Methods
Outerwoven measures public language: which words and phrases are being used more than usual, why, and whether the change lasts. Every figure on the site comes from the rules below. When a rule changes, it gets a new version, listed at the end, so a figure can always be traced to the rule that made it. The archive is young (collection began in mid-2026), so early figures rest on short histories.
What we collect
News is measured from The GDELT Project's index of about 100,000 English headlines a day from roughly 6,500 outlets, counted the same way every day since July 2025. Podcasts and newsletters are measured from our own feeds: podcast episodes (with their transcripts where the show publishes them) and Substack newsletters. Our own news collection, headlines and the opening text of items from news outlets' RSS feeds, supplies the conversations behind a spike. We count items, not views: a story read by millions and a post read by dozens each count once. Every rate below is measured against how much was counted that day.
What counts as a use
A phrase is a single word or a run of two or three words that appears literally, in order, in an item's headline or the first 600 characters of its text. Nothing is stemmed or invented. Phrases never cross punctuation, single words need at least three letters, and phrases can't start or end on a filler word like “of” or “the”. A phrase counts once per item however often the item repeats it. Different spellings and forms of one word (“far right” and “far-right”, “noncitizen” and “noncitizens”) are added together under one page.
Times its usual rate
“6× its usual rate” compares a word's uses in the last 7 days with its average week over the 4 weeks before. It's only shown when the word averaged at least 3 uses a week before (anything less is too thin to call usual) and is running at 1.5× or more. On Words of the Week, a spike's size is its busiest 7 days against its usual level, both as a share of everything collected.
Rising words
Each word gets a score from four independent signals, each scaled from 0 to 1:
- Usage (35%): growth in use, the last 7 days against the 4 weeks before: news from GDELT, podcasts and newsletters from our feeds.
- Lookups (25%): growth in views of its Wiktionary entry, a sign people are meeting it without knowing it.
- Novelty (25%): how new it is to written English, from its first year in books or a recently created Wiktionary entry.
- Spread (15%): how many kinds of source used it this week.
A missing signal is left out and the rest re-weighted. The score is then multiplied by a part-of-speech weight: verbs ×1.15, expressions ×1.15, adjectives ×1.1, nouns ×0.85. A rising noun is usually a news event; a rising verb or adjective is more often language changing. Words found only through lookups count a little less than words with both kinds of evidence. Names, everyday words and words from other languages are left out of the list.
Spikes
A word's usual rate is measured separately in news (the share of each day's outlets using it), podcasts and newsletters, and each day's expected uses are that rate times how much of each kind came in. A spike can only start once the word has 11 weeks of history. It starts when the word's uses over 7 days reach 2.5× what was expected (with at least 8 uses). Its usual rate is frozen at that moment, so a surge can't quietly become the new normal. The spike ends when uses have stayed under 1.5× expected for 14 days in a row (a return within those days is the same spike), or when they have settled at a new level: the last 6 weeks averaged within 1.25× of the 6 before them, with no new high. Those last weeks then become the word's usual rate. Inside a long spike, a new surge starts when uses fall below 50% of the last peak and then rise 2× off their low.
Each spike keeps the conversations that drove it as they were at the time, so an old spike reads the same later. Words that spike from the same story are linked for that spike only (“Spiked together with”), never permanently.
Peaked
A word in a spike is shown as “peaked” once its last 7 collected days fall below 60% of its best 7-day stretch since the spike began, both as a share of everything collected. Days with no collection are skipped, so a gap in our sources doesn't read as a fall.
Will it stick?
A spike sticks when it leaves the word at a new, higher level that lasts. The old level is the word's share of everything collected over the 8 weeks before the spike began (at least 14 days of collection are needed, and a word we had never seen is treated as half a use a week).
- Too early: the spike is still on, or it ended under 4 full weeks ago.
- New baseline: 2× or more of the old level in every week for the last 8 weeks.
- Back to before: under 1.25× the old level over the last 4 weeks.
- Settling: anything in between.
How it's spreading
On a word's page, two lines compare this week with the four before: uses in our sources and views of its Wiktionary entry.
Tone is rated by software for each item, from −2 (very negative) to 2 (very positive), and averaged per day on days with at least 10 rated uses. It describes the coverage a word appears in, not the word.
Words of the Week
Each issue covers an ISO week, Monday to Sunday (UTC). Its words are the spikes that started or peaked that week, ranked by their busiest week against their usual level, new spikes first. The top word leads the issue. Only proper terms are included, and a word the headlines behind its spike use as part of a person's name is left out.
An issue is worked out live while its week is on. 3 days after the week ends, once spike detection has caught up, it's published: saved as it stands, with the version of these rules that made it, and never changed afterward. The current rules are version 1.
Meanings
“How it's used now” is written by software from the headlines and opening text of recent items that use the word, with the dictionary's senses as background only. It isn't copied from any dictionary, and it's checked for people's names before it's saved. Dictionary definitions are shown separately, as written, from Wiktionary.
Names
Outerwoven tracks language, not people. Words that are a person's name don't get pages or places in lists, and everything written by software (conversation names, meanings) is checked to be free of names. Publishers' own headlines are shown as published, credited and linked to their source.
Sources and credits
- Definitions: Wiktionary, via Kaikki.org, under CC BY-SA 4.0.
- Lookups: Wikimedia pageviews for English Wiktionary entries (CC0).
- Books: the Google Books Ngram data (CC BY 3.0), uses per million words in English books.
- News counts and conversations: English-language headlines from The GDELT Project.
- YouTube content via YouTube API Services, under the YouTube Terms of Service.
- Headlines and excerpts belong to their publishers and are always linked to the original.
Citing Outerwoven
Every phrase page and every spike has a permanent link: outerwoven.com/phrase/<word>#spike-<year>-<month>. The Share button on a phrase page copies a ready citation with the date you accessed it. Published Words of the Week issues never change, so they can be cited as they are: outerwoven.com/words/<year>-<week>. Please cite the rule version below when a figure matters to your work.
Versions
- Words of the Week rules, version 1 (October 30, 2026): ranking, the name check and publishing as described above.
- Will it stick, version 1 (October 30, 2026): replaces an earlier measure of how much of a spike's peak remained.
- Spikes, version 2 (November 2, 2026): news measured on GDELT's article index; every earlier spike recomputed.
- Spikes, version 1 (October 18, 2026): per-source usual rates, frozen at the start of a spike.
- Rising words, version 1 (October 18, 2026): four signals and part-of-speech weights.
Questions about a figure? contact@outerwoven.com. See also About.