Many patients now arrive at an appointment after asking an AI chatbot about their symptoms, lab results, or what to ask their doctor. About a third of U.S. adults (32%) say they used AI chatbots for health information or advice in the past year. (KFF Tracking Poll on Health Information and Trust, March 2026) For a patient with limited English proficiency, that research raises questions a language access team has to answer. What language did the patient research in? Can they reach a qualified human interpreter when the conversation turns serious? And when they leave, did they understand the plan well enough to act on it? Reaching an interpreter is the first step. Teach-back is how a care team confirms the patient understood the plan.
Those questions concern one kind of AI in particular. Patient-directed AI is AI that a patient chooses and uses on their own, outside the health system, to understand their health or prepare for care. It is different from institutional AI, the tools a hospital deploys to manage its own work. That difference was the focus of a panel at the Patient Experience Symposium in Boston, “The New Patient Experience Is Happening Without You. Now What?” The panelists included Sue Sheridan, MIM, MBA, DHL, founding member of Patients for Patient Safety; Laura Adams, senior advisor at the National Academy of Medicine; Dave deBronkart, author and patient engagement advocate; and Hugo Campos, patient advocate. The panel points below are paraphrased from that discussion.
How are patients using AI before their appointments?
The panel described patients using AI for quick information, not physical care. Some enter their symptoms to see what might be causing them, and others ask AI to explain their lab results. That preparation can pay off: patients who arrive with better questions can make the appointment more efficient. Still, the panel was clear that AI does not replace clinicians. Instead, it lets patients and caregivers take part in their own care.
KFF’s national survey points the same way. In the past year, 29% of adults used AI tools for information about their physical health and 16% for their mental health. Health care professionals (80%) and internet search engines (68%) were still more common sources. (KFF, March 2026)
Why do patients turn to AI instead of waiting for an appointment?
Clinicians remain patients’ main source, so why turn to AI at all? The panel named three problems that push patients toward AI: long waits for appointments, bias in care, and limited access to information. KFF’s data supports the access point. Among adults who used AI for health information, 18% said a major reason was that they had no regular provider or could not get an appointment, and 19% said they could not afford to see a provider. (KFF, March 2026)
Whatever the reason, a patient who looks things up is engaged, and the panel advised clinicians to welcome that engagement.
What does patient AI use mean for people with limited English proficiency?
Access is harder still for people with limited English proficiency. Nearly 29 million people in the U.S. age five and older speak English less than “very well,” about 9% of that population. (U.S. Census Bureau, 2024 American Community Survey 1-Year Estimates, table S1601)
The panel asked AI developers to meet people where they are, including their reading level. Language belongs in the same request. A patient may research in Spanish, Vietnamese, or Somali, then sit across from a care team that works in English. Culture shapes that research too. Which questions a patient types into an AI tool, and which answers they believe, depend on whom their community trusts and how it talks about illness. That is one reason teach-back matters: it shows the care team whether its explanation made sense in those terms. The research helps the patient only if the visit lets them use it, and for many patients, that means a qualified interpreter in person, on the phone, or on video.
National polling leaves a gap here. KFF fielded its survey in English and Spanish, so it does not show how patients who speak other languages use AI. Your own patients can fill that gap. Ask about it at intake, with an interpreter when the patient needs one.
What should language access teams do? Five steps.
- Welcome the research. Start before an interpreter is involved: coach care teams to listen with humility when a patient brings AI-generated questions. The panel’s advice was to treat engaged patients as partners, never as if they were foolish for looking.
- Help patients search better. Some of that research will be wrong. When a patient arrives with incorrect information, help them improve how they search instead of dismissing the search. Point them to translated materials your organization stands behind, in their language. Share those materials through someone the patient already trusts, such as a member of the care team or a community health worker, so they come with a conversation, not as a handout alone.
- Make the human interpreter option visible. Good materials still leave some questions for a person. Every patient should have a clear way to request a qualified human interpreter in their language when scheduling, checking in, and in the patient portal. A patient who researched with AI should still know a person is available.
- Route each call by conversation type. Not every call needs the same kind of interpreter. AI interpretation can handle routine calls. Credentialed human interpreters should handle consent discussions, new diagnoses, and other high-stakes conversations. Decide which conversation types go where before you deploy AI interpretation and write the decision down.
- Check understanding with teach-back. Even the right interpreter does not guarantee understanding. AHRQ describes teach-back as asking patients “to state in their own words what they need to know or do about their health.” (AHRQ, Use the Teach-Back Method: Tool 5) Use it in every interpreted conversation, whether a human interpreter or AI interpretation is in use. Ask about the AI research too: if the plan differs from what the patient read before the visit, teach-back is where that shows up.
How can language access teams tell the program is working?
To see whether these steps are working, track three measures each quarter, by language, by conversation type, and by human or AI interpretation, so the routing decision in step 4 rests on your own results:
- Teach-Back Results: The share of interpreted visits where the patient can explain the plan back in their own words
- Interpreter Requests at Scheduling: How many requests come in before the visit, compared with requests made once the patient is already in the room
- Follow-Through: Kept follow-up appointments and other next steps after interpreted visits
How should health systems hold AI interpretation accountable?
Steps 4 and 5 only work if the AI interpretation tool can be checked and can hand a call to a person. Before deploying AI interpretation, ask any vendor four questions: which language pairs are validated, how high-stakes conversations reach a human interpreter, and who reviews quality. As one example, here is how Piedmont Global handles these questions in Connexus™ today:
- Languages: Spanish-English is the first language pair.
- Human Review: Piedmont Global’s own interpreters tested and reviewed the service.
- Routing: Each organization configures which call types go to credentialed human interpreters and which go to the AI Interpreter.
- Human Interpretation: For every other language, and for any patient who asks for a person, Connexus covers 300+ languages and dialects over the phone, by video, and on-site, including American Sign Language (ASL) by video and on-site.
Whatever tools a health system chooses, patients will keep asking AI before they walk in. After they walk out, the measure is whether they understood the plan and followed it. A language access program that plans for patient AI use and checks understanding keeps patients talking to their care team. A program that ignores it leaves the hardest part of the conversation to a tool the patient found alone.