You're about to upload a video. Pick Walkthrough and the AI captures clicks. Pick SOP and it also captures decision rules and the apps you used. Pick Tutorial and it explains the "why" behind each step. Pick Bug Report and the AI tags where things went wrong. Pick AI Agent SOP and the export gives you decision rules and actions an agent can execute. The pill you pick changes how your video gets read. Seven types, one decision.
What each type does
| Type | Use when | What changes in the AI |
|---|---|---|
| Walkthrough | Demonstrating a single feature or task end-to-end. Quick demos, screencasts, "let me show you how this works" content. Default. | Action-by-action breakdown: clicks, form entries, navigation, visual feedback. Generic action titles. |
| SOP | A business procedure your team follows correctly and consistently. | Names exact apps and services by their official name (Gmail, Sheets, Slack). Captures decision rules and rationale the narrator says, not just clicks. Flags validation checkpoints. |
| AI Agent SOP | A procedure designed for an AI agent to execute, not a human. | Same as SOP plus a structured decisionRule field for agent execution. Conditional logic and business rules get extracted. Output feeds into the manifest.json from the "Export For AI Agents" download. |
| Tutorial | Shipping "how to use" documentation for an app or product. End users learning your software for the first time. The "why" matters as much as the "what". Right pick when you're publishing user-facing docs. | Cursor-on-target screenshot selection for click actions (best frame for "this is what to click"). Learning-oriented titles ("Configure the database connection" instead of "Click Settings"). Concept explanations on first introduction. Reads like product onboarding, not an internal checklist. |
| Bug Report | Capturing a reproducible defect. | Bug step gets a "BUG:" title prefix. Focuses on the moment unexpected behavior appears, expected vs actual, and any error messages on screen. |
| How It Works | Explaining how a system behaves under the hood. | Focuses on mechanisms instead of clicks. Captures cause-and-effect ("when X happens, the system does Y") and transitions between states. |
| General | When none of the above fit. | Generic step-by-step. Less specialized analysis. |
Tips
- Pick Walkthrough if you're not sure. It's the default for a reason. Action-by-action works for most product demos.
- Tutorial scales across multiple apps. Documenting one product? Pick Tutorial. Documenting ten? Pick Tutorial for each upload. Same output style means your help center reads coherent across your portfolio, not pieced together one type at a time.
- Tutorial uses cursor-on-target frame selection. For clicks, it picks the screenshot where your cursor is hovering on the target button, not after. Best frames for "this is what to click" docs.
- SOP and AI Agent SOP capture decision rules. These are the only types that explicitly extract the narrator's "why" alongside the clicks. If your video has conditional logic ("if enterprise client, send to Plan B"), pick one of these or that nuance won't make it into the steps.
- Type is fixed at extraction time. You pick it once before clicking Generate Flow. Changing your mind later means re-uploading the video and spending minutes again. Pick deliberately.