Why macOS Autocorrects the Wrong Words

macOS autocorrect changes the wrong words because it uses a dictionary-matching approach with no context awareness. It compares your typed text against a built-in word list and picks the closest match, regardless of what you intended. It cannot tell the difference between "ther" meaning "there" versus "their" - it just picks whichever dictionary word is nearest. ML-based tools like Charm use sentence context to make far more accurate decisions.

Why macOS autocorrect has no context awareness

The system macOS uses for autocorrect is built around NSSpellChecker, Apple's spell-checking framework. NSSpellChecker compares each word you type against a static dictionary. If your input is within a small edit distance of a known word, it substitutes that word. This approach has not changed significantly since around 2010.

The problem is that edit distance tells you nothing about meaning. Consider the sentence: "I went ther." The word "ther" is one character away from both "there" and "their." macOS will pick one based on which is closer alphabetically or more frequent in its word list - not based on what the sentence means or what you were trying to say.

This is the there/their/they're problem in miniature. Homophones and near-homophones are indistinguishable to a dictionary-based corrector. It has no model of grammar, no understanding of what came before the word, and no sense of what a sentence is trying to communicate. It is pattern-matching, not comprehension.

The result is a false positive rate of approximately 12%. That means roughly one in eight corrections macOS makes is a word you typed correctly and intentionally, changed to something you did not want.

Why macOS overcorrects proper nouns and technical terms

The dictionary that NSSpellChecker uses is built from common English words. It does not include most proper nouns - people's names, brand names, product names, programming terms, medical terminology, or any domain-specific vocabulary that does not appear in a general-purpose word list.

When you type a name like "Ankur" or a company name like "Vercel" or a technical term like "useState," macOS sees a string that is not in its dictionary. It then searches for the dictionary word closest to what you typed. If your name is one or two characters away from a real word, macOS will substitute that word every time you type it.

This is particularly frustrating for developers and anyone who regularly types product or people names that happen to be near-misses for common words. The corrector is not making a judgment about meaning - it is purely measuring character distance.

The practical fix for individual terms is to right-click the word while it is highlighted and select Learn Spelling. This adds the word to your personal dictionary at ~/Library/Spelling/LocalDictionary. For a large list of terms, you can edit that file directly in a text editor, one word per line, to add everything at once.

A secondary source of unexpected substitutions is text replacements. If you have entries in System Settings under Keyboard and Text Replacements, those fire before autocorrect runs. A replacement that partially overlaps with a word you are typing can produce substitutions that look like autocorrect errors but are actually text replacement conflicts.

Why Charm is more accurate

Charm uses a machine learning model rather than dictionary distance to evaluate corrections. Instead of asking "is this word close to a known word?", it asks "given everything in this sentence, is this word plausible, and if not, what is the most likely intended word?"

That shift from local pattern-matching to contextual inference is what closes the accuracy gap. In the sentence "I went ther store," a context-aware model recognises that the slot requires "the" or "there" based on surrounding grammar, not just which word has the smallest edit distance from "ther." It considers position, surrounding words, and sentence structure simultaneously.

The practical result is a false positive rate below 3% - compared to macOS's approximately 12%. Charm changes something you typed correctly less than one-third as often as macOS's built-in corrector does. For writers, developers, and anyone who types a lot, that difference is noticeable in daily use.

Charm also learns your personal vocabulary over time. Names, technical terms, and domain-specific words you use regularly get weighted appropriately rather than constantly triggering dictionary substitutions.

How to reduce wrong autocorrections on Mac

There are four practical steps you can take today, in order of effort.

Step 1: Add incorrectly corrected words to your personal dictionary. Right-click any word that macOS keeps changing and select Learn Spelling. This is the fastest fix for individual terms and works immediately.

Step 2: Bulk-add proper nouns via LocalDictionary. Open ~/Library/Spelling/LocalDictionary in any text editor and add words one per line. This is the fastest approach when you have a long list of names or technical terms to protect from correction.

Step 3: Install Charm for ML-based context-aware correction. Charm replaces macOS's dictionary approach with a model that understands sentence context. It works across every app on your Mac and drops the false positive rate from 12% to under 3%.

Step 4: Configure Charm's personal dictionary settings. Charm includes per-app settings and its own personal dictionary. You can mark specific apps where you want minimal intervention - for example, a coding editor where technical strings should never be touched.

Frequently asked questions

Why does macOS keep changing words I typed correctly?

macOS autocorrect uses a dictionary-based approach with no context awareness. It compares your typed text against a built-in word list and substitutes whenever your input is close to a known word, even if what you typed was intentional. It cannot tell the difference between a deliberate choice and a typo.

How do I stop autocorrect from changing a specific word?

Right-click the word while it is underlined or just after it has been changed, then select Learn Spelling from the context menu. macOS adds it to your personal dictionary and stops correcting it in future. You can also edit ~/Library/Spelling/LocalDictionary directly to add multiple words at once.

Is there a smarter autocorrect for Mac?

Yes. Charm is an ML-based writing assistant for Mac that uses sentence context rather than dictionary distance to decide whether and how to correct a word. Its false positive rate is below 3%, compared to approximately 12% for macOS's built-in autocorrect. Charm works across every app on your Mac for a one-time $9.99 payment.

Why does macOS autocorrect names incorrectly?

macOS autocorrect checks each word against its dictionary. Proper nouns such as people's names, company names, and technical terms are usually not in that dictionary. When macOS finds a name that is only a short edit away from a real dictionary word, it substitutes the dictionary word. Adding the name via right-click and Learn Spelling prevents future corrections.