Medication authorization letters, patient handouts, and patient triage are among top use cases of AI in DermGPT and are likely the top applications in dermatology across AI tools broadly, dermatologists told Medscape Medical News.
"A lot of dermatologists have started using AI to help them with prior-auth letters or even handouts for patients," said Steve Daveluy, MD, professor and program director of dermatology at Wayne State University in Detroit, in an interview. "Uptake is very dependent on people's general views of technology and AI. Some are nervous about touching it, and some are fully in. A lot are dabbling in it."
One key to successful use of AI, he and others said, is an upfront investment of time to learn effective prompt crafting, or what's known as prompt engineering-the art of posing a question to an AI tool that is specific enough to lead to high-quality outputs.
"It's really a learned skill," said Hadar Lev-Tov, MD, associate professor of dermatology at the University of Miami, Miami. "You need to learn how to prompt AI, otherwise it will look like garbage and you'll get frustrated."
Lev-Tov and Faranak Kamangar, MD, a dermatologist and the founder of DermGPT, spoke about prompt engineering and other AI issues at a panel discussion on integrating technology in dermatology held at the 2025 Integrative Dermatology Symposium (IDS).
"Specify everything you need, do it as if you're talking to your medical assistant," Kamangar advised at the meeting. Specify demographics, provide clinical context, define a format, and set a technical level based on your audience, for instance. "Give it some rules...and ask for citations" if needed.
"For a patient handout, for instance, your prompt might say 'This is a dermatologist's office. Give me an 8th-grade-level handout on Retin-A, how to use it, when to expect side effects, and when not to use it," she added in a later interview.
Or, for patient triage -- to assist an overburdened staff or lower-level non-RN staff -- a prompt might read, "You are a dermatologist nurse. You have received the following three requests from patients and want to know who would need priority to be seen. This is for triage. The three requests are then listed verbatim."
"And if you're triaging for more serious issues, you can ask that [your AI tool] inform you if there is [a report of] bleeding, for instance, or blisters on the hands or mouth," Kamangar, chair of dermatology at the Palo Alto Medical Foundation, told Medscape.
An example of DermGPT output for a three-person triage request shown at the IDS meeting summarized each patient's condition, rated urgency, and presented the rationale for the rating in bullet points. For a case of a patient who reported severe pain (10/10) in the armpit due to a flare of hidradenitis suppurativa, despite antibiotics that had been prescribed for an emergency, for instance, AI rated urgency as "high" and provided this rationale:
* The patient is experiencing severe pain, indicating a possible acute exacerbation or infection.
* HS can lead to significant complications if not treated promptly, including abscess formation and systemic infection.
* The request for an injection suggests the need for immediate intervention to manage pain and inflammation.
(As part of her message, the patient had asked, "Can I come for an injection?")
Proofread or Disclose Generative AI Use, California Law Says
AI assistance with patient handouts can increase the quality of information provided to patients by increasing thoroughness, personalization, and conciseness, dermatologists told Medscape.
Kamangar's AI-generated patient handouts are "better than anything I would have written," and "easier to proofread than to create fresh," she said at the meeting, noting that one's prompt history can be saved in most generative AI models, and that handouts can be produced in different languages, including in "Minion-speak" (from Despicable Me) for children.
Similarly, AI-assisted medication authorization notes can be more thorough, which is important when "you're trying to do everything you can to prevent a first denial," and they can save time, trimming down the 5-plus hours of time spent on the electronic health record per 8 hours of scheduled patient time, Kamangar maintains.
Per a California law that took effect in January 2025, Kamangar said that healthcare professionals are required to disclose when communication is generated by AI, unless, for example, she or a fellow physician or nurse has read and reviewed it.
Several other states have AI laws that similarly include disclosure requirements, particularly with regards to patient communication, and medical organizations such as the American Medical Association have issued policies or position papers that include calls for transparent disclosure.
In general, Kamangar told Medscape Medical News, a "trust but verify" mentality is currently needed with all AI models. "We read everything verbatim before it goes out to a patient," she said, and medication authorization communication is routinely proofread.
The Role of GPT Systems, Other AI tools
DermGPT utilizes a large language model that queries dermatology-specific peer-reviewed articles and other published texts from curated sources before querying other information. The system's top use case, just above prior authorizations, patient handouts, and triage, involves clinical decision-making for tough cases -- specifically, use of a "second consult agent" on the system to review differential diagnoses, Kamangar told Medscape Medical News.
In a small study of DermGPT and ChatGPT published in 2025, dermatologists utilized both models to answer a set list of questions but were blinded to which model produced which response. Overall, DermGPT's answers were preferred for clarity and conciseness, while ChatGPT's answers were preferred for their source citations.
Manuel Valdebran, MD, a dermatologist with the Medical University of South Carolina, Charleston, South Carolina, who participated in the technology panel discussion at the IDS meeting, uses Doximity GPT (DoxGPT), another generative AI tool designed for clinicians, to edit Assessments and Plans for complex diagnoses and to generate some after-visit summaries and letters regarding specific school- or work-related issues. The tool can also help with coding, he said.
(Valdebran doesn't use a GPT system for most prior authorizations as his practice has a pharmacist team handling these tasks, but he said he agrees that GPTs can be valuable time-saving tools for generating prior authorization letters.)
For answering specific clinical questions, GPT models (including DermGPT and DoxGPT) "aren't always up to date" in his experience. "They don't [always] include the latest articles or are not able to pull out the most relevant articles or guidelines for specific questions," he said, "such as recommendations in pregnancy and lactation."
GPT models are not currently integrated into electronic medical record (EMR) systems -- a fact that necessitates the use of different logins and time for prompt creation -- but EMRs are integrating AI in other ways. For instance, Valdebran uses DAX Copilot, an AI clinical documentation tool embedded into the Epic EMR that uses ambient speech recognition to create clinical notes.
The tool is helpful for documenting history, but "it struggles with physical exam and assessment and plan sections, particularly [with regard to] full body skin examinations," he said. Editing and correcting is required overall, and "significant" editing is required for skin cancer checks.
Overall, "it may save time and does help patient interaction," he said. But thus far, a "good human scribe trumps AI."
And what's on Daveluy's AI menu? "I use ChatGPT, Claude...and Open Evidence, which is focused on medical literature," he said. "I use AI a lot for prior authorization letters, and if I'm trying to learn something about specifics of insurance coverage, AI can sometimes be helpful."
Kamangar reported being the founder of DermGPT. Daveluy, Lev-Tov, and Valdebran reported having no relevant disclosures.
