Detect overused words in text

Empty text, whitespace-only paste, and missing thresholds are rejected with invalid_input so a blank form cannot become a silent empty success. Use it free in the browser while drafting, or call the API at $0.002 per successful request when you need reproducible overuse lists in CI, CMS hooks, or editorial tooling.

What overused words mean and what the response contains

Overused-word detection answers a practical writing question: which meaningful terms dominate this passage relative to its full length, once everyday stopwords are set aside. Total length includes every token, stopwords and content words alike, so the share always reflects how large a fraction of the whole text that content word occupies. Stopwords are counted in the length denominator but never appear in the overused list, which keeps articles, prepositions, and auxiliary verbs from drowning the signal. There is no stemming, no thesaurus, no language detection model, and no machine-learning ranker that could drift between runs. If you need a full frequency table of every word including stopwords, use a word-frequency counter; if you need SEO density against one target phrase, use a keyword-density tool. A word is flagged when its count divided by total_words is strictly greater than the threshold you supply. The response echoes the source text and the numeric threshold, reports total_words as every token found, content_words as tokens that are not stopwords, stopwords_excluded as how many tokens were filtered, unique_content_words as the number of distinct content keys, overused_count as how many content words crossed the threshold, and overused as a ranked array of word, count, and share rounded to six decimal places. The engine either builds an exact overuse report inside the supported character length or rejects the request with invalid_input. Rejected cases include a missing text field, a null value, an empty string, whitespace-only content, objects and arrays that are not plain text, payloads longer than the documented maximum, strings with only punctuation and zero letter or digit tokens, and a missing or out-of-range threshold.

How the text and threshold fields are read and how rankings stay reproducible

The value must be a non-empty string after a meaningful content check; finite numbers and bigints are stringified first so a bare product code typed as a number still produces a stable digit token. Outer spaces around real words are separators only; they do not create phantom tokens. A value of 0.05 means any non-stopword that occupies more than five percent of all tokens is reported as overused. Thresholds of zero, negative numbers, values above one, non-numeric strings, and missing threshold fields are rejected. Tokenization uses Unicode letter and digit runs with optional internal apostrophe segments so contractions stay whole when the source uses straight apostrophes. All keys are lowercased so Really and really merge into one content word. Counting uses a single tokenize pass, a Map for content tallies, and a pure comparison sort with no network dictionary, no locale table that reorders letters differently across hosts, and no random sampling. The preferred text field is text, but aliases input, string, value, content, and s are accepted so query strings, spreadsheets, and typed forms can reuse the same contract without renaming columns. Booleans, plain objects, and arrays are rejected with a clear message so a mistaken JSON body does not become the word true or an unhelpful object dump. Empty string and whitespace-only paste are always invalid_input: the product promise is detect overused content words, not invent a zero success for a blank submit. Length is capped at one hundred thousand characters so a runaway paste cannot monopolize a Worker; longer payloads fail with an explicit limit message rather than a truncated silent cut. The preferred threshold field is threshold, with aliases max_share, share, max_frequency, and freq. It must be a finite number greater than zero and at most one, representing the maximum allowed share of total tokens. Pair the free browser run on this page with the priced API when you need the identical report inside an automated pipeline.

Where overused-word detection fits in writing, teaching, and software

On this page the work runs entirely in your browser; through the API each successful analysis costs $0.002, failed validation is not charged, and nothing is stored after the response returns. That split—free interactive exploration and a priced, identical API—matches the rest of the KIT tier-A catalog and lets you move from a classroom demo to an automated pipeline without changing the numeric result.