Can an AI proofreader put a wrong word into a drug label?
Yes, it can. An AI proofreader generates text rather than looking words up, so a suggested fix can itself be a correctly spelled wrong word. The US National Institute of Standards and Technology (NIST) names this risk confabulation, and warns that people may over-rely on AI output. That's why a person still has to read the final line for meaning.
Why a fix can be wrong
NIST's generative AI risk profile (NIST AI 600-1, July 2024) defines confabulation as the production of confidently stated but erroneous or false content. It explains that generative models produce outputs that approximate the statistical distribution of their training data, for example by predicting the next word. Usually that gives the right word. Sometimes it gives a plausible one that's wrong for this sentence, and on a label a plausible wrong word is exactly the error spellcheck can't see.
Why people miss it
The same NIST profile describes automation bias as excessive deference to automated systems, and notes that people may unjustifiably see AI content as higher quality than content from other sources. Our reasoning: a reviewer who accepts a run of good suggestions may be primed to accept the next one.
What each kind of tool can and can't do
- Spellcheck flags strings that aren't words. It can't flag a real word in the wrong place. See does spellcheck catch wrong words.
- Grammar checkers flag broken grammar. A wrong word that keeps the grammar intact gets through.
- AI proofreaders can notice that a sentence doesn't make sense. They can also change words, and their changes need checking too.
The side-by-side is in spellcheck vs grammar vs AI proofreading.
Practice advice (our view)
Treat every accepted AI suggestion on a warning, direction or dose line as a change that needs a read for meaning. Compare the final text with the approved source, not with the AI's last version. See is a tool enough for labeling QC.
Sources
Last reviewed: October 14, 2026
More on contextual errors in FDA DailyMed labels.
Questions about wording on your own labels? Talk to us.
We scan freshly published FDA DailyMed labels every day and pull out these high-stakes wrong words.
Every Tuesday and Friday, a new case like this lands in your inbox if you subscribe.
FAQ
Can AI proofreading add errors to a label?
Yes. A generative tool can suggest a plausible but wrong word, which NIST calls confabulation, and a correctly spelled wrong word passes spellcheck.
What is automation bias?
NIST describes it as excessive deference to automated systems. Accepting AI suggestions without checking them is one example.
Do AI proofreaders replace reading for meaning?
No. They can flag some sentences that don't make sense, but their own changes still need a person to read the final line for meaning.