the largest changes, then inspect the passages
The largest changes in normalized frequency appear first. To check a term you expected the edit to keep, search for it and select the row to read its occurrences in A and B. The bigrams view does the same for two-word sequences. A missing phrase calls for a look at the passages it came from: the absence alone does not establish that an explanation or attribution was lost.
Pasted text and reviewed URL text are compared in your browser. URL input is fetched through the server without retaining the page content. The public comparison API uses its existing whole-source behavior; it does not inherit the extraction selection from this web interface.
What the exports include
The screen displays up to 500 matching rows. A CSV export contains the returned rows from whichever view you selected, words or bigrams. JSON includes both returned comparison tables along with the summary metrics. Searching the table does not narrow either export. Each table returns at most 1,000 rows; a visible notice tells you when the result is partial.
WORKED EXAMPLE · WEBPAGE EXTRACTION
Same article, different menu: why the extracted text matters
The two downloadable pages contain the same article. Page B adds repeated navigation and footer labels. With English selected and Keep stop words enabled, analyzing the whole page produces different word counts even though the article has not changed.
A clear revision checklist A revision checklist helps an editor keep useful facts, remove unnecessary repetition, and explain each change. Before publishing, read the revised page once for accuracy and once for flow. Record the source of every changed claim.
| Extracted source | Words | revision count | revision share | contact count | contact share |
|---|---|---|---|---|---|
| Page A · whole page | 45 | 2 | 4.44% | 2 | 4.44% |
| Page B · whole page | 49 | 2 | 4.08% | 5 | 10.20% |
| Either page · main content | 40 | 2 | 5.00% | 0 | 0.00% |
“revision” still appears twice, but its share drops from 4.44% to 4.08% when the extra template words enter the denominator. “contact” rises from two to five occurrences. Selecting the main content gives the same 40-word input on both pages: “revision” is 5.00%, and “contact” is absent.
To reproduce this, use Word Frequency Counter with the example URLs below. Preview the extracted text, compare whole-page and main-content selection, then run the calculation. The preview should include the heading and article paragraph in main-content mode, without the navigation or footer.
When comparing article vocabulary, use the same extraction scope on both sources. When reviewing the full page, retain the template text deliberately. A cleaner extraction does not prove that the article is better or more likely to rank.
For two-page work, compare the extracted texts side by side. The web tools fetch URL content through the stateless server, then calculate from the reviewed text in your browser. The public analysis API keeps its existing whole-source behavior; it does not inherit your web preview selection.
LIMITATIONS
Does this comparison check similarity or plagiarism?
This comparison measures changes in surface vocabulary. There is no check against a database of published work, and the result cannot establish plagiarism or tell you whether two sentences have the same meaning. For a cosine score based on shared vocabulary, use the text similarity calculator and read its limitations before interpreting the score.
The vocabulary may stay the same while a claim changes. Conversely, an important term may appear more often because the revision adds a useful example. Frequency changes and Zipf diagnostics cannot choose a winning version or predict which will rank higher. Judging the revision still means reading it for intent, readability, factual accuracy, and support from its sources.