Sonja: Thank you for sharing your human-at-the-core process at Fairview. You said it wonderfully: "Maintaining a human-centered review process is one important way we can uphold both patient safety and trust while responsibly exploring new tools."
I'd add that review processes work best when they're grounded in something more structured: a shared, principled human-centered framework for how teams create with AI, use AI, and evaluate what it produces. Where organizations have invested in training their teams on a common approach, I see meaningfully better consistency, fewer errors reaching patients, and standards (both internal and regulatory) being met more reliably. That's because the framework creates shared mental models across teams rather than leaving each person to figure it out individually.
On AI translation specifically: it has real promise and has improved significantly. But, accuracy varies considerably across languages. Languages like Tagalog, Hindi, and Haitian Creole, all in the top 10 in the US by number of speakers after English and Spanish, are what researchers call "low resource languages" and tend to fare worse in standard LLMs. AI translation is not transcreation.
A few practical strategies that have worked well for organizations we've supported: identifying which content poses the highest risk if mistranslated, understanding which languages and what kinds of content surface which kinds of errors, and developing human-curated dictionaries of key terms and phrases that can be provided to LLMs at the time of translation to positively influence outputs.
One area worth paying close attention to: what happens when AI translation moves from static content to the point of care? These are situations where clinicians or staff might use AI translation during a patient interaction or where, for example, IT enables AI translation across your website, including content that was intentionally developed to meet health literacy standards. The human review process that works well for documents doesn't easily transfer to real-time use. That's a different challenge that also requires a human-centered approach to AI. It often involves vendor evaluation, technology decisions, and policy frameworks that may sit outside your team's direct control but that your team should absolutely have a voice in shaping. A shared, structured, human-centered framework for approaching AI makes a real difference in situations like this, not just for static content, but across the organization.
Thank you, Connie, for sharing the PubMed article. It's an important resource for this conversation.
If anyone is working through how to build a structured approach to AI for content creation, translation, or evaluation in your organization, I'm always happy to talk through what we're seeing. Feel free to reach out.
Best,
Temese
------------------------------
Temese Szalai
Principal
Subtextivetemese@subtextive.com------------------------------
Original Message:
Sent: 04-07-2026 01:19 PM
From: Sonja Casey Lara
Subject: AI translation
Thank you for sharing this resource. At our organization, we also follow a human-in-the-loop-or what I often refer to as a human-at-the-core-approach to translation. This practice supports regulatory requirements, but more importantly, it helps ensure that translated content is accurate, high quality, and truly supports understanding, trust, and positive experiences within the communities we serve.
AI-powered translation has made meaningful advances, and it can be a useful tool when applied thoughtfully. That said, performance varies widely across languages and subject areas, and for many languages there remain well-documented gaps in accuracy. Without qualified human review and approval, there is a real risk of introducing errors or bias into patient-facing content-risks that can undermine health literacy and, in some cases, cause harm.
For that reason, we apply the same standards to AI translation that we do to any other drafting tool: if we would not publish AI created English content without review, we should not rely on machine translation outputs without human validation by a trained professional. Ultimately, accountability for patient communication rests with clinicians and organizations, not with the technology itself. Maintaining a human-centered review process is one important way we can uphold both patient safety and trust while responsibly exploring new tools.
------------------------------
Sonja Casey Lara
Translation Program Manager
Fairview Health Services
sonja.caseylara@fairview.org
Original Message:
Sent: 04-06-2026 07:33 AM
From: Connie Feiler
Subject: AI translation
Hi Cori,
I was searching for regulatory guidance on using AI to translate documents. I found this pubmed article referring to HHS recommendations:
https://pmc.ncbi.nlm.nih.gov/articles/PMC12918658/#:~:text=Level%202:%20formal%20regulations,is%20given%20to%20a%20patient.&text='%20This%20creates%20a%20regulatory%20'human,efficiency%20gains%20of%20AI%20deployment.
Here's the section that was relevant:
Level 2: formal regulations from HHS:
At the second level are formal regulations, which implement statutory requirements through detailed rulemaking. The 2024 HHS Final Rule implementing Section 1557 provided the first explicit guidance on machine translation (referred to here as AI translation). This regulation introduced a critical distinction between 'critical' and 'non-critical' documents. For critical documents-a category that includes consent forms, discharge instructions and notices of eligibility18-the regulation mandates that 'a qualified human translator must review and correct the content before it is given to a patient.19' This creates a regulatory 'human-in-the-loop' floor for high-stakes clinical communication. However, this regulation was developed based on the performance capabilities of older model generations and defaults to a manual review process that may negate the efficiency gains of AI deployment.
Nursing Consultant
Wolters Kluwer Patient Education
------------------------------
Connie Feiler
Nursing Consultant
Wolters Kluwer
Pittsburgh, PA United States
Connie.Feiler@wolterskluwer.com
------------------------------