AI translation: A hidden liability for pharmacies
Across the U.S., pharmacists are quietly using ChatGPT and Google Translate to serve patients with limited English proficiency (LEP), often with a liability waiver pushed onto the pharmacy.
This isn’t a small-scale issue: more than 8% of the U.S. population is considered LEP, meaning that some 26 million Americans over the age of five may struggle to read a prescription label in English.
Pharmacy staff are already overloaded. As wait times for prescriptions increase, it’s tempting to accelerate with AI. But when AI translates the prescription label, who protects the patient?
Prescription label translation is a critical patient safety issue, and a regulatory wave is coming. Multiple U.S. states now require translated prescription labels; but for any pharmacy leader with a significant LEP population, it’s already a crisis hiding in plain sight. With approximately 22% of U.S. hospitalizations caused by medication non-adherence (taking a medication incorrectly or not taking it at all), translated prescription labels are a patient safety concern for every pharmacy.
The impact on LEP patients
In a recent interview on the podcast HITea with Grace, Language Scientific/RxTran CEO Sharon Blank identified three key risks when the prescription journey breaks down:
- LEP patients are afraid to take their medication at all. When a patient gets home and can’t understand a prescription label, they’re unlikely to call the pharmacy with questions. More likely, they simply don’t take the medication at all, leading to a higher chance of complications or hospital readmission.
- LEP patients try to take their medication, but misunderstand the instructions. Many medications are intended to be taken with food, without food, at a specific time, or according to a specific interval. Faced with English-only instructions, an LEP patient may misinterpret these specifics and take their medication at the wrong time or in the wrong way.
- LEP patients may ask a child, relative, or friend to help interpret their prescription instructions. This is a scary reality, as these individuals, particularly children, are not qualified medical translators. They may not understand clinical abbreviations or may underestimate the seriousness of instructions like “Never drink alcohol while taking this medication.”
The prescription is safe…only if the patient can read it
When patients can’t read English labels, an otherwise correct prescription becomes a preventable patient-safety risk.
In community feedback related to the state of Oregon’s dual-language prescription labeling laws, a Korean-speaking doctor and public health professor testified that his own sister had been incorrectly medicated by his LEP parents, and that one of his LEP patients landed in the emergency room after misunderstanding a label and taking a dangerously high dose of blood pressure medication.
“The persistent gap is that language access has largely been treated as a compliance requirement rather than a clinical safety mechanism,” noted Sharon Blank in a recent interview with Healthcare IT News. A time-consuming and expensive hospitalization can ultimately fail in a very simple way: because the focus is on compliance rather than patient safety.
The patient-safety risks of AI translation, and a better approach
The challenges facing pharmacies are real. The National Community Pharmacists Association reports that 70% of pharmacies are struggling to fill crucial roles, and enrollment in pharmacy schools dropped by more than 60% between 2011 and 2021.
Yet, free translation engines aren’t the answer. Let’s take a look at five key risks associated with AI translation of prescription labels:
| AI translation risk | What can go wrong | Potential impact on the patient | Safer approach |
| Incorrect dosage instructions | AI engine mistranslates quantity, frequency, or timing | Patient takes too much or too little medication; health consequences or ineffective treatment | Use clinically validated translations reviewed by bilingual experts |
| Clinical context is missing | AI engine cannot reliably interpret abbreviations (“PRN”) or specific terminology (“dropperful”) | Unknown terms left in English; confusing or contradictory instructions | Use a translation system designed specifically for prescription labels, not a translation tool trained on general material |
| Translation output is fluent but incorrect | Translation sounds convincing, even when it contains errors | Patients may follow dangerous instructions, assuming they’re correct | Have a qualified human review the translation before it reaches the patient |
| Zero accountability | Free, public AI engines generally provide no quality guarantee or protection against clinical liabilities | Liability is pushed onto the pharmacy or the patient | Use an auditable, insurance-backed translation solution |
| Disclosure of confidential information | Public AI engines generally do not comply with HIPAA or other healthcare privacy laws | Patient confidentiality and regulatory requirements may be violated | Use a secure system that integrates with existing pharmacy management software |
“Non-adherence”: A language-access failure?
When patients don’t take their medication correctly, the problem is often framed as non-compliance: the patient knows how to take the medication, but chooses not to, or “forgets” to take it.
This is a real issue; a recent review found that 44-75% of patients with complex medical conditions are classified as non-adherent, resulting in massive direct costs (hospitalization, wasted medications) and indirect costs such as lost work time.
What’s being ignored? Patients who can’t read or understand their prescription instructions are often “non-adherent,” through no fault of their own.
LEP patients are readmitted to the hospital at a much higher rate than fluent English speakers, often because they are too afraid to take medication based on instructions they can’t understand, or because they misunderstand the instructions and take the medication incorrectly.
Many pharmacies see language access as a compliance issue (“Are we legally required to translate this?”) or a customer-service benefit (“Can we increase our business by translating this?”). In reality, it’s a clinical safety issue; one that cannot be solved by AI alone.
Patients, families and pharmacists shouldn’t have to guess
When a validated translation isn’t available, where do patients turn? In the pharmacy, an employee under pressure may reach for ChatGPT, Google Translate, or another free tool, in an effort to be helpful. Given the pharmacy labor shortages cited above, this workaround is understandable…and completely inappropriate.
As shown in the risk table above, AI translations, especially those done with free, public AI engines, are simply too risky for a pharmacy setting. Automated translation has come a long way, but AI never admits that it doesn’t know an answer: it guesses. Patient safety should come from a validated clinical tool, not a “best guess.”
Once the patient leaves the pharmacy, the risks pile up, when the LEP individual calls on a child, relative, or friend to help interpret their medication instructions. A 2023 commentary by the Children’s Hospital Association noted, “Compared to certified medical interpreters, ad hoc interpreters commit twice as many errors. …asking children to serve as ad hoc interpreters for patients with LEP is both harmful to the patient’s clinical course and unfair to the child.”
Additionally, these risks are unnecessary. Validated, clinically appropriate prescription label translation solutions, such as RxTran, integrate with existing pharmacy management software and are simple and affordable to implement.
Safe prescription translation is more than word substitution
AI has the potential to transform some areas of healthcare. A recent paper in the Interactive Journal of Medical Research highlights some of the risks and benefits. When used correctly, AI may:
- Result in better data-driven healthcare decisions
- Help to predict a patient’s risk of certain diseases
- Identify risks that result from multiple factors such as medication interactions
At the same time:
- AI diagnostic tools do not always outpace the diagnostic skills of a good clinician
- Data sharing and confidentiality is a huge issue
- AI cannot be held accountable for diagnostic mistakes
Generating a translation and validating its clinical accuracy and appropriateness are two different things. AI can help, but it’s far from the end of the story. A safer model looks like:
- Any translation (whether automated or done by a human) for use in a pharmacy must be reviewed by a qualified bilingual pharmacist or medical translator
- The validated translation must then be integrated directly into the pharmacy workflow and printed directly on the prescription label, not a separate sheet of paper, and definitely not scribbled on a pharmacy note pad
- The goal is not to reject technology, but to keep patients safe by ensuring that prescription labels accurately communicate the prescriber’s intent before the labels reach the patient
When it comes to translated prescription labels, communication breakdowns lead to non-adherence, and unreviewed AI translation leads to unsafe workarounds. Validated, insurance-backed prescription label translation tools like RxTran offer a clinically safe workflow that protects pharmacies from liability and keeps patient safety at the forefront.
