Spellcheck vs grammar checkers vs AI proofreading: which catches wrong words on drug labels?
None of them reliably. Spellcheck misses every correctly spelled wrong word, grammar checkers catch only the ones that break the sentence, and AI proofreaders can miss a fluent wrong word or suggest one of their own. ContextRx has documented these errors on live FDA DailyMed labels, and in every case the fix needed a person who knew what the line was supposed to say.
Each of these tools is good at something. The trouble is that a contextual word error, a real word, spelled correctly, that is wrong for the sentence it sits in, falls between what they check.
The short version
| Tool type | What it checks | Catches a correctly spelled wrong word? |
|---|---|---|
| Spellcheck | Whether each word is in its dictionary | No |
| Grammar or style checker | Sentence structure, agreement, part of speech, and in some tools a list of commonly confused words | Sometimes, if the wrong word breaks the grammar |
| AI proofreader or general chat assistant | Whether the wording looks likely in context, based on patterns in the text it was trained on | Sometimes, but not reliably, and it can introduce fluent errors |
Spellcheck: right tool, wrong question
Spellcheck is fast, consistent and very good at what it does: it catches misspellings and typos that produce non-words. Medical dictionaries extend it to clinical vocabulary, which matters on labels.
But the only question it asks is whether each word exists. Every human-verified case ContextRx has published passes that test. "Attach" is a word. So is "doctor", and so is "does". A spelling tool has no way to know that a heart-risk warning needed "attack", that a first-aid line needed "water", or that a directions line needed "dose".
Grammar checkers: catch the grammar breaks, not the meaning breaks
Grammar and style checkers look at how words fit together: subject and verb, word order, which part of speech belongs in a slot. Some also carry lists of commonly confused words. That gives them a real chance at some wrong words. We have not benchmarked any product, so treat these as likely behaviour, not test results:
- does→dose, directions line. A verb sits where a noun belongs. A grammar tool might flag it, because the sentence structure is off. FDA DailyMed directions said "does" instead of "dose"
- doctor→water, first-aid line. Both are nouns, and the sentence still parses. Likely silent. FDA DailyMed wipe label said "doctor" instead of "water"
- attach→attack, heart-risk warning. Both are valid words in a grammatical sentence. Likely silent. FDA DailyMed NSAID warning said "attach" instead of "attack"
The pattern: when the swap breaks the grammar, a grammar checker has something to grab. When the swap only breaks the meaning, it usually doesn't.
AI proofreaders: better at sense, still not accountable
AI proofreaders and general chat assistants work differently. They judge whether wording looks likely in context, which means they can sometimes notice that a first-aid instruction naming a person instead of a fluid reads oddly. That is closer to reading for sense than a dictionary or grammar rule gets.
The limits are real, though. An AI tool can read past a wrong word that fits the rhythm of the sentence. It can smooth a wrong word into a confident, fluent line. And it can "fix" stock regulatory wording that is required to stay exactly as approved. It also can't be held responsible for the label. Whatever it suggests still needs a reviewer who knows what the line is for.
What a person catches that none of them do
A trained reviewer brings what none of these tools have: knowledge of the line's job. A heart-risk warning has to name the event. A first-aid line has to name the fluid. An alcohol warning has to state the limit. When the word that carries that job is missing, a person who expects it notices, especially when they read the line aloud and hear it the way a patient would. FDA DailyMed warning said "ever" instead of "per" is a clean example: every word is real, and the limit is gone.
How to use them together
Use the tools first, for what they do well. Let spellcheck clear the misspellings and let a grammar checker flag broken structure. Treat any AI suggestions as flags for a person to judge, not edits to accept, particularly on stock wording.
Then do a read-for-sense pass on the lines where one word changes what the reader does: warnings, directions, indications and first aid. Name the key word each line must contain, and read it aloud. Expert review stays the last step before release.
None of this means the tools are failing. Each one answers the question it was built for. The mistake is assuming that a clean spelling, grammar or AI pass means the label says the right thing. Those are different questions, and only a person who knows the line's job can answer the last one.
FAQ
Does a grammar checker catch wrong words on drug labels?
Only when the wrong word breaks the grammar. If the sentence still parses, it usually stays quiet.
Can AI proofreading catch correctly spelled wrong words?
Sometimes, but not reliably, and it can introduce fluent wrong words of its own. A reviewer still has to check.
Is spellcheck useless for labels?
No. It catches misspellings well. It just can't judge meaning.
Should AI tools edit regulatory stock wording?
Be careful. Required wording often has to stay exact, so treat AI suggestions as flags for a person to judge, not edits to accept.
What catches the wrong words that spellcheck, grammar tools and AI miss?
A person reading the high-stakes line for meaning, ideally aloud, knowing the key word it must contain.