Here’s how it works: Apple has built a hash table of “blocked” phrases into iOS. This is not about storing plain text of words, but about matching hashes of word IDs to a set of prohibited hashes. If there is a match, the iOS keyboard will not suggest the word. And if you’re swiping, you’ll enter some other word for the swiped word or replace a previous word in your phrase with a different word that has a similar swipe pattern.
Apple has not made the list of phrases public, and the company does not reveal how it chooses what appears in it. Kauffman reverse engineered the hash table within iOS.
The sentences are mostly what you would expect. They include vulgarities, politically and socially charged phrases, racist terms, various forms of intolerance, etc. You can literally type the phrase “smoke weed,” but if you type “smoke weed…”, the keyboard’s suggestions for ending that second word will never include the obvious “weed.” Instead, you’ll get “weekend” or “weekly.” And if you’re using Apple’s QuickPath (Swipe to Type), you’ll get “smoke Wednesday” or some other word, no matter how accurate you are at swiping the word “weed” correctly.
The Swipeblocked site has a list of all the phrases banned in English in iOS 27. Warning: many of them are not safe for work and are very offensive. But some are things you’re likely to want to type or swipe, especially in a particular context (illegal immigrant, NBA surveillance, etc.), and the iOS keyboard’s refusal to complete input or suggest the next natural word is one of the reasons users are frustrated with it.
In addition to phrases being locked in the suggested words area or on swipe, there is a mechanism to prevent the autocomplete and suggested word list from suggesting individual words. The iOS dictionary labels words as vulgar, offensive, or sensitive. Swear words are not suggested until the user manually types them three times. Offensive words are never suggested or completed by swiping (although you can always type them yourself). Likewise, sensitive words are never offered as a prediction or completion, and some of them cannot be deleted.
Why is Apple doing all this? It’s probably less about controlling your speech than about avoiding lawsuits and bad PR. All it takes is for one person to post a screenshot of the iOS keyboard suggesting the word “good” after typing “Nazis were” or “real” after Santa isn’t” (both included in the list) to cause a bad viral moment criticizing Apple’s intentions, even if the keyboard suggestions would only suggest things the user had typed several times before.
