Increasingly, customers are asking me to revise completed translations based on AI feedback. It’s understandable to want a second opinion, especially when a tool offers a quick, confident explanation. But how can you tell whether it has found a genuine error, suggested a stylistic preference, or introduced a new problem?
To explore that question, I asked AI to review four English-to-Spanish informed-consent forms. In our usual process, one professional translator prepares the first draft, and two more professional translators review it before delivery. For this informal experiment, I used the initial drafts. At that stage, they could still contain typos, awkward phrasing, or other first-pass issues.
The results were mixed. AI caught some basic editing issues, but it also missed errors and made questionable recommendations. This was an informal experiment, not a systematic evaluation of AI tools, but it raised practical questions about how customers should assess the feedback.
A helpful suggestion isn’t always the right correction
A polished explanation can make a proposed change sound convincing. But sounding convincing doesn’t mean the change preserves the intended meaning. A suggested revision might introduce ambiguity, while the review itself might overlook an existing error. In patient-facing documents, that matters: medical translation needs to communicate complex information accurately and in language patients can understand.
Clearer wording or more confusion?
While reviewing catheter-placement instructions, AI suggested wording that didn’t clearly preserve the anatomical relationship between the catheter and the veins. When I challenged it, the AI acknowledged that it had missed the logical connection.
The issue wasn’t simply whether the Spanish sounded natural. The instructions needed to explain the procedure accurately and make clear which anatomical structure or device they were referring to.
If you can’t independently assess the Spanish and the medical context, a smoother-sounding alternative may look like an improvement even when it makes the explanation less precise.
A confident answer can still get the facts wrong
In another exchange, I asked whether “ortesis” needed an accent mark. AI confidently answered that only “órtesis” was correct and attributed that rule to the Royal Spanish Academy. In fact, the Academy accepts both forms. The issue wasn’t a missing accent; it was a confident answer that treated an accepted spelling as an error.
A customer relying on that explanation could reasonably think the translation was wrong when it wasn’t. Real Academia Española
A detailed review can still miss an omission
AI also failed to flag a sentence in which the verb "es" (is) was missing. This was a straightforward omission.
That’s worth keeping in mind: detailed feedback doesn’t mean the review is complete. An AI suggestion shouldn’t automatically be accepted, and the absence of a warning shouldn’t be taken as confirmation that everything is correct.
An apology isn’t the same as verification
When I challenged its recommendations, AI readily apologized and promised to be more rigorous. But its summary of what it had ‘learned’ repeated questionable guidance. Accepting a correction or apologizing for an answer wasn’t the same as verifying it.
A confident recommendation and a reassuring apology don’t show that the issue has actually been checked. The key question isn’t just whether AI can give feedback. It’s whether that feedback can be supported by the source text, linguistic guidance, and clinical context.
Questions to ask before approving a change
These concerns aren’t unique to AI. I have also found serious errors in translation samples submitted for professional screening. Feedback from bilingual staff needs careful assessment as well. Human involvement alone doesn’t guarantee quality.
Questions about a translation are always welcome. Professional medical translators should be able to explain their choices, and genuine errors should be corrected. But suggested revisions deserve the same careful look as the translation itself.
Before approving a change, ask:
- What exactly is the concern: an error, an unclear passage, or a stylistic preference?
- How does the proposed revision compare with the English source?
- Does it preserve the clinical meaning while making the Spanish clearer?
- What evidence supports the recommendation?
If you’re unsure whether to accept a proposed change, ask a qualified medical-translation reviewer to assess it. If a substantive disagreement remains, an independent qualified review can help resolve it.
AI can help, but professional judgment matters
I’m not against AI. I use it often to design my dream home or come up with a great stroganoff recipe. But in my work as a medical translator, I need more than a confident answer. In healthcare, clear communication can affect patient safety, and an incorrect recommendation can have serious consequences.
My experiment showed that AI can help with editing, but its feedback needs to be evaluated rather than accepted automatically. The practical rule is simple: treat AI comments as questions to investigate, not corrections to implement.
The goal isn’t to choose between AI and people on principle. It’s to use technology responsibly, with qualified professionals accountable for the final translation. When patient understanding and safety are at stake, a suggestion should earn its status as a correction.
Can AI Review Medical Translations? What Customers Should Now