Word frequency analysis: seeing what your text actually overuses

Ask a writer which word they lean on too hard and they almost never know. We all have crutch words — a favourite intensifier, a pet transition, a noun we reach for three times a page — and we are structurally unable to notice them, because to us they just feel like the natural word. A word frequency analyser is the bluntest possible instrument for fixing that blindness: it counts every word and shows you the ranking, and the surprises near the top are usually the words you'd never have guessed.

It's the oldest trick in text analysis, but it underpins a lot of more sophisticated work — keyword research, authorship attribution, readability, even the term-frequency weighting that search engines were built on. This piece walks through what a frequency count actually tells you and how to read the output without drawing the wrong conclusion.

The stopword problem

Run a raw frequency count on any English text and the top of the list is completely predictable: "the", "of", "and", "to", "a", "in". These are stopwords — the high-frequency function words that hold sentences together and carry almost no topical meaning. In nearly any document, "the" alone is around 5–7% of all words. A raw count, then, mostly tells you that English is English.

That's why the word frequency analyzer lets you toggle stopword filtering on and off. With filtering on, the function words are dropped before counting, so what rises to the top are the content words — the nouns, verbs, and adjectives that actually reflect what the text is about. There's also a minimum-length filter to screen out short tokens that slip past the stopword list. Turning filtering on is the move that converts a count of grammar into a portrait of subject matter.

When is the raw view useful? When you specifically care about style and rhythm rather than topic — an unusually high rate of a particular function word ("that", say, or "just") is itself a writing tic worth noticing. So the toggle isn't a "right vs wrong" setting; the two views answer two different questions.

Why the percentage matters more than the count

The tool reports, for each word, both a raw count and a percentage of the total words counted. The percentage is the number to trust, because counts don't compare across texts. "Budget" appearing 12 times means something very different in a 300-word memo (4%) than in a 6,000-word report (0.2%). The percentage normalises for length, so you can compare a word's prominence between documents, or against a known baseline for the language.

Note one subtlety in how the percentage is computed: it's a share of the counted words, not of the whole text. If you've enabled stopword filtering, the denominator is the content words only — so the percentages tell you each content word's share of the meaningful vocabulary, which is usually exactly what you want when judging whether something is overused.

What to do with the table

Once you have a ranked, filtered table, a few patterns are worth hunting for:

1. The accidental crutch. A content word sitting surprisingly high — not a key term, just a word you happen to overuse. "Really", "important", "various", "leverage". These are candidates to vary or cut.
2. The buried keyword. For SEO or content work, the opposite check: is the term you're actually trying to rank for present at a sensible frequency, or did you dance around it with synonyms and never quite say it? The frequency table makes keyword presence (and keyword stuffing) visible at a glance.
3. The topic fingerprint. The top ten content words are a compact summary of what the document is genuinely about. If they don't match what you intended the piece to be about, your focus has drifted.

Because the table exports to CSV, it's also a handy front end for larger jobs — comparing the word profile of several drafts, tracking how an edited version's vocabulary shifted, or feeding the counts into a spreadsheet for charting.

The limits of counting

A frequency count is deliberately dumb, and it's worth knowing where that bites. It treats inflected forms as separate words: "run", "runs", and "running" are three entries, so a concept can be more dominant than any single row suggests. It has no idea about meaning, so it can't tell you that "bank" appeared in two different senses. And high frequency isn't automatically bad — repeating your core subject noun is often correct and clear; the cure for repetition is sometimes worse than the disease.

So use the count as a starting point for your judgement, not a rulebook. It surfaces candidates; you decide which are real. Paste a draft into the word frequency analyzer, turn stopword filtering on, and read the top of the list with fresh eyes. The word you didn't realise you'd used eleven times is usually sitting right there, waiting to be noticed.