Cohesion is not the same as connective words
Start with the thing this tool cannot measure. A piece of writing holds together when each sentence gives the reader something to attach the next one to, and most of that work is done without any connective at all — by repeating a key noun, by using a pronoun that clearly points back, by putting old information at the front of a sentence and new information at the end. Connectives are the explicit layer on top of that implicit structure. They are useful, and they are not the structure itself.
Which means a low count is not automatically a problem and a high count is not automatically good. What the counts tell you is how much of your cohesion is being signposted rather than built in. If the answer is "all of it", the joins may be doing work the sentences should be doing. If the answer is "none of it", the argument may still be perfectly clear, and it is worth reading the paragraph openings in sequence to confirm.
Six functions, and why the balance matters
| Function | Typical words | What a high share suggests |
|---|---|---|
| Addition | also, furthermore, moreover, in addition | A list rather than an argument; points accumulate without ranking |
| Contrast | however, but, although, on the other hand | A piece built on qualification; check for hedging |
| Cause and effect | because, therefore, thus, consequently | A genuine argument; the healthiest concentration to have |
| Sequence | first, then, next, finally | Procedure or narrative; wrong shape for an analytical piece |
| Illustration | for example, such as, in particular | Evidence-led; check the examples are doing work |
| Summary | in conclusion, overall, in short | More than two or three is usually restatement |
The distribution is more diagnostic than the total. An analytical piece dominated by addition words is often a list wearing an argument's clothes: each paragraph adds a point, nothing follows from anything, and the reader finishes with a pile rather than a conclusion. A piece dominated by contrast words is frequently one where the writer has not decided what they think. Cause-and-effect connectives are the ones that carry an actual claim about the world, and a piece that never uses them may not be arguing anything.
The word that ruins its own effect
However is the connective most likely to be overused, and it has a particular failure mode: it is long, it is formal, and it takes the front position of a sentence where it is maximally visible. Three howevers in a page of prose is enough for a reader to notice the word rather than the contrast it marks. It also has an unusual number of near-substitutes with different weights — but is lighter and faster, yet is more literary, though at the end of a sentence is conversational, that said is casual, nevertheless is heavier still. Swapping between them is nearly free and it is why the repeat table exists.
Consecutive paragraphs opening with the same word are called out separately because they are the most visible repetition available. Readers scan the first word of a paragraph before committing to reading it, so two Howevers stacked one after another register even in a skim. Whatever the total count says, fix those first.
Ambiguity in the word lists, stated plainly
Several entries do more than one job in English and the tool cannot tell which job is being done in your sentence. While means both during and whereas. Since means both from that time and because. Still is a contrast marker and also a plain adverb meaning yet now. Rather is a connective in rather, the opposite is true and an intensifier in rather cold. Consider and that is are frequently just ordinary verbs and pronouns. All of them are on the lists because excluding them would miss real transitions, and the consequence is that your counts are an upper bound.
Matching is case-insensitive on whole words, and multi-word phrases tolerate any whitespace between their words, so a phrase broken over a line break still registers. When a connective is found at the start of a paragraph, leading quotation marks are skipped, which means a paragraph opening with quoted dialogue that begins with However is still detected. Phrases are tested longest-first, so in addition is recognised as itself rather than being missed. Turning on the sentence-openings-only option counts at most one connective per sentence and gives a much cleaner picture of signposting at the cost of missing every clause-joining because and although.
Using the paragraph opening table
The last table is the one to read carefully. It lists every paragraph in order with its opening connective, its function and its first few words, and reading that column top to bottom gives you the skeleton of the piece in about ten seconds. Three things are worth looking for. A run of the same function — four contrast openings in a row — usually means the argument is stalling rather than advancing. A summary connective appearing before the final third means you concluded early and then kept writing. And a long stretch of paragraphs with no opener at all, in the middle of a piece that signposts elsewhere, often marks the section you drafted fastest and thought about least.
Once the joins are sorted, the sentence length analyzer covers rhythm inside the paragraphs, and the bullet outline maker is useful if the structural problem turns out to be the order of the sections rather than the connections between them.
Questions people ask
How many transition words should a piece have?
No published figure is worth quoting, because the right number depends entirely on genre. Procedural writing needs sequence markers at nearly every step and would be confusing without them. Literary nonfiction can go pages without an explicit connective and hold together through repeated nouns and clear pronoun reference. The useful reading is the distribution across the six functions and the repeat table, not the total, and the most useful comparison is against another piece of your own writing rather than against a benchmark.
Why is "while" counted as a contrast word when I meant "during"?
Because a pattern matcher cannot tell the two apart, and excluding "while" would miss the genuine contrastive uses, which are common. The same ambiguity affects "since" (time or cause), "still" (contrast or plain adverb), "rather" (connective or intensifier) and "that is" (connective or an ordinary pronoun and verb). Everything on the lists is included knowingly, and the consequence is that your counts are an upper bound rather than an exact figure. The paragraph opening table is unaffected by most of this, because those words rarely open a paragraph in their non-connective sense.
My paragraphs are not being detected.
Paragraph boundaries are blank lines, meaning two consecutive line breaks. A draft where every paragraph is on a single line with no blank line between them reads as one paragraph to the tool, and a draft that uses indentation instead of blank lines does too. Pasting from a word processor usually preserves the blank lines; pasting from a PDF often does not, and pasting from a code editor with soft wrap can produce line breaks that are not paragraph breaks at all. Check the paragraph count in the summary before trusting the openings table.
Should every paragraph open with a transition?
No, and a piece where they all do usually reads worse than one where half do. Constant signposting tells the reader what relationship to expect before they have seen the content, which is helpful in dense technical material and patronising in most other registers. If the tool reports that every paragraph is signposted, try reading the piece with them all deleted. Whatever still connects was being carried by the sentences, and those connectives were labelling something already obvious.
Does the text get uploaded?
It does not. The word lists are part of the page and the matching runs in your browser on your own machine. Nothing about the input is transmitted, stored or logged anywhere, which is why the page keeps working with the network disconnected.