MedEdMentor Mondays, and the question I get asked most frequently

If you have ever typed a question about your teaching or your research into ChatGPT or Claude and gotten back something confident, fluent, and not quite right for our field, this post is for you.

Last month (August 2026, as a reference as this post ages), I was at the AMEE conference in Vienna, Austria. I spoke to lots of people curious about MedEdMentor. The question I got asked more than any other was, "how is Uni different from ChatGPT or Claude?" It is a fair question, and I have wanted to write out a proper answer for a while now. It is also a good place to start something new.

What starts today

Every Monday from now on, there will be one new thing here on the blog. Some weeks it will be an idea, like this one. Other weeks it will be a video from our new YouTube series, UniCorner, where I show how I actually use Uni, our AI Mentor, to do my own work. And some weeks it will be a workshop, meaning a facilitator guide and a slide deck that you can download and run with your own people, for free.

We are calling it MedEdMentor Mondays, because who doesn't love alliteration? Honestly, probably a lot of people, but sometimes I am a simple human, and maybe it will help you to remember to check our website for new content!

The answer to the question

The short version I give people in person is that we are surrounding a commercial LLM with field-specific knowledge and skills and giving it specific instructions to act like a mentor and coach.

Uni runs on the same class of model that powers the tools you already use, like ChatGPT and Claude. As of today, Claude Sonnet is the model behind Uni, but that will change over time as models improve and costs change.

What is different is everything that surrounds the model. This includes a curated library of health professions education (HPE) literature we call the Atlas (about 119,000 papers from 151 journals), a set of skills that teach the model how experienced mentors coach people through specific tasks (designing research questions, performing various literature searches, choosing methodologies, designing surveys, writing introductions, etc.), a database of the theories our field actually uses, and, most importantly, a mentoring/coaching posture. Uni was built to mentor you rather than to produce things for you, so it won't do your work for you, even when you ask it to.

What that looks like in practice

Say you are a brand-new hospitalist, and you have noticed that your interns struggle when they have to give feedback to medical students. You mention it to your chair, and your chair says, "You should write that up." You have never written anything up, and you are not entirely sure what that even means.

If you ask a general chatbot for help, you will get a lot of material very quickly: a study design, a twelve-item survey, a literature summary, maybe even a draft introduction. Some of it will be useful, some of the references will turn out not to exist, and at the end of it you will have learned almost nothing about how to do this kind of work yourself, which was presumably the whole point of your chair's suggestion.

If you ask Uni the same thing, the first thing it does is ask you a question, and then another one. What exactly did you notice? Who was in the room? What do you think is going on? Somewhere in that conversation, it will search the Atlas and come back with something like, "There are dozens of papers on feedback in the clinical learning environment. If you only read three, start with these, in this order, and here is why."

Real conversation between Uni and a "hospitalist"

It will also help you turn your specific problem (my interns, on my ward) into a phenomenon the field recognizes (how trainees learn to give feedback under hierarchy), because that is the move experienced mentors make. In fact, we spent six years studying them to learn how they do it. What it will not do is write your survey for you; instead, it will help you figure out whether a survey is even the right tool, which (as anyone who has run a bad survey knows) is the more useful question.

That last part frustrates some people at first, because most of us are used to tools that give us exactly what we asked for (although it is often half-baked). However, having watched dozens of people work with Uni, even after just five minutes, this is the part they appreciate most. Because they didn't come to HPE scholarship to have someone/something else do their work for them. They are genuinely curious people who want to learn and grow. And they can clearly see that is what Uni does.

In fact, I ran a workshop a few weeks ago with folks from The Patricia A. Tietjen, MD, Teaching Academy at Northwell Health. I asked learners to use Uni on their own projects. During the debrief, one of the audience members said:

Deeper, not faster

MedEdMentor exists for two reasons.

The first, and the one that got Greg Ow and me started on all of this, is to democratize access to mentorship in health professions education. Greg and I were lucky enough to train at places where good mentors were down the hall. As we traveled deeper into the field, we realized that most people working in HPE do not have those resources. You can see this in our field's literature, where a small number of well-resourced institutions produce most of the scholarship and therefore guide the conversations. That is where the name Uni comes from: universal mentorship, for anyone doing this work, wherever they are.

The second reason is to help people go deeper, not faster, which is how we think about AI as a vehicle for mentorship. The general tools are built for speed, and speed is a fine thing to want, but we don't think it helps anyone become a skilled educator or scholar.

What I would love from you

You do not need to be a MedEdMentor user to read these posts. If you have a question you would like Uni to take on in a future video, or a workshop you wish existed, tell me. We want MedEdMentor to serve your needs.

Write to me at geoffstetson@mededmentor.org. Talk to you next Monday!