Dialogue Punctuation Checker

What it actually measures

The checker walks quotation marks. After each closing quote it looks at the next words and asks a narrow question: is this a tag or a beat, and did the punctuation agree? A tag is a speaking verb — said, asked, whispered, replied, and a long but finite list of cousins. A beat is a body: walked, smiled, nodded, turned. Tags keep the sentence open with a comma and a lowercase continuation. Beats start a new sentence with a period and a capital.

It also looks for unbalanced quotes, a mix of straight and curly marks, spaces before punctuation, questions shaped like statements, and adverbs glued to tags. The adverb flag is a note, not an error. An adverb that contradicts the line (“I’m fine,” she said brightly, while bleeding) is doing honest work. An adverb that repeats the line is usually the line failing to act. The tool cannot tell those apart. You can.

It will not understand screenplay format, plays, or thought written without quotation marks. If it says it found no quoted speech, that is a format mismatch, not a verdict on the scene. British logical punctuation — comma outside the quotes — will look “wrong” to a checker trained on the American manuscript default. If that is your house style, ignore those flags and read the comparison guide before you ‘correct’ a UK imprint into a US one.

How to use the flags

Run it after the scene exists, not while you are still finding the want. Early punctuation panic produces correct commas around lifeless lines. When you do run it, fix errors before notes. An action beat punctuated as a tag changes who the reader thinks is speaking and how fast the line lands. An adverb note might be the best sentence on the page.

Load the messy sample first and read each explanation before you look at your own pages. The sample is built so that a period-before-tag, a walked-as-tag, and a comma after a question mark all fire. Then load the clean sample. It should stay quiet. If your own pages produce a pile of capitalised-tag errors, you are probably hitting return and letting the software capitalise She. That is a habit, not a style, and it is worth one dedicated pass.

Do not accept a zero as absolution. The tool cannot hear that three people sound identical, that the subtext was explained, or that the scene began with hello. Those are the other guides. Punctuation is the part we can make boring on purpose, so the unboring part has a clean bed.

Limits, on purpose

Nested quotes, interruptions with em dashes, and speeches that run across paragraphs are only partly seen. A human copy editor still wins on those, and should. The tool also cannot know your character’s name from a pronoun, so “Mara said” after a comma is treated as a legal name-led tag, which it is, while “She said” after a comma is flagged, which it should be.

False comfort is the real risk. A page can be punctuated like a handbook and still be a briefing in costume. Use the share link to show a collaborator the counts — error total and flag total — not the scene. The link never contains the pasted text. If you are working on a machine you do not trust, the text is still in that browser’s memory until you close the tab; the Lab does not have a copy.

If you want a person rather than a heuristic, a manuscript critique or a book editor will look at rhythm as well as commas. This checker is the pass you can run at 1 a.m. without sending the chapter anywhere. It is not a script coverage service and it will not tell you if the scene belongs in the book.

A teaching note

When a flag fires, rewrite one line by hand instead of globally replacing commas. The difference between “I’m late,” she said and “I’m late.” She put the keys down is an acting choice. Global fixes erase the choice. Students who only chase a green light learn to fear said and to trust smiled, which is the opposite of the lesson.

Pair this pass with the tag auditor. Punctuation tells you whether the verb was attached legally. The auditor tells you whether you are hissing your way through a scene that only needed said and a hand on a door. Together they cover the surface. Subtext still requires a person.

A worked pass, in the order that helps

Start with a passage you already know is messy, not with the scene you are proud of. The messy sample on the tool is there so the first result is a lesson rather than a verdict on your book. You should see a period where a comma belongs, a walking beat punctuated as if the body spoke, and a question wearing a comma it does not need. Read the explanation beside each flag before you touch the text. The explanation is the product. The highlight is only the index.

Then fix one class of error and run it again. If you repair everything in a single blur, you will not remember which rule you were obeying. Commas before tags first, because they change the rhythm a reader hears. Capitalised She next, because that is usually a word processor, not a decision. Action beats last among the errors, because turning a beat into its own sentence is an acting choice and you should feel yourself make it. Leave the adverb notes until the errors are gone. Some of those notes are the best sentences you have.

When the sample is quiet, paste a page of your own and do not accept a zero as a grade. Ask the questions the tool cannot: who wants something, who refuses, and which line would still be there if you deleted the greeting. If the page is clean and dead, the punctuation was never the problem. Go to the cutting guide or the subtext generator and give the scene a hidden want. Come back for commas when the want is audible.