Word Frequency Counter: Practical Tips and Common Mistakes
Reviewed by the FreeOnline.fyi team · Updated 2026-09-16
What gets counted, and what doesn't
Paste a draft, a transcript or a subtitle file into the Word Frequency Counter and a ranked list builds while you are still typing — there is no submit button to press and nothing is uploaded anywhere. The header line is the first thing worth reading: it states "Top 20 of N words · M unique words", which tells you both how much text the tool actually ingested and how much of the vocabulary you are not seeing.
The counter splits text on whitespace and trims punctuation from the edges of each token, so "policy," at the end of a sentence and "policy." land on the same row. Numbers count as words too. What it deliberately does not do is stem or lemmatise: run, runs and running stay on three separate rows, and so do analyse and analysis.
Two switches under the textarea change the output more than anything else on the page. Matching is case-insensitive by default, so The and the merge. The stopword box is ticked on load, so the, and, of and to never reach the table unless you untick it.
When to switch off the stopword filter
Ignoring common English words is the sensible default for the question most people bring here: what is this document actually about? Strip the function words and the ranked list becomes a rough topic fingerprint — names, terms and repeated concepts rise straight to the top.
There are two situations where you want the filter off. The first is style editing: repetition of the, that, of and there is the classic tic in flabby prose, and hiding those words hides the evidence. The second is translation review, where an unnatural density of small words often reveals sentence structure copied from the source language.
A common mistake is leaving stopwords on and then deciding a piece is thin because only a handful of distinct words survive in the table. Toggle the box and watch the header counters move — same text, different totals. Also remember the stopword list is English-only, so filtering French, Spanish or German text gains you very little.
Case sensitivity changes the story
Switching matching to case-sensitive splits proper nouns from their common-word twins. Apple and apple become two rows with two counts, and so do Data and data. If your text is heavy with headings, product names or title-case subheads, this setting can reshape the ranking completely.
Case-sensitive mode is the right choice for entity work — checking how often a brand, person or place is named — and for spotting sentence-initial noise, because a capitalised The at the start of every sentence inflates the lowercase count in the merged view.
The mistake to watch for is leaving case-sensitive mode on without noticing, then reading a partial count as the real one. If a word seems rarer than your memory of the text suggests, check the matching control before you draw any conclusion.
Reading the percentage column honestly
The share column is a word's count divided by the total number of counted words — and that denominator shrinks the moment you tick the stopword box. Percentages therefore jump when you filter, even though the source text has not changed by a single character. Never compare percentages taken with different settings.
A worked example helps. In a 1,000-word article where "the" appears 62 times, the share reads 6.2%. Turn stopwords on and the denominator might fall to roughly 600 words, so a keyword with 12 mentions climbs from 1.2% to about 2%. Both numbers are correct; they simply answer different questions.
Natural language roughly follows Zipf's law: the most frequent word is about twice as common as the second, three times the third, and so on. That means the top-20 table is only the head of a very long tail. If the header reports several thousand unique words, the CSV holds the full ranking, which is where the interesting mid-frequency terms live.
Exporting a clean CSV
The Download CSV button produces word-frequency.csv with four columns — rank, word, count, percentage — for the full ranked list, generated entirely inside your browser through a local Blob URL. Nothing is sent to a server, so the file is safe to keep beside confidential drafts.
Export formatting is handled properly rather than approximated. Fields containing commas, quotes or line breaks are quoted and escaped following RFC 4180, the common CSV specification, and a UTF-8 byte-order mark is prepended so Excel opens accented characters and non-Latin words without turning them into mojibake.
The mistake we see most often is skipping the download and copy-pasting the on-screen table straight into a spreadsheet. You lose everything below rank 20, and long lists pasted this way often lose their sort order along the way. Take the file, then sort or pivot in the spreadsheet.
Where the counts still mislead
A word counter is not a linguistic analyser. No stemming means near-synonyms and inflections scatter across rows; no phrase detection means "climate change" is two unrelated entries; hyphens and apostrophes may keep compounds together as one token. Compare the header's total word count with your word processor's count — a gap usually points to punctuation handling rather than to a bug.
Scripts that do not separate words with spaces are a genuine limitation. Paste Chinese or Japanese prose and you get very few, very long tokens, because there are no whitespace boundaries to split on. For those languages a frequency counter of this kind is close to useless, and saying so is more honest than pretending otherwise.
Finally, remember that transcripts carry speaker labels, subtitles carry timestamps and raw HTML carries markup. Those artefacts will dominate your list until you clean them out. Treat the top 20 as a fast orientation, spot-check two or three counts by searching the original text, and reach for the rest of the free online tools at FreeOnline.fyi when you need a different kind of pass over the same document.