Use of Artificial Intelligence in AAMC Service Programs

The AAMC is leading efforts to explore how artificial intelligence (AI) can enhance our tools and services to better support learners, medical schools, and academic health systems. By integrating AI responsibly and transparently, we aim to improve access to information, streamline processes, and deliver more personalized and efficient experiences across academic medicine.  

We are committed to openly sharing how AI is used at the AAMC, including the purpose of AI tools in our work and how it supports (but does not replace) human decision-making. While AI can help organize information and improve some steps in the medical school and residency application process, it does not replace human reviewers, program leaders, and experts making decisions about applicants. We take seriously the risk of bias in any algorithm system, and our approach includes in-depth testing and assessment to ensure fair outcomes for all applicants. 

In February 2024, the AAMC convened a technical advisory committee to create the Principles for Responsible AI in Medical School and Residency Selection. The group consisted of seven individuals with expertise in data science, industrial and organizational psychology, and residency selection. The group met several times for six months, reviewing standards from other industries and discussing the current and future state of AI in medical school and residency selection before drafting the AI Principles. These in-depth principles will be integrated into all the AAMC’s use of AI. 

Additional details on artificial intelligence and academic medicine are available in the AAMC AI Collection.  

In the American Medical College Application Service® (AMCAS®), Medical College Admission Test® (MCAT®), and PREview® program, the AAMC is exploring how AI may support, but not replace, human expertise. All core admissions services are designed and operated by people. The AAMC is committed to responsible and transparent use of AI operated by people.

Generative Artificial Intelligence (genAI) has the potential to support your letter-writing process, including crafting initial drafts of the LOR and editing its content and tone. Regardless of how you use genAI, you remain the author of the letter and are responsible for its content, accuracy, and the assessment it conveys. Although genAI may reduce the effort and time required to produce LORs, consider the following critical factors when deciding whether and how to integrate genAI in your process. Additionally, consult your institution’s policies on AI use, as many institutions are developing governance frameworks for responsible AI adoption:

• Human Oversight: The core value of a letter of recommendation is your authentic, firsthand assessment of the applicant. GenAI should support your writing process, not replace your professional judgment.
  • Do not rely on genAI to form your assessment of the applicant. Use it only to articulate an assessment you have already developed through direct observation and interaction.
  • Ensure the final letter reflects your voice, perspective, and genuine evaluation. A letter that reads as AI-generated may undermine the credibility of your recommendation.

• Privacy: Sensitive applicant data or any information that could be used to identify the applicant should not be entered into genAI systems where data privacy is compromised. Instead, those data should be protected vigorously with robust security measures.
  • Ensure applicant data is only entered into AI tools that are institutionally approved and comply with relevant privacy regulations (e.g., FERPA).
  • Do not enter identifiable applicant information into AI tools without first confirming that the tool meets your institution’s data governance requirements.
  • Avoid including applicant data while using personal versions of web-based AI tools, because those tools often do not keep your data private.

• Fairness: The prompts and content that go into genAI systems will critically influence the nature and quality of the output. Carefully review and standardize your prompts and uploaded content.
  • Screen applicant content to ensure that all content relevant to past performance or future performance potential is included, and all content that is irrelevant is excluded.
  • Standardize content input across applicants to ensure fairness (e.g., include transcripts for all, not a subset, of your letters).
  • Carefully develop and iterate the prompts used in your process to ensure they are sufficiently comprehensive and specific.
  • Review AI-generated drafts for potential linguistic bias. GenAI may introduce differential language patterns when describing applicants from different backgrounds (e.g., agentic vs. communal descriptors). Compare drafts across applicants to ensure consistency in structure, tone, strength of language, and descriptor quality.

• Accuracy: Your role in the process is vital. GenAI can be a powerful assistant, but ultimately you drive and are responsible for the content, structure, and tone of the LOR.
  • Expect to read, edit, and iterate through multiple rounds to refine the LOR.
  • GenAI output can be contaminated with errors, lack all the facts, or be otherwise inaccurate or exaggerated. Review and edit content so it accurately captures the applicant’s past performance and future potential.
  • AI-generated text may produce overly enthusiastic and generic language, or reuse the same letter structure. Make sure the final letter reflects your own personal assessment and the unique strengths of each applicant.

For more information, see:

Principles for the Responsible Use of Artificial Intelligence in and for Medical Education

Using Generative Artificial Intelligence When Writing Letters of Recommendation

The ERAS platform itself does not use AI to analyze, sort, or evaluate applications. All application data is transmitted exactly as submitted by applicants.

At present, Thalamus uses optical character recognition (OCR) in Cortex to extract core clerkship grades from transcript PDFs and apply a normalized percentile and grade distribution within and across medical schools. Thalamus Cortex also uses AI to read personal statements and assesses certain applicant characteristics, starting with Academic Career Interest for the 2026 ERAS season.

Thalamus does not use AI to score applications; automatically filter out, search, sort, select, or reject applicants; or determine whether essays were written with AI tools (avoiding biases that disproportionately affect applicants). Thalamus AI tools continue to progress forward and evolve based on user feedback. Additional information regarding Thalamus’s AI philosophy can be found on the Thalamus website.

The AAMC has several research projects underway that utilize advanced analytics, such as Natural Language Processing (NLP) and Large Language Model (LLM), to analyze, understand, and gain insights from essays and other dense text. Additionally, we have research and development projects, being conducted in collaboration with constituents, underway to explore the viability of using advanced analytic techniques to improve predictions about where applicants will be invited to interview. If successful, these techniques could be used to help applicants identify programs where they are competitive candidates and to help programs prioritize applications so they can conduct a more in-depth review of applications.