Text & Writing Tools

What these tools are for

Most of the work here is small, dull and necessary. You have a block of text and it is not quite the shape you need. Line Tools sorts and tidies lines. List Converter turns one kind of list into another. Invisible Character Remover strips the hidden characters that ride along when you copy from a web page or a chat window.

Others serve a particular trade. Keyword Density Checker counts how often each term shows up, which matters if you write for search. Filler & Cliche Checker points at phrases most drafts are better without. Accent Remover & Transliteration turns accented letters into plain ones, which helps when a filename or a slug has to stay simple. Teleprompter scrolls your script while you talk to camera.

The rest lean towards play and planning: Anagram Maker, Cipher Tools, Letter Frequency Counter, Kanban Board, Habit Tracker. Writers, editors, coders and students all end up somewhere in this list.

Text is small, and often private

Text is tiny next to photos or video. A whole novel is a few megabytes. Your browser can sort, count and rewrite that without help, so there is no honest reason to send it to a server first.

There is a second reason to keep it local. Text is often the most private thing on your machine: an unpublished draft, a client's copy, a contract, a note to yourself. Nothing here is uploaded. The build checks it and fails if any tool tries to reach the network, and every page keeps working with the network switched off.

The trade-off is memory. There are no accounts, so nothing syncs. A board you build on the laptop stays on the laptop, and it goes if you clear your browser data. Keep a copy elsewhere when it matters.

Worth knowing before you start

Counting is less exact than it looks. Word counts depend on where you decide a word ends: hyphens, apostrophes, numbers, dashes. Two honest counters can differ by a percent or so. If someone else enforces the limit, do the final count with whatever they use and treat every other figure as a guide.

Clean first, then convert. Strip invisible characters and tidy the lines before you reformat, not after. And keep the original. Some changes lose information for good: once the accents are gone, nothing can tell which letters used to have them.

When two tools sound alike, choose by what you want back. Text Repeater hands your own words back again and again. Lorem Ipsum Generator gives placeholder text that nobody will mistake for real copy.

When a paste arrives as é instead of é

Text that comes back full of é, ’ or п has not lost anything yet. It was stored as UTF-8, then read back one byte at a time as though it were an older single-byte character set. Two bytes became two characters, and that wrong reading was then saved as perfectly valid Unicode. It is a decoding fault, not a font fault, and that is why it can usually be undone.

Mojibake & Encoding Repair runs the fault backwards. Each damaged character is turned into the byte that single-byte set would have used, and the resulting byte string is read again as UTF-8. When the guess is wrong the bytes do not form valid UTF-8 and the attempt fails cleanly, so the tool can afford to try everything: eleven single-byte sets, among them Windows-1252, strict Latin-1, Mac Roman, Windows-1251, KOI8-R and code page 866, each on its own and then every pair of them for text that went through the mistake twice. The candidates are ranked by how many suspicious runs disappeared, and the best eight are listed with a plain sentence saying what happened to get there.

The byte tables are not shipped with the page. Each one is built when the tool loads, by asking the browser's own decoder what all 256 bytes mean in that character set and turning the answer round, so a table can never disagree with the decoder that will read the repair afterwards. One kind of damage is beyond all of this: a replacement character, the black diamond with a question mark in it, means some earlier program threw the byte away. Those are counted and reported as lost rather than covered up with a confident guess.

Comparing two drafts: a view, a number, or a document

Three tools here take two versions of the same text, and they are not interchangeable. Choose by what you want to be holding at the end.

Text Compare gives you a view. It lines the two up with a longest-common-subsequence table, then makes a second pass word by word inside lines that only changed a little, so one altered word reads as one altered word instead of two lines rewritten in red and green. The table is capped at twelve million cells, which leaves room for two files of around three thousand lines each; past that the tool drops back to a straight line-by-line comparison rather than freezing the tab.

Text Similarity gives you a number, and on purpose it gives you five: edit distance, Jaro-Winkler, Dice over adjacent letter pairs, Jaccard over the distinct words, and cosine over word counts. Each is scaled so that 100 means identical under that measure, which is not the same as identical text. Two paragraphs built from the same words in a different order score 100 on Jaccard and much less on edit distance, and that gap is the useful part of the answer. Edit distance suits typo-level differences, Jaro-Winkler is built for names and weights a matching start, and Jaccard is the one to read when you only care which words the two share.

Redline Document Maker gives you something to send. It matches whole paragraphs first, because comparing a long contract in a single pass finds clever matches that pair a word in clause 2 with the same word in clause 40, and the marked-up result is unreadable. Where a deletion and an insertion land in the same place and share at least forty per cent of their words, the pair is treated as one edited paragraph and compared word by word; below that they stay a separate cut and addition. A paragraph removed in one place and put back unchanged somewhere else is labelled a move, instead of appearing as a deletion plus an insertion the way a word processor shows it. If two paragraphs need more than two thousand edits to reconcile, the search stops and they are reported as wholly replaced, and the page says that is what happened. The .docx it writes carries real revision marks with an author and a date, so Word, LibreOffice and Google Docs all offer the changes to accept or reject one at a time.

The two tools that build pages instead of printing a scroll

Nearly everything in this category hands back text. Two tools hand back pages, and both lay those pages out themselves rather than trusting a browser to print a long scroll.

Screenplay Formatter treats a script as a fixed geometry. Twelve point Courier is ten characters to the inch and six lines to the inch, so a page can be measured in characters instead of boxes: scene headings and action sixty characters wide starting an inch and a half in, dialogue thirty-five characters from two and a half inches, parentheticals twenty characters, the character name at 3.7 inches. The body runs from an inch below the top to an inch above the foot, which on US Letter is nine inches, fifty-four lines, and roughly one minute of screen time. That last equivalence is why nothing is allowed to drift, and why the tool measures the real width of a character when it loads and adjusts the point size until ten of them span an inch. On a machine with no Courier New the substitute face has a different advance width, and the page count quietly stops meaning anything. Page breaks then follow the rules a script supervisor would apply: a scene heading never sits alone at the foot of a page, a character name is never parted from their first line of speech, and a speech split across a break gets (MORE) at the foot and CONT'D at the head of the next page.

Essay & Manuscript Formatter paginates for a different reason. A browser will print with the right margins, but there is no CSS that any browser actually supports for putting a surname and a page number at the top of page four, and a running head with the page number in it is exactly what MLA, Chicago and a standard manuscript all require. So each page is assembled here as a real sheet with its own head, and the printer is handed those. Where a paragraph has to break across a page, the words are added back until the block is exactly as many lines as the page has left, found by halving the range rather than creeping forward a word at a time.

The measurement almost everyone gets wrong is double spacing, which is not twice the point size. A word processor doubles the font's own single-line height, and for twelve point Times that height is about 13.8 points, so a double-spaced line is nearer 27.6 points than 24. Get it wrong and a page holds twenty-seven lines instead of twenty-three, the page count comes out about a sixth short, and the old convention of 250 words to a page stops holding. The single-line height of each face is written into the tool, and every other number on the page is built from it.