💬 *MEGATHREAD*: Say Hello to Role Play! 💬

MarinaT

Hi @Boris035,

I'm already back with a quick update from the team. They confirmed that for now, it's best to stick with simpler role plays.
The good news is that they're planning to introduce difficulty levels (like Medium and Advanced) in future releases, which will give you the flexibility to design more complex scenarios for your learners.

While we don’t have a timeline yet for when those will be available, please rest assured we’ll keep you posted with any updates.

Boris035

Great @MarinaT ! Thanks a lot, I appreciate it!

PuneetGupta453

Udemy is competing with Google Gemini by introducing roleplay. I recently experienced Gemini and it is awesome. I hope Udemy can build the tool and build it better than Google Gemini.

Roleplay should be able to access a particular lecture from our course content. It should be able to build a story around what is being taught in the lecture. I mean Udemy should think how it can enhance roleplay and embed story telling into it to make is a future ready tool.

GaryLloyd

I’m delighted to see Udemy experimenting with AI role-play—it’s a bold and necessary step toward supporting real practice in skill-based learning. I’m currently building an AI-first course on Udemy, explicitly designed around this new feature, and I’ve had experience developing similar role-plays outside the platform as well.

There’s huge potential here, and some strong foundations already. The interface is clean, and learners can step into structured practice in a way that’s far more active than a multiple-choice quiz. Even in its current form, it’s adding value.

However, if this feature is to support truly rich learning experiences, I believe there are a few critical limitations that need to be addressed.

Right now, there’s no way to give instructions to the AI character that are hidden from the learner. That’s a deal-breaker for anything beyond surface-level exchanges. If I want the AI to subtly test the learner, hold a specific tone, or escalate resistance across turns, I simply can’t. I can describe the character in the scenario and first line, but it’s a blunt instrument—and learners see it all.

This limitation also blocks another powerful use case: Socratic tutorials. These are interactions where the AI asks probing questions to help the learner reflect, reframe, or reconsider their assumptions. I’ve used these effectively elsewhere, but here, because I can’t invisibly guide the AI’s behaviour, I’m forced to flatten the experience into a pre-set sequence. It becomes more of a quiz than a conversation.

As a workaround, I’ve ended up rewriting the learner-facing “scenario” and “personality” fields to double as indirect instructions for the AI. It’s not ideal, but it’s the only way to nudge the behaviour I need. I’ve also had to frame some role-plays more like coaching simulations or guided walkthroughs, because that’s the only level of complexity the current system seems able to handle.

Despite these constraints, I’m pressing ahead. I believe strongly in the potential of AI-driven practice, and I want to help shape how this evolves on the platform. But for it to become a genuinely transformative tool for instructors, we need:

• A way to privately prompt or instruct the AI
• Better clarity on which fields influence behaviour
• And ideally, an option to use GPT-4o for stronger instruction following

Finally, if we are to create experiential learning experiences, Udemy needs to reconsider the need for

30 minutes of video content in paid courses. This was fine when we were limited to mostly video information giving,

GaryLloyd

I never quite finished the previous post. What I wanted to say was:

Finally, if we’re serious about creating experiential, AI-first learning, Udemy needs to reconsider the 30-minute minimum video requirement for paid courses. That threshold made sense when courses were mostly built around video lectures, occasionally broken up by quizzes. But in an AI-driven model, video is used differently—short, 2–3 minute bursts of instruction followed by hands-on role-plays or tutorials.

In the Giving Feedback and Difficult Conversations course I’m currently building, 10–15 short, purposeful videos already feel like a lot. Padding them just to meet a time requirement risks weakening the course rather than strengthening it.

AHardin

Hi @MarinaT, I just created my first role play, love it's overall initial launch functionality, and plan to create much more.

However, one issue I found is that the mic recording limits responds to the AI character to 22 seconds, which when testing my first role play with simple responses, I frequently exceeded 22 seconds. And I could easily see students doing as well. I had to train myself to speed talk in my role play testing, so I didn't exceed 22 seconds.

It might be beneficial to show students a recording countdown time of some sort, otherwise I could see student quickly getting frustrated when their mic cuts off and it tells them they exceeded the recording limits. Also, do you know if there's plans to increase the limit a bit? 22 seconds is quite short.

MarinaT

Hi @AHardin,

Thank you so much for your feedback! Yes, having a countdown added for students to see how many more seconds they have for their answer is a great idea. Let me forward it to the team right away!

AHardin

Thanks @MarinaT!

MarinaT

Hi @PuneetGupta453,

Thank you so much for your feedback. It’s been forwarded to our team for consideration as they continue to improve the tool.
Since @WilliamStewart had a similar request about role plays using instructor content, I wanted to take this opportunity to share an update from the team here:

At the moment, our role play feature uses OpenAI models through their API, combining different models, settings, and modalities. These aren’t tailored to an instructor’s specific course content yet. That said, we completely agree! Customizing role plays to align with your unique course material not only makes them more relevant, but also shows the expert knowledge you bring to your course. This is something we’re actively working on for future releases.

Again, thank you so much for your feedback, and please don’t hesitate to reach out if you have any other questions or thoughts to share.

Stay tuned!

MarinaT

Hi @GaryLloyd,

Thank you so much for taking the time to share this feedback about our role play feature. We’re glad to hear you like the interface! You’re right, there are still some limitations with this early version, which is why feedback like yours is so important.

I completely understand where you’re coming from with your suggestions, especially around probing questions and the ability to add private instructions to the AI persona without displaying them to learners. I’ll make sure all of your feedback, including your point about allowing courses under 30 minutes in the context of role plays, is shared with the internal team for consideration. What we can already say is that the team does have plans to introduce difficulty levels like Medium and Advanced, which may support the kinds of scenarios you’re looking to create.

If there’s anything else you’d like to see in the role play feature, please don’t hesitate to let us know. We know there’s still a lot of potential to explore!

WilliamStewart

Thanks @MarinaT . If the role plays do not reinforce the Udemy course content, carefully curated by human experts to provide very specific and valuable advice, and instead reflect back generic OpenAI information from across the Internet, I extremely strongly with every fiber of my being, including the atoms at the end of my fingernails, recommend you do not publicly role out this functionality. Until you can provide it run with a Udemy specific LLM, that reflects back the specific course content it is being run for.

I have done several tests. The role play AI feedback actually dilutes and detracts from my highly specific course value. And the big picture is that this amounts, in effect, to turning over our courses to providing services from one of Udemy's biggest competitors 😱. That just does not sound like a good idea.

(Parenthetically, fwiw, imho, with love, if you had a senior instructor internal to Udemy, on payroll and under NDA, paid to represent Udemy's interests, in the discussions and at the table during planning, this issue would have been picked up much earlier. "Consultation" just does not do it. Think of it like this: it is like trying to run an engineering company without any engineers on the executive team.)

LawrenceMMiller

I am with @WilliamStewart on this. In most of my courses I teach models, a process for whatever the topic is. The role plays need to be about applying the models or methods that I am teaching based on my field experience. I haven't figured out how to make that work yet, but I will keep trying.

Boris035

@LawrenceMMiller - it does not have to be that way, at least in the first stage of the Role Play development. For example, you can start with some general concepts and frameworks - e.g., to create one about the ADKAR Model for your Change Management course, applying SWOT analysis for your Strategy course, or giving feedback for your Facilitation course. You can improvise with more specific concepts once you better understand how the feature works.

I think it's a fantastic tool, but it takes a lot of testing and iterations, at least initially.

LawrenceMMiller

Thank you @Boris035. You know my stuff! 😊

Boris035

@MarinaT , can you clarify one thing about the Role Play? Once you make it public, all Role Play activities will last as long as those activities are pre-set, meaning either 3, 5, or 10 minutes. And, after that period, the feature will automatically stop working?

Why am I asking this? While testing some of them, I have realized more time is needed for some of them - simply, you can complete only the basic Role Plays within 3-5 minutes. Yet, for more complex activities, even 10 minutes might not be enough. And, during this test period, we (as instructors) are allowed to play it even longer than the stated 3, 5, or 10 minutes. I would like this functionality/option to stay if possible ;-)

Related question - if a student is not able to complete a Role Play within the stated timeframe, how is this going to affect his/her course completion, including a Certificate (since for some of them, that's an important question)?

In addition, some Role Plays can be created as real business cases, and for that, it will be useful if we are eligible to create a page (or attach a PDF) for students with some specific details like numbers, hints, etc. Of course, those things can be specified within the "Scenario" section - yet, it will be helpful for students to have them handy while crunching the case.

Thanks,

Boris

OxfordLearnLab

@MarinaT I went through the guide and also the thread. I can grasp how to apply Role-Play to soft skills scenarios. However, my courses are all about Marketing (strategy or tactical), so I am not sure how to use it. Are you thinking about providing more sector-specific scenarios that can inspire role-play settings?

MarinaT

Hi @Boris035,

Thank you so much for your thoughtful questions and for sharing your feedback about the option to attach a PDF to the roleplay! I’ll check in with the internal team and get back to you as soon as I have an update 😊.

MarinaT

Hi @OxfordLearnLab,

Thanks so much for your comment. I'm glad you had a chance to check out the thread and explore the role-play feature!

As you mentioned sector-specific scenarios: Are you thinking of examples tailored to marketing in particular (or any other subject), to help you see how role-play might be applied in your own courses?

Just want to make sure I’m understanding you right. Either way, it’s a great suggestion, and I’ll definitely pass it along to the team.

OxfordLearnLab

@MarinaT yes exactly. I played with the tool and created a soft skill scenario similar to the examples given. Dealing with a problematic client of a Marketing Agency. The result is interesting but is not what i teach in the course itself. So my concern is that the learner will not have the instruments to answer. I will keep it anyway as it looks potentially useful to understand when dealing with clients.

Paul_Imre

In reply to @WilliamStewart
Thanks for sharing your observations. We should all watch out for the repeat of your findings in our work.

Possible reason. In terms of general LLM use, such as ChatGPT, prompt engineering plays a huge role, excuse the pun, in the quality of output. The better the prompt, the better the results. Logically speaking, how we bound the Udemy system with student role, learner role and AI character will make a big difference. The key therefore is to understand how these elements map onto a prompt. Examples:

R – Role
A – Action
C – Context
E – Expected Output

Prompt: Act as a nutritionist (Role). Explain (Action) the importance of fiber in the diet (Context). Provide 3 simple tips (Expected Output).

Or

F – Format
O – Objective
R – Role
M – Method

Prompt: Write a paragraph (Format) to inform teens about the dangers of vaping (Objective), as a health educator (Role), using simple language and a friendly tone (Method).

Udemy sits in between us and the LLM. How much we can influence things is a good question as @RobertBrown028 managed to work his magic. Sounds like we need a hackathon or a lab or heaven forbid a Udemy instructor roleplay :)

@MarinaT can Udemy share any info on prompting techniques?

PuneetGupta453

I’m eagerly anticipating the launch of Roleplay and curious to see how students engage with it.

That said, Udemy might have benefited from introducing this feature initially to a limited group of students—ideally those who already have positive experiences with Udemy courses—to gather feedback and refine the experience. Already the feedback from instructors looks like, it is half-baked cake.

If Roleplay isn’t thoroughly optimized before a full rollout, there’s a risk it could lead to unintended dips in course ratings.

Early-adopter instructors, like myself, may end up disproportionately impacted by experimenting with an untested feature, despite our enthusiasm for innovation. Just a humble suggestion to prioritize a phased approach!

MarinaT

Hi @OxfordLearnLab,

Thank you for confirming. I've already shared your feedback with our internal team and will keep you updated here in the community on any developments. If anything comes to mind while you're working on the role play or reviewing learner feedback, please feel free to share it.

Wishing you a great weekend!😊

Paul_Imre

Another thought on the @WilliamStewart v @RobertBrown028 success or not.

During my own roleplay testing, I found that the learner was missing the objectives. To investigate further, I asked ChatGPT for feedback. The response regarding the AI character’s first line was particularly insightful — see below:

Weaknesses (in relation to your goals):

1. Too broad and deferential
   The line places the learner in a “supporting expert” role but doesn’t challenge them to influence or persuade — a key part of their goal. It frames them as a helper rather than someone navigating organizational friction or resistance.
2. Steers toward operational fix, not strategic or ethical insight
   Focusing the first question on “unofficial AI use” narrows the scope to reactive governance (device policy, access control), instead of prompting a conversation about foundational issues like misuse of data, ethical oversight, or the risks of relying on inaccurate outputs from LLMs.
3. By skipping the tension, it softens the urgency
   There's little indication of Suzi being pulled between innovation and compliance, or of her being tempted to downplay risks. This misses a chance to trigger learner advocacy and influence — particularly around how to "sell" privacy in a business-driven context.

Take away? I think @WilliamStewart is warning us not to rush into this. You might risk the quality of your course. Perhaps warn your students that this is a new thing.

WilliamStewart

Thanks @Paul_Imre , to clarify, I think role play functionality would be very very useful, if the AI is trained on our course content. It could then be used by the student to confirm in a realistic interaction that they understand our course content. Very very cool indeed.

The current implementation is just generic ChatGPT. Any role play you create could be run directly through the ChatGPT site. It does not add to our courses or Udemy core value. And, in my own tests, since it contradicted my course material, by saying that the drivers of project success were the standard ChatGPT response, whereas my course focuses on three much more specific and actionable drivers, the role play was counter-productive.

If the role play reinforced my course focus on three very specific and actionable drivers of success, and the very specific actions I describe that the student can implement to ensure those drivers make their project successful, then it would be very very useful. As generic ChatGPT, I could not get it to reflect back any of my course material, and instead it responded with completely different success drivers that were much higher level, general, and not useful to a poor project manager trying to make their project successful right now in the real world. That is, it actually contradicted my course.

With my strategic hat on, I believe, as one of the smartest people I ever met once told me, "every step, every thought, every action you take should reinforce your core value".

Addition of generic ChatGPT to the Udemy site moves away from Udemy core value. On the other hand, an LLM trained specifically on course content would take large steps towards reinforcing Udemy core value. I believe Udemy should implement this functionality, once, and right, with an LLM trained on our courses.

I'm just trying to be genuinely helpful here.

Boris035

@Paul_Imre - that's interesting, and I think this kind of exchange between instructors is very important. And yes, I would like Udemy was able to provide us with more useful cases and instructions before we entered this journey.

Nevertheless, here are some findings from my side:

• I found the tool works better if you put AI in the role of a coach, mentor, or boss, while a student is the one who should explain things and explore possible solutions.
• There are a few levels of difficulty we can create. For example:
  • Basic one, to use Role-Play for an easy interactive exercise - e.g. a student should understand and classify certain things based on a well-known framework, like the BCG Matrix, to classify if the company belongs to Cash Cows, Stars, or some other type, and give some basic recommendations about future strategy.
  • The next level of difficulty is where AI navigates a student via some framework where the student should think critically and give explanations. For example, to elaborate on market attractiveness based on Porter's 5 forces framework, to create a SWOT analysis, and similar.
  • The most complex level refers to unstructured conversations (meaning without a clear framework to be followed) - that's like solving a real business case. Once you do it right, it's awesome Role Play. Yet, it might require over 10 minutes to be solved, and a student might struggle to find a solution without clear guidelines (hints) provided in the Scenario section.
• You have to test each Role Play you created because "loops", meaning AI repeating the sentences, can happen, so you have to find a solution and solve it before publishing.

Those are the key findings from my side for now. In general, I like it, and I hope students will love it. Of course, the risk is on our side if something goes wrong.

Boris

LawrenceMMiller

I very much appreciate this discussion. I have been hesitant to dive into role play creation because I am unsure that role plays that do not reinforce what I am teaching will be experienced as just more work for the student. I love the idea of role plays but I am struggling to make them work.

AbiyotAderi862

Role play, this things is new for me, please post clear instruction

OxfordLearnLab

@Boris035 like your post. Did you manage to make it work? I was thinking at SWOT that seems quite straight forward but if you do not have enough info to start with 10’ is a very short time. This is the reason why 90% of SWOT’s are useless, they are not better than a guess work unless you have solid data about what you are analysing.
Thank you for the valuable conversation :)

OxfordLearnLab

@MarinaT i tested and tried with some external people i know showing the preview - just a bit worried to going live. The answer was a “fun” thing to do but was not directly relatable to what i am teaching. I am sure there is a much better way to design the exercise and is my fault if i did not found the “right” one. I guess following the community and Udemy future examples this tool can be very powerful.

Boris035

@LawrenceMMiller , @OxfordLearnLab ,

Check my SWOT Analysis in Action Role Play. I'm sure you can copy/paste it and see how it works. It would be nice to change something if you want to keep it - it's not OK to have exactly the same Role Play activity.

And, as you will see, it takes some preparation if you want to do it in under 10 minutes.

SWOT Analysis in Action:

Plan Role Play

Scenario:

You’ve just joined Evercharge Mobility, a European startup that builds and operates electric vehicle (EV) charging stations in mid-sized cities. The company was founded in 2020 and has grown quickly, with 2,100 charging points across 6 countries and a €42M Series B funding round closed last year. Evercharge focuses on affordable fast-charging hubs placed in suburban retail zones, partnering with local municipalities and real estate groups.

Recently, growth has slowed. New competitors—including oil giants like VoltFuel and tech-backed startups like SparkNet—are entering the space aggressively, offering smarter hardware, dynamic pricing, and app-based loyalty perks. Evercharge’s utilization rate has dipped to 38%, and customer satisfaction is being hit by occasional downtimes.

As the company prepares for a strategy offsite, the CEO has asked you to lead the creation of a refreshed SWOT analysis. You’ll meet with Sofia Richter, an experienced strategy coach, to challenge your thinking and sharpen your output. You’re expected to go beyond buzzwords and identify what truly matters: What should the business build on? Where is it vulnerable? What trends can it leverage? What risks could derail it?

This SWOT will set the stage for Evercharge’s next growth chapter—so clarity and strategic depth are key.

Learner Role:

Junior Strategy Associate – Evercharge Mobility

AI Character Customization:

Name: Sofia Richter

Character Role: Strategy Coach (former EV executive and now advisor)

Personality and background:

Calm, analytical, and experienced in scaling cleantech startups. Sofia pushes for clarity, avoids fluff, and expects thoughtful strategic logic. She will challenge vague ideas and help you sharpen your insights.

AI Character’s First Line:

"Welcome aboard! I saw your first draft of the SWOT—good start, but let’s make it sharper. I’ll challenge you on each part so we build something truly strategic. Let’s start with Strengths—what do you think sets Evercharge apart in the EV charging space?"

Role Play Setting:

Meeting title:

SWOT Strategy Prep – Evercharge Mobility

Meeting goals:

Help you develop a clear, realistic SWOT analysis

Push you to avoid generic or surface-level answers

Prepare you to use this SWOT in a broader strategy discussion