Confusable-Word Checker
Everything runs in your browser. Nothing you type is uploaded or stored.
Paste a sentence, a paragraph, or a whole draft, and this tool reads it against the 375 confusable word pairs that have their own researched page on this site — affect and effect, accepted and excepted, its and it's, complement and compliment, lay and lie, principal and principle, and the several hundred others. Where the surrounding words make the answer mechanically decidable, it tells you which word your sentence actually needs and why: a determiner in front of the slot means the word is a noun, so "a strange affect" wants "effect"; a comparative in front of "then" means the word is "than"; "its" in front of a verb can only ever be "it's". Where the answer depends on what you meant rather than on what you wrote, it says so instead of guessing, shows you the distinction, and links you to the pair's full page so you can settle it yourself. Everything happens in your browser. Nothing you paste in is uploaded, logged, or stored anywhere, which matters because the text people most want checked is usually the text they least want sent to a stranger's server.
How it works
The checker has two layers, and knowing which one produced a given flag tells you exactly how much weight to put on it. The first layer is an index of every confusable pair on the site, built at publish time from the pages themselves. When a word in your text appears in that index, the checker can always tell you that the word has a twin and quote the distinction written on that pair's own page. The second layer is a set of hand-written contextual tests that look at the words immediately around the match and try to decide the question for your specific sentence. Those tests are deliberately narrow. They look for the kinds of evidence that are actually visible in the text — a determiner, a modal verb, a comparative, a negation, a plural count noun, an infinitive marker — and they are written to stay silent rather than guess when that evidence is absent. That is why a result comes back in three bands rather than one. "Likely wrong" means a test fired that is close to watertight: "alot" is not a word in any dictionary, "it's own" expands to "it is own" and cannot be a sentence, "less people" wants the count word "fewer". "Worth checking" means the word is a known confusable and the sentence around it does not settle which one you need — the tool has found the question rather than the answer, and the pair's page is where the answer lives. "Checked, reads fine" means a test fired and confirmed the word you used, which is why those are collapsed by default: a clean result should be quiet, not a wall of green. The checker also stays deliberately quiet about the most common words in English. "To", "so", "no", "one", "for", "see" and several dozen others all belong to genuine confusable pairs, and flagging every one of them in a paragraph would bury the findings that matter under findings that do not. Those words produce a result only when a contextual test genuinely fires on them. What the tool does not do is worth stating as plainly as what it does. It is not a spell-checker, a grammar-checker, or a style-checker. It has no model of your sentence's meaning. It will not catch a mistake involving a word pair that has no page on this site, and it will not notice a missing verb, a mangled tense, or a sentence that simply does not say what you wanted it to say. It checks confusable pairs, and its usefulness comes from doing that one job transparently rather than from doing many jobs opaquely.
How to read a result you disagree with
A contextual test can be right about the pattern and wrong about your sentence, and the honest thing to do with that is to make it easy to check rather than to hide it. Every flag carries the reasoning that produced it and a link to the pair's full page, so you can read the actual distinction and decide. Two cases come up often enough to name. "Affect" really is a noun in psychology — "the patient displayed a flat affect" is correct, and the checker skips the flag when one of the clinical adjectives precedes it, but a phrasing it does not recognise will still get flagged. And "lay" is genuinely the past tense of "lie", so "I lay down for an hour yesterday" is correct even though "I lay down now" is not; the checker marks that one as worth checking rather than wrong for exactly this reason. Disagreeing with a flag after reading the rule is a perfectly good outcome. Being told you are wrong with no reasoning attached is not, which is why no flag here arrives without one.
Sharing and saving a result
The text you check is held in the page's own address, so the result is a link. Copy it and the person you send it to sees the same text with the same flags, without either of you needing an account or an upload. Very long passages are the exception — past roughly 1,400 characters the text no longer fits comfortably in a URL, and the tool says so and offers you the plain-text report instead, which you can paste into an email or a document. The report lists each finding, its reasoning, and the sentence it came from, in the same order as the page.
Frequently asked questions
Is any of my text uploaded when I use this?
No. The pair index ships with the page and the scan runs in your browser, so the text never leaves your device. That is also true of the shareable link — the text is encoded into the address in your own browser, and the link only reaches anyone else if you choose to send it.
Why does it flag a word without telling me which one is right?
Because for that sentence it genuinely cannot tell, and saying so is more useful than guessing. Plenty of confusable pairs are only distinguishable by what the writer meant — "the council sought counsel" and "the counsel sought council" are both grammatical. When the surrounding words carry no evidence either way, the tool shows you the distinction and the pair's page rather than inventing a verdict.
It flagged something that is correct. Is that a bug?
Not necessarily. The contextual tests read patterns, not meaning, and a few correct constructions match a pattern that is usually an error — the psychology noun "affect" and the past tense "lay" are the two that come up most. The reasoning behind every flag is printed with it so you can see what the tool was looking at and overrule it. If a flag is clearly wrong on ordinary English rather than on an edge case, that is worth reporting through the contact page.
Why are common words like "to" and "see" usually ignored?
They belong to real confusable pairs, but they are also among the most frequent words in the language, and flagging every one of them would produce dozens of notes per paragraph and hide the two that matter. Those words are checked, but they only produce a result when a specific contextual test fires — "to much" gets flagged, "to the shop" does not.
Does it replace a proofreader or a grammar checker?
No, and it is scoped so narrowly on purpose. It knows nothing about spelling, punctuation, tense, agreement, or style, and it only knows the word pairs that have a page on this site. It is a fast second pass over one specific category of mistake — the category that survives spellcheck, because both spellings are real words.
How many pairs does it actually know?
Every confusable pair with its own page on the site — 375 of them at the time of writing, and the number moves as pages are added, because the index is rebuilt from the pages themselves rather than maintained separately. A word with no page on the site is invisible to the checker.