Getty Images For Unsplash+
How A Patient Portal Message Is Written May Affect Whether The Doctor Responds
In A Nutshell
- A study of 3.6 million patient portal messages found that Black, Hispanic, and lower-income patients were significantly less likely to get a response from their own doctor.
- Writing style, not medical content, explained roughly half of that gap, including things like message length, greeting, and punctuation.
- Messages with a doctor’s last name in the greeting had a 39% response rate, compared with 26% for messages with no greeting at all.
- Researchers warn that AI triage tools trained on past data could repeat, rather than fix, this same bias.
Millions of Americans now message their doctors through online health portals, expecting a response. But a large new study finds that for patients from historically marginalized groups, including Black, Hispanic, and lower-income patients, those messages are far less likely to get a reply from the doctor they were meant for. The reason may surprise you: it’s not what patients are writing about. It’s how they’re writing it.
Researchers examined more than 3.6 million patient message threads sent through a large academic hospital system, then analyzed the wording of nearly 990,000 English-language messages in detail, publishing their results in JAMA Network Open. Writing style, things like whether a message opened with the doctor’s name or included a question mark, accounted for roughly half of the gap in how often marginalized patients heard back from their intended doctor. The actual medical content of those messages explained very little of the disparity.
Put plainly, even across similar types of medical requests, shorter or less formal messages were associated with lower rates of response from the intended clinician, regardless of what the patient was actually asking about.
A Doctor’s Last Name in the Greeting Was Tied to Higher Response Rates
Before a message ever reaches a doctor, it often passes through a triage nurse or medical assistant who decides whether it’s urgent or involved enough to forward. Most practices in the study operated this way, with messages first landing in a shared pool where a staff member makes fast calls with limited context.
How a message is phrased appears to matter here, sometimes more than patients might expect. According to the study, messages that opened with the doctor’s last name, such as Dear Dr. Smith, had a target doctor response rate of about 39%. Messages with no greeting at all had a rate closer to 26%. Messages with multiple question marks showed noticeably higher response rates than messages with none.
Shorter messages also fared worse. Messages with four to 25 words had a doctor response rate of about 21%, while messages between 101 and 200 words climbed to roughly 43%. These patterns held across different medical topics, from medication questions to symptom updates.
Black and Hispanic Patients Saw the Widest Response Gaps
Data for the study came from the Mass General Brigham health system, covering messages sent between January 2021 and December 2023. Over 511,000 unique patients were included. About 83% of message threads came from White patients, 5.9% from Hispanic patients, 4% from Asian patients, and 3.7% from Black patients.
After accounting for factors like which doctor a message was sent to, when it was sent, and a patient’s health history, clear gaps in response rates emerged. Compared with White patients, Black patients had a 3.7 percentage point lower rate of getting a response from their intended doctor, representing an 11.6% reduction from the average. Hispanic patients, Medicaid recipients, and patients with only a high school education also received lower rates of doctor-specific responses.
When researchers broke down what was driving those gaps, writing style accounted for 48% of the gap for Black patients, about 35% for Hispanic patients, roughly 61% for patients with only a high school education, and about 43% for Medicaid recipients.
Patients from historically marginalized groups tended to send shorter messages and were less likely to use the doctor’s name in a greeting or include question marks. Medicaid patients and those with lower educational attainment were also more likely to open with a negative tone, which was associated with lower doctor-specific response rates.
AI Triage Tools Risk Repeating the Same Bias
Artificial intelligence tools are increasingly being used in clinical messaging workflows. Some handle triage, some summarize messages, and others draft responses. But the researchers caution that AI isn’t a guaranteed fix. Systems trained on historical data could learn and repeat the same biases found in human triage decisions.
Solutions could come from several directions, the study’s authors suggest. AI applications designed specifically to strip away stylistic differences and surface the core clinical content of a message could be part of the answer. Patient-facing tools that help people draft clearer, better-structured messages are another possibility. Even something as low-tech as updating portal interfaces to prompt patients for specific details could help standardize how messages are written across the board.
Raising awareness among triage staff may also matter. Nurses and medical assistants sorting messages are likely unaware that the phrasing of an opening line could be influencing how they prioritize a request.
That takeaway is both simple and unsettling. In a system meant to give every patient equal access to their care team, something as ordinary as a greeting or a sentence count may be shaping who gets heard.
Disclaimer: This article summarizes findings from a peer-reviewed observational study. It does not establish that any single factor causes a doctor to respond or not respond to a specific message, and it should not be used as guidance for medical communication in place of advice from a healthcare provider.
Paper Notes
Limitations
Data came from a single academic health system, which may limit how broadly the findings apply to other types of practices with different staffing arrangements or patient populations. Text analysis was limited to English-language messages, excluding non-English messages that may carry their own distinct patterns and disparities. While researchers controlled for many potential factors, the study cannot establish definitive cause-and-effect relationships. Triage staff responding via telephone may not have been captured in the primary outcome measures, though secondary outcomes attempted to account for other types of care encounters.
Funding and Disclosures
No specific funding source for this study was identified in the provided content. One co-author reported receiving grants from the Agency for Healthcare Research and Quality, the Physicians Foundation, and the American Medical Association outside of this work. Another co-author reported receiving personal fees from the Peterson Health Technology Institute and serving as an unpaid member of the Scientific Advisory Board of the German Society for Digital Medicine outside of this work. No other conflicts of interest were disclosed.
Publication Details
Authors: Mitchell Tang, PhD; Lisa S. Rotenstein, MD, MBA, MSc; Ariel D. Stern, PhD; Rebecca G. Mishuris, MD, MPH, MS; Michael L. Barnett, MD, MS | Author affiliations include: Columbia Mailman School of Public Health; Harvard Kenneth C. Griffin Graduate School of Arts and Sciences; Brigham and Women’s Hospital; University of California, San Francisco; Hasso Plattner Institute, University of Potsdam; Icahn School of Medicine at Mount Sinai; Mass General Brigham Digital; Brown University School of Public Health. | Journal: JAMA Network Open | Paper title: Care Team Response to Patient Portal Message Content and Writing Style | DOI: 10.1001/jamanetworkopen.2026.30379 | Published: August 24, 2026
