AI Word Counter
Words, characters, sentences and paragraphs, updating as you type, with reading and speaking time worked from published rates. Written for AI drafts in particular, because those count higher than they read and nothing on screen explains why.
Your draft is 280 characters and the field says it is too long.
This is the complaint that brings most people to a counter twice. The text looks short enough. You count it, and it is short enough. The platform disagrees, and nothing on screen accounts for the difference.
Text that came out of an AI assistant is carrying characters you cannot see. Some of them are invisible Unicode: zero-width spaces, non-breaking spaces, word joiners, byte-order marks. Some of them are ordinary whitespace that has simply accumulated — trailing spaces at the ends of wrapped lines, doubled gaps after punctuation, tabs from Markdown output. All of them are real characters, all of them count toward a limit, and none of them render as anything you could point at.
So a meta description that reads as 150 characters is rejected at 160. A post that looks well inside the limit will not send. An essay comes in over a word count that the writer measured as under. The count is not wrong in any of those cases. The text has more in it than it appears to.
This counter shows both figures — characters including whitespace, and characters excluding it — so the gap between what you see and what you have is visible rather than mysterious. When the gap is larger than it should be, that is the signal to clean the text before counting it again.
Four counters, four numbers, no explanations.
Paste the same paragraph into four word counters and you will get four numbers. They are all correct, because each has made a different decision about what a word is: whether a hyphenated compound is one word or two, whether a URL is one word or five, whether a standalone number counts at all.
This matters because a word count is almost never wanted for its own sake. It is wanted because someone else is going to check it — a marker with a limit, an editor with a brief, a client paying by the word. A count that disagrees with theirs is not a count, it is an argument waiting to happen.
So the rules used here are written out below rather than left for you to reverse engineer, and they are chosen to match Microsoft Word and Google Docs, since that is what the person checking will almost certainly be using.
How each number is arrived at.
- WordsA word is a run of characters with whitespace on either side. That makes "don't" one word, "mother-in-law" one word, a URL one word and a standalone number one word. It is how Microsoft Word and Google Docs count, which is the only thing that matters: a count that disagrees with the one your marker or editor is looking at is worse than no count.
- CharactersEvery character, spaces and line breaks included. This is the number that platform limits are measured against, so it is the one to watch for a meta description or a post. The separate no-spaces count is what some academic submissions ask for instead.
- SentencesA sentence ends at a full stop, question mark or exclamation mark. Abbreviations therefore inflate the count slightly, since a tool cannot tell the full stop in "Dr." from the end of a sentence without understanding the text. Treat it as an indicator rather than a precise figure.
- ParagraphsA block of text separated from the next by a blank line. Text that uses single line breaks between paragraphs rather than blank lines will therefore count as one paragraph, which is usually a sign the text needs its line breaks resolved rather than a sign the count is wrong.
- Reading timeCalculated at 238 words per minute, the figure from Brysbaert's 2019 meta-analysis of adult silent reading rates. It is an average across a lot of studies, so technical material runs slower and light prose faster, but it is a defensible number rather than a rounded guess.
- Speaking timeCalculated at 130 words per minute, which is the conventional presentation pace. Conversational speech runs faster, at 150 or more; a formal talk with pauses runs slower. If you are timing a script, this is the number to trust and then rehearse against.
Five places the number has consequences.
- Essays and assignmentsWord limits are usually enforced with a tolerance of ten per cent either way, and going under is penalised as often as going over. The thing to check is what your institution counts: footnotes, references and appendices are excluded by some and included by others, and the difference is easily several hundred words.
- Titles and meta descriptionsSearch engines truncate by pixel width rather than character count, but character count is the practical proxy: around 60 for a title and 160 for a description. Set the target option and the counter tells you how many you have left rather than making you subtract.
- Posts and captionsEvery platform counts differently, and several count a URL as a fixed length regardless of how long it actually is. The character count here is the raw one, so treat it as the floor rather than the exact figure the platform will apply.
- Freelance and commissioned writingPer-word rates make the count the invoice. Agree in advance whose count is authoritative, because a client counting in Word and a writer counting in Google Docs will get slightly different numbers on the same file and both will be right.
- Manuscripts and submissionsPublishers ask for a rounded figure rather than an exact one, and expect it to match a standard manuscript count. A novel quoted at 90,000 words means approximately that; quoting 89,412 reads as inexperience rather than precision.
Clean the text before you trust the number.
Counting measures characters, and characters you cannot see are still characters. A character count that includes invisible Unicode is not wrong, but it is not the number a platform limit will apply either, and the difference is why a field that accepts 280 characters rejects text that visibly contains fewer.
Two things throw a count off most often. Surplus whitespace inflates the character figure, which the AI space remover resolves. Invisible characters do the same without any visible cause at all, and the zero-width space remover names and removes those.
Both are part of what the ChatGPT text cleaner does in a single pass, which is worth reaching for when a draft needs more than one of these jobs doing. If the text came out of an assistant and the count is what matters, the ChatGPT watermark remover reports every hidden character it finds by code point, so you can see exactly how many of your characters are ones nobody will ever read.
A third is worth knowing about for paragraph counts specifically. Text extracted from a PDF uses a line break at the end of every line rather than blank lines between paragraphs, so it counts as a single paragraph however long it is. Resolving the line breaks fixes both the count and the text.
Word count questions.
Why is my character count higher than the text looks?
Because the text is carrying characters that do not render. Zero-width spaces, non-breaking spaces and byte-order marks are all real characters that count toward a limit and show as nothing on screen. Surplus whitespace does the same thing more visibly. Text that came out of an AI assistant is the most common source of both.
Why does a platform say my post is too long when it counts as short?
The platform is counting characters you cannot see. It is not applying a different rule; it is counting the same characters you are, plus the invisible ones. Cleaning the hidden characters out and counting again usually resolves the discrepancy exactly.
Does AI-generated text really carry extra characters?
Routinely, though most of it is layout rather than anything deliberate. Chat interfaces insert zero-width spaces to control line wrapping, Markdown output arrives with indentation intact, and any editor the text passes through adds non-breaking spaces of its own. By the time you paste a draft somewhere, several tools have each contributed.
How do I get an accurate count for an AI draft?
Clean it first, then count. Compare the characters figure against characters without spaces: if the gap is far larger than the number of spaces you can see, something invisible is in there. Removing it changes the count, which is the point.
How does the word counter count words?
A word is a run of characters separated by whitespace. Contractions like don't count as one, hyphenated compounds like mother-in-law count as one, numbers count as words and a URL counts as one word. This matches how Microsoft Word and Google Docs count, so the number here should agree with the number your editor or marker sees.
Why does my word count differ from Microsoft Word?
Usually because of what is included rather than how words are counted. Word counts text boxes, footnotes, endnotes and headers depending on the settings and the version, and its status bar count changes when text is selected. Paste only the body text here and the two figures should match closely.
Does it count characters with or without spaces?
Both, as separate figures. Characters includes spaces and line breaks, which is what platform limits measure. Characters without spaces excludes all whitespace, which is what some academic submissions ask for instead.
How accurate is the reading time?
It uses 238 words per minute, the figure from a 2019 meta-analysis of adult silent reading rates across many studies. Real speed varies with difficulty and purpose: technical material runs slower, familiar prose faster. Treat it as a well-founded average rather than a promise.
How long does it take to speak 1,000 words?
About seven and a half minutes at a presentation pace of 130 words per minute. Conversational delivery is faster, around 150. If you are writing to a time slot, write to the slower figure and rehearse, because nervous delivery speeds up rather than slows down.
Is there a limit on how much text I can count?
No. Everything runs in your browser, so there is no upload and no size limit. A full manuscript counts as fast as a paragraph.
Does it save or upload my text?
No. Nothing is transmitted and nothing is stored. The text stays on your device and closing the tab discards it, which makes it usable for unpublished or confidential work.
Can I check my text against a limit?
Yes. Set a target and the counter reports how many characters or words you have left, or how many you are over. The presets cover SEO titles, meta descriptions, posts and the two most common essay lengths.
How are sentences counted?
By counting full stops, question marks and exclamation marks that end a run of text. Abbreviations therefore inflate the figure slightly, since nothing short of understanding the sentence can distinguish the full stop in an abbreviation from the end of a sentence.
Why does my whole document count as one paragraph?
Because paragraphs are separated by blank lines, and your text is probably using single line breaks instead. That is common in text extracted from a PDF. Resolving the line breaks first gives an accurate paragraph count and generally makes the text more usable besides.
Do numbers and dates count as words?
Yes. A standalone number is one word, and a date like 12 March 2026 is three. This is the standard convention and matches what word processors do.
Does it work on other languages?
For anything that separates words with spaces, yes: French, German, Spanish, Russian and so on all count correctly. Chinese, Japanese and Thai do not use spaces between words, so a whitespace-based count is not meaningful for them, and no simple counter handles them properly.