Platform certification lets you train and certify your team on any software tool using Yoodli's AI Tutor. The AI walks learners through the tool's interface, answers their questions, and then coaches them step-by-step as they practice in the real tool on their own screen.
This guide covers setup for two scenarios:
Well-known tools (Claude, ChatGPT, Gemini, Copilot, HubSpot, Salesforce, etc.) — faster to set up because the AI already understands how these tools work.
Custom or internal tools — requires more detailed setup since the AI has no prior knowledge of your tool.
Setting up certification for well-known tools
If you're training your team on a widely used tool like Claude, ChatGPT, HubSpot, or Salesforce, setup is straightforward. The AI already has deep familiarity with these products, so your job is to define the learning flow and capture a few key screenshots.
Step 1: Define the learning outcome
Before building anything, decide what the learner should be able to do by the end of the session. Be specific. "Learn Claude" is too broad. "Draft a LinkedIn post using Claude, including prompt iteration" is a concrete task you can build a session around.
Good examples:
Use Claude to draft and refine a customer email
Create a HubSpot deal pipeline and move a contact through stages
Build a prompt in ChatGPT that summarizes meeting notes into action items
Step 2: Create a new roleplay of type Tutor
Navigate to the Roleplay Builder as an Admin and create a new roleplay. Select Tutor as the template.
Step 3: Write the roleplay instructions
The context field is where you tell the AI exactly how to run the session. Think of it as a lesson plan. The AI will follow this sequence with the learner.
A strong set of instructions typically includes:
How to greet the learner and assess their experience level
Which parts of the tool's interface to walk through (and in what order)
When to stop sharing and ask the learner to take over
What hands-on task the learner should complete
How to coach them during the task
Here is an example for a Claude training session:
You are conducting a focused, hands-on training session to make the user proficient at using Claude. Start by introducing yourself and directly asking about the user's prior experience with Claude. Do not engage in small talk or personal conversation. Walk the user through the following
steps in accordance with their stated experience level.
Share your screen and detail the key elements of the Claude interface, explaining each one clearly.
Adjust your level of detail based on the user's stated experience level. For beginners, explain the purpose of each feature in detail and give examples for when they would be used. For intermediate users, explain the basics with one example. For advanced users, briefly describe each feature but do not give examples.
Show ClaudeChat.png and cover the new chat button and how to start conversations.
Then show ClaudeProjects.png and explain Projects and how they help organize work.
Last show ClaudeArtifacts.png and explain Artifacts and how Claude generates usable outputs like documents, code, or other structured content.
After covering the interface basics, stop sharing your screen.
Transition into a practical exercise: how to write an effective LinkedIn post using Claude. Ask the user to share their screen. Have them browse to claude.ai and log in. Walk the user through how to craft a strong prompt, what kinds of instructions produce better results, and how to iterate on Claude's output to refine it.
Have the user draft a LinkedIn post about how AI is improving their personal productivity. Coach them through the process in real time by offering tips on prompt structure, suggesting refinements, and helping them understand why certain approaches get better results from Claude.
End by instructing the user to copy and paste Claude's output into a LinkedIn post.
A few things to notice in this example:
Experience-level branching. The AI adjusts depth based on what the learner says. You write the branching logic directly into the instructions.
Screenshot sequencing. Each image is referenced by filename and tied to a specific teaching moment. The AI shows them in order.
The screen-share handoff. The instructions explicitly tell the AI when to stop sharing its own screen and ask the learner to share theirs. This transition is the core of the session — it's where the learner moves from watching to doing.
A concrete task. The learner isn't just clicking around. They're producing something real (a LinkedIn post) while the AI coaches them.
For well-known tools, you don't need to explain how the tool works in the instructions. The AI already knows. You're directing what to teach and in what order.
Step 4: Capture and upload screenshots
Take screenshots of the key screens you want the AI to walk through. Name them clearly — the filenames you use here must match the filenames referenced in your instructions.
For the Claude example above, you'd capture:
ClaudeChat.png — The main chat interface with the prompt box visible
ClaudeProjects.png — The Projects section in the left navigation
ClaudeArtifacts.png — An example of an Artifact output
Upload these in the roleplay's visual content section.
Tips for good screenshots:
Capture the full screen or the relevant section — avoid cropping so tightly that the learner loses spatial context.
If there are multiple steps within one screen (e.g., a dropdown menu that needs to be opened), capture each state as a separate image.
Use clear, descriptive filenames. The AI references these by name, so HubSpotDealPipeline.png is better than screenshot_3.png.
Specify instructions to display these images under display instructions of visual content
Example:
You have three interface screenshots available to display during the conversation. When walking the user through the Claude interface, you should display each one individually as you discuss the corresponding feature.
Display the Claude chat interface screenshot when you are explaining how to start a new conversation and pointing out the new chat button and main input area. This screenshot shows the welcome screen with the input field and navigation options.
Display the Projects page screenshot when you are explaining what Projects are and how they help organize work. This screenshot shows an empty Projects section with the option to create a new project.
Display the Artifacts screenshot when you are explaining what Artifacts are and how Claude generates structured outputs. This screenshot shows a grid of various AI-powered tools and templates that illustrate the range of outputs Claude can produce.
Once you have finished walking through all three screenshots, you should not display any further files for the remainder of the conversation.
Step 5: Configure the persona
Set up the AI persona for the session. Give it a name, a professional tone, and a demeanor appropriate for training (empathetic, enthusiastic, friendly).
Step 6: Set up goals and assessment (optional)
If you want to certify learners — not just train them — add custom goals to evaluate whether they completed the task successfully. For example:
Did the learner write a prompt that included specific instructions for tone and audience?
Did the learner iterate on Claude's first output at least once?
Did the learner produce a final LinkedIn post?
Step 7: Add the roleplay to a program
To create a full certification track, add this roleplay (along with any others) to a Program. Programs let you sequence multiple sessions, track completion, and gate progression.
For example: Video, AI Tutor Roleplay and Accreditation
Step 8: Test the experience
Run through the full session yourself before assigning it to learners. Pay attention to:
Does the AI show screenshots in the right order?
Does the screen-share handoff feel natural?
Is the coaching during the learner's screen-share session helpful and accurate?
Does the AI stay on-task or drift into unrelated topics?
Alternative: generic instructions for well-known tools
If you don't need a tightly scripted session, you can write looser instructions and let the AI adapt to the learner's own environment. This works well for tools the AI knows deeply.
Here's an example for HubSpot:
You are an experienced trainer helping a Customer Success Manager learn how to use HubSpot effectively. Focus on practical, hands-on guidance for managing renewals, tracking customer progress, and handling key accounts. Start by understanding their current experience level and specific challenges. Ask the user to share their screen and bring up HubSpot. Provide step-by-step instructions, share best practices, and offer tips for common workflows using the user's own HubSpot instance and accounts. Be patient and encouraging, breaking down complex processes into manageable steps. Ask questions to ensure they understand each concept before moving forward.
With generic instructions, you don't need to upload screenshots. The AI coaches entirely from the learner's shared screen and its own knowledge of the tool. This is faster to set up but gives you less control over the exact flow.
Setting up certification for custom or internal tools
If you're training your team on a proprietary tool, an internal platform, or any software the AI wouldn't already know, setup requires more preparation. The AI has no prior knowledge of your tool's interface, terminology, or workflows — so you need to provide that context explicitly.
What changes
| Well-known tools | Custom tools |
Instructions | Define the flow; AI fills in product knowledge | Define the flow and explain how the tool works |
Screenshots | Key screens for the teaching sequence | More granular — every screen state the AI needs to reference |
Documentation | Optional | Required — the AI needs written context about your tool's features, terminology, and workflows |
Testing | Standard QA pass | More thorough iteration; expect to refine instructions and screenshots based on how the AI performs |
How to approach it
Provide detailed documentation in the context field. Describe what the tool does, what each screen or section contains, what the key workflows are, and what terminology the learner will encounter. Write it as if you're explaining the tool to a new hire who has never seen it.
Capture more screenshots and label them precisely. For a well-known tool, three to five screenshots might cover a session. For a custom tool, you may need significantly more — every distinct screen, dialog, dropdown state, or configuration panel that the AI will need to reference or explain.
Be more explicit in the instructions. Instead of "explain the dashboard," write "Show Dashboard.png and explain that the top section displays active alerts sorted by severity, the middle section shows a timeline of recent deployments, and the sidebar contains filters for team and environment." The AI cannot infer what's on screen for a tool it doesn't know.
Plan for iteration. Custom tool setups typically require a few rounds of testing and refinement. The AI may misinterpret a screenshot, skip a step, or coach inaccurately if the documentation has gaps. This is expected. Run the session, note where it breaks, add detail to the instructions or context, and re-test.
Working with Yoodli on complex setups
For custom tool certification, we recommend working directly with your Yoodli account team during setup. Your CSM can help you:
Structure the documentation for best results
Identify where additional screenshots or context are needed
Run test sessions and troubleshoot AI behavior
Refine the experience before rolling out to learners
Reach out to your account team or contact [email protected] to get started.
Summary
Setup type | Effort | When to use |
Well-known tool, prescriptive | Medium — screenshots + detailed instructions | You want a controlled, consistent learning flow |
Well-known tool, generic | Low — instructions only | You want the AI to adapt to each learner's environment |
Custom tool | High — documentation + granular screenshots + iteration | You're training on an internal or proprietary tool |
For questions or help with setup, contact your Yoodli account team or reach out to [email protected].