Following these best practices will significantly improve the accuracy of Visual AI Checkpoints and the participant experience.
Capture High-Quality Reference Images
✓ Do
- Use the Take Snapshot button whenever possible, as described in Taking a Snapshot, which improves the accuracy of the Visual AI Checkpoint algorithm.
- Use full, uncropped screenshots.
- Match the student VM screen resolution.
- Upload .png (RGBA) format.
✗ Avoid
- Manually resized or cropped screenshots.
- Images with a blinking cursor or system clock.
- Screenshots taken with browser UI visible.
- Images that don't reflect the participant's VM environment.
Select Specific, Static Checkpoint Areas
When marking the Checkpoint area on your reference image, choose regions that uniquely identify the desired end state:
- Select specific UI elements, such as buttons, form fields, table rows, or status indicators.
- Avoid selecting large background areas, generic text, or regions that look similar regardless of participant state.
- The more unique and static the selected area, the lower the chance of false positives.
Guide Participants to the Right VM and Screen
A Visual AI Checkpoint is tied to a specific visual element on a designated machine.
- Provide clear Guided Journey instructions that direct participants to the correct VM.
- Make sure participants are on the same screen/product view that was used when the Checkpoint was created.
- A Checkpoint will fail if the participant is on a different machine or has a different screen open.
Test Before Going Live
Always validate your Checkpoints in the Editor environment before running the experience with participants:
- Run through the Guided Journey as a participant would confirm each Checkpoint triggers correctly.
- Verify that failure hints are accurate and helpful.
- If a Checkpoint fails unexpectedly, review your reference image quality and selected area.
Keep Checkpoints Relevant to Instructions
Each Checkpoint should directly correspond to the objective or task described in the section instructions.
If participants are asked to configure a firewall rule, the Checkpoint should verify a visual indicator confirming the rule is set, rather than an unrelated part of the screen.
Understanding Current Limitations
Be aware of the following constraints when designing your Checkpoints:
- One Checkpoint can be defined per section.
- Checkpoints cannot be defined for multi-step class environments.
- Each Checkpoint is evaluated independently and does not depend on the result of any other section.
- Checkpoint results do not block a participant from progressing to another section or chapter, meaning that participants can always skip and move on.
Pro Tips for Accurate AI Matching
- Originals Only: Use full, uncropped screenshots without manual resizing.
- Match the Student View: Ensure your screen resolution matches the student's VM environment.
- Select Unique Areas: Mark clear, static regions. Avoid blinking cursors or system clocks.
- Stay Consistent: The more your source image reflects the live environment, the better the AI performs.