One of the most interesting features announced with ChatGPT for Clinicians was not another AI model, another medical search engine, or another documentation tool.

It was CME.

More specifically, OpenAI proposed something that makes enormous sense for practicing physicians: turn the clinical learning we are already doing into continuing medical education credit.

The concept is excellent.

The implementation, at least in my experience, is not.

And after trying to get the feature to work, I am beginning to wonder whether CME was launched as a mature clinician feature or primarily as an attractive bullet point for the April 2026 rollout.

CME Without the Artificial Separation Between Learning and Practice

Physicians learn continuously.

A patient walks into the office with an unusual presentation. We look something up.

A medication raises a question about lactation, renal function, drug interactions, or an adverse effect. We investigate it.

An ECG raises a question about electrophysiology. We review the mechanism.

A guideline has changed. We check the current recommendation.

A difficult occupational medicine case requires reviewing the literature, regulatory standards, imaging findings, latency, epidemiology, or causation evidence.

That is medical education.

Yet the traditional CME system often separates "education" from the actual process of practicing medicine. A physician may spend an hour researching a complex clinical question during the workday and receive no CME credit for it, then later spend another hour completing an activity specifically constructed to generate a certificate.

That distinction has always seemed artificial.

ChatGPT for Clinicians appeared to offer a much better model.

OpenAI announced ChatGPT for Clinicians on April 22, 2026, specifically advertising “CME from real clinical questions.” The proposition was straightforward: eligible evidence review performed while researching clinical questions in ChatGPT could automatically count toward CME, without separate courses or additional paperwork.

That sounds exactly like what physicians need.

The problem is that I cannot get it to function reliably.

I Am Doing the Kind of Work That Is Supposed to Generate CME

I use ChatGPT for Clinicians for exactly the type of work that appears to fit this model.

I ask clinical questions.

I examine mechanisms.

I request evidence.

I work through differential diagnoses.

I review medications and adverse effects.

I investigate guidelines.

I ask follow-up questions until I actually understand the issue rather than merely obtaining a one-line answer.

In other words, I am doing precisely what OpenAI describes: using ChatGPT to investigate real clinical questions arising from clinical work.

Yet I am not seeing a clear and dependable process in which those eligible activities are recognized, converted into CME, accumulated somewhere understandable, and made available as usable CME documentation.

If the system is recognizing these activities somewhere in the background, it is certainly not obvious to me as the clinician using it.

And if I cannot tell whether an activity qualified, why it qualified, why another did not qualify, how much credit I earned, or where that credit went, then from the user's perspective the feature is effectively not functioning.

The Strange Lack of Discussion About the CME Feature

Another thing has begun to stand out to me.

There is remarkably little public discussion specifically about the CME functionality.

There was coverage when ChatGPT for Clinicians launched in April 2026. Healthcare publications reported on the new clinician product, and several mentioned the ability to earn CME.

But months later, there appears to be very little follow-up discussion about the CME system itself.

Where are the articles from physicians saying:

“I earned 20 hours of CME this year simply by researching my real clinical questions in ChatGPT”?

Where are the demonstrations of the CME dashboard?

Where are the discussions explaining which clinical inquiries qualify?

Where are physicians comparing how much CME they are accumulating?

Where are the screenshots showing the workflow?

Where are the detailed reviews of whether the system actually recognizes evidence review accurately?

For a feature that could fundamentally change how physicians obtain continuing education credit, the public conversation around it seems remarkably quiet.

That does not prove clinicians are not using it. OpenAI does not publicly provide, at least that I have found, detailed utilization data showing how many clinicians have earned CME through ChatGPT for Clinicians or how many credits have been issued.

But the lack of visible discussion raises an obvious question:

Are many clinicians actually trying to use this feature yet?

Because if large numbers of physicians were routinely earning meaningful CME credits through their daily ChatGPT use, I would expect considerably more discussion of the experience.

Perhaps adoption is simply early.

Perhaps most clinicians do not know the feature exists.

Perhaps they are using ChatGPT for Clinicians but not attempting to claim CME.

Or perhaps other clinicians are encountering the same problem I am.

I do not know.

But the silence around such a potentially significant feature is notable.

Trying to Get It Fixed Is Almost More Frustrating Than the Problem

What has been particularly disappointing is the support experience.

A feature designed for physicians is supposed to eliminate friction.

The marketing proposition is essentially:

Ask the clinical questions you already ask. Learn while you work. Automatically earn CME. No extra paperwork.

That is a beautifully simple concept.

But when the feature does not appear to be working, the simplicity disappears.

In my experience, contacting OpenAI support about the problem has led to requests for increasingly technical troubleshooting information.

Screenshots.

Recordings reproducing the problem.

Browser information.

Reproducing the behavior under specific conditions.

And eventually discussion of creating a HAR file.

For those unfamiliar with the term, a HAR file is essentially a technical recording of browser network activity that developers can use to investigate what is happening between a web browser and a website.

This is not something the average physician normally does.

OpenAI's own support documentation describes a process involving opening browser developer tools, navigating to the Network tab, recording network requests while reproducing the problem, exporting the resulting HAR file, and supplying it to Support.

That may be perfectly reasonable troubleshooting for a software engineer.

It is a remarkably poor workflow for a physician attempting to use a feature that was explicitly marketed as requiring less administrative work.

There is something almost absurd about being told that CME should happen automatically while I practice medicine, only to find myself inspecting browser network traffic because the CME is not appearing.

Physicians Should Not Have to Become Software Debuggers to Earn CME

This is more than an inconvenience.

It reveals a mismatch between the target user and the support process.

A physician who reports:

“I have performed multiple evidence reviews in ChatGPT for Clinicians and the CME feature does not appear to be recognizing them.”

should not have to become the quality-assurance department for the product.

OpenAI presumably knows:

• what constitutes an eligible CME activity;

• what internal event triggers CME recognition;

• whether that trigger occurred;

• whether an activity was transmitted to the CME system;

• whether a credit was generated;

• whether the user's account is eligible;

• whether there was an application error; and

• whether the CME service is functioning normally.

Those are server-side questions.

The system should have sufficient internal logging to allow Support to inspect what happened.

At most, the physician should need to provide the date, approximate time, and conversation involved.

Instead, if the burden of proving that a feature failed is shifted back to the clinician through recordings, browser debugging, network captures, and HAR files, then the product is imposing exactly the sort of technical and administrative friction it claims to eliminate.

Is CME Actually a Priority?

That experience leads to a larger question.

How important is this feature to OpenAI?

ChatGPT itself evolves at an extraordinary pace.

Models change.

New tools appear.

New integrations are announced.

Healthcare functionality continues expanding.

Yet the CME component seems strangely invisible.

There is little obvious public education about how to maximize it.

There is little discussion of CME usage metrics.

There is little visible conversation about improvements to it.

There does not appear to be a prominent clinician-facing CME workflow explaining exactly what has qualified and why.

And when I report that it is not working, the support pathway feels less like support for a core professional product and more like troubleshooting an obscure software bug.

That creates the impression—fair or not—that CME may not currently be a major product priority.

A Great Idea Can Still Become a Gimmick

The word gimmick is intentionally provocative.

I do not mean that CME integration is inherently gimmicky.

Quite the opposite.

I think CME integration may be one of the smartest ideas in ChatGPT for Clinicians.

But a genuinely useful idea can become a marketing gimmick if it is featured prominently in a product launch and then receives insufficient attention afterward.

ChatGPT for Clinicians was introduced in April 2026 with a compelling list of capabilities.

Trusted medical search.

Clinical documentation.

Medical research.

Clinician-specific tools.

And CME from real clinical questions.

That last item immediately attracts the attention of practicing physicians because we all have continuing education requirements.

“Use ChatGPT for work and automatically earn CME” is an extremely effective marketing proposition.

But six months later, the important question is no longer whether the idea sounds good.

The question is:

Does it actually work?

At least for me, not reliably enough to be useful.

And when a supposedly automatic feature requires substantial technical troubleshooting just to determine why it is not generating credit, it starts to feel less like a finished professional capability and more like a feature that was announced before the workflow was fully mature.

This Is Not a Minor Feature

CME could easily be dismissed as an accessory attached to a much larger AI platform.

I think that would be a mistake.

For physicians, CME integration may be one of the most consequential ideas in the entire ChatGPT for Clinicians product.

Why?

Because AI is rapidly becoming part of the physician's cognitive workflow.

If I ask an AI system why premature atrial contractions can initiate AVNRT, examine the difference between fast and slow AV nodal pathways, ask follow-up questions about retrograde atrial activation, and then review supporting medical evidence, I have engaged in a genuine educational process.

If I investigate whether a medication is compatible with breastfeeding, compare recommendations, review pharmacology, and examine the supporting literature, I am learning.

If I research whether an occupational exposure plausibly caused a disease and review imaging findings, latency, epidemiology, and causation evidence, I am learning.

These are not artificial CME exercises.

They are practice-driven education.

And arguably they are more educationally relevant than many traditional CME activities because the learning occurs at the exact moment a physician has identified a knowledge gap.

AI Could Create a New Model of Continuing Medical Education

There is a bigger idea here.

The traditional CME system primarily measures educational activities.

AI potentially allows us to measure learning episodes.

Imagine a system that recognizes that a physician:

1. encountered a meaningful clinical question;

2. investigated the issue using reputable medical evidence;

3. explored the question beyond a superficial lookup;

4. reviewed relevant sources;

5. synthesized the information; and

6. applied that knowledge to professional practice.

That is a much richer model of continuing professional development.

The computer does not need to invent the educational activity.

The physician's work already created it.

The system simply needs to recognize it.

ChatGPT Is Particularly Well Positioned to Do This

This is where ChatGPT has an unusual advantage over conventional CME platforms.

A traditional CME provider knows that I completed its course.

ChatGPT potentially knows how I learned.

It can see the progression from the original clinical question to follow-up questions, clarification, evidence review, and synthesis.

A conversation may begin with:

What causes this arrhythmia?

Then progress to:

Why does the premature beat initiate the circuit?

Then:

What is different about the fast and slow pathways?

Then:

What happens to atrial activation during the tachycardia?

That progression contains information about the learning process itself.

The physician is not passively consuming information.

The physician is interrogating a problem.

That is exactly the kind of learning that continuing medical education should encourage.

Eligibility Cannot Be a Black Box

There is another problem that becomes apparent when CME is described as applying to eligible clinical questions.

What makes a question eligible?

That needs to be transparent.

Does the physician have to use clinical search?

Do citations have to be opened?

Does a minimum amount of time need to be spent reviewing the subject?

Does the conversation need to contain evidence from particular sources?

Does the physician need to answer a reflection question?

Does the system automatically decide whether meaningful learning occurred?

If a conversation does not qualify, the physician should be able to see why.

Otherwise, “eligible evidence review” becomes a black box.

That is particularly problematic in a professional credentialing context.

The Fix Is Not Complicated From the User's Perspective

Whatever complexity exists behind the scenes, the physician-facing workflow should be extremely simple.

ChatGPT for Clinicians should have a dedicated CME section showing:

• eligible activities;

• activities still requiring attestation;

• CME credit earned;

• the clinical topic;

• date of activity;

• evidence reviewed;

• accreditation information;

• downloadable certificates or transcripts;

• activities that did not qualify and why; and

• year-to-date totals.

There should also be a simple indicator inside qualifying clinical conversations:

CME eligible

or

CME activity recorded

And if something fails:

CME activity could not be recorded — contact support.

Support should then be able to see the failure internally without asking the physician to reconstruct the network behavior of the ChatGPT website.

I Still Think the Idea Is Excellent

I want to emphasize that this is criticism of the implementation, not the underlying concept.

The idea is excellent.

In fact, I think OpenAI may be pointing toward the future of continuing medical education.

Physicians should receive educational credit for meaningful evidence-based learning that occurs during the actual practice of medicine.

We should not have to pretend that learning only occurs when we log into a separate CME website.

Clinical practice generates questions.

Questions generate investigation.

Investigation generates learning.

That learning should count.

ChatGPT has the technological infrastructure to make that process measurable at a scale that previously would have been difficult or impossible.

But a Promised Feature Has to Function

As of my experience in November 2026, the CME component of ChatGPT for Clinicians has not delivered the seamless experience suggested by its April launch.

Perhaps there are eligibility criteria that are insufficiently visible.

Perhaps certain workflows trigger CME while others do not.

Perhaps my account has a technical problem.

Perhaps the functionality is still evolving.

Perhaps adoption is limited enough that few clinicians have encountered or publicly discussed these issues.

I cannot determine the underlying explanation from the outside.

What I can say is much simpler:

I am using ChatGPT for Clinicians to perform substantial evidence-based clinical learning, yet I am not seeing that work reliably translated into an accessible CME workflow.

When I have attempted to resolve the problem, the process has involved considerably more technical effort than one would expect from a feature marketed around eliminating extra work.

And months after launch, there appears to be remarkably little public discussion specifically about physicians successfully using the CME functionality.

None of those observations proves the feature is universally broken.

But together they raise a legitimate question about whether CME is currently a mature, actively prioritized component of ChatGPT for Clinicians—or whether it was a compelling April 2026 launch feature that has not yet received the attention required to fulfill its promise.

The Bigger Opportunity

AI is already changing how physicians retrieve information.

The next stage will be changing how medicine recognizes learning itself.

Instead of asking:

How many CME courses did you complete this year?

we may eventually ask:

What did you learn while practicing medicine this year?

That is a far more meaningful question.

ChatGPT for Clinicians could help make that transition possible.

But first, the CME feature needs to become transparent, reliable, and usable.

Physicians should not need to create HAR files to find out where their CME went.

If we are learning while we work, the technology should be capable of recognizing it.

And if the product promises CME while we work, we should be able to see the CME.

Author's note: This article describes my personal experience with ChatGPT for Clinicians as of November 2026. It does not establish that the CME feature is malfunctioning for every clinician or that few clinicians use it. OpenAI continues to advertise CME support for eligible clinical questions. I have not found publicly available utilization figures showing how many clinicians have earned CME through the feature or how many credits have been issued. The concern raised here is the gap between the advertised functionality and the workflow I have personally experienced.