Some people learn by watching; others want a written procedure they can skim and search. Teams that maintain both separately often find they drift apart.
If your training video is already broken into clear steps, it holds most of a written standard operating procedure (SOP): the task, the steps in order, what to do in each, and a picture of each screen. This guide shows how to get both from one recording, and what you still need to check and add by hand.
Start with a step-by-step video
The written guide is only as good as the steps behind it. Before you think about the document, build the video:
- Record the task in your own voice, one task per recording.
- In Make it safe, click Prepare video, then check every item in Review blurs. Draw a blur box over anything the scan missed.
- In Studio > Callouts, add numbered step cards (eyebrows like "STEP 1", "STEP 2"). Suggest callouts can draft them from your narration, screen text and script.
- Trim, frame and render as usual.
Each numbered step card becomes one step in the written guide, so make the step titles clear and action-first: "Open the quote", "Add a line item", "Save and check the total".
Turn the steps into a written guide
In GlideClip this is the AI data area (we call it Codify). Exporting the bundle needs Pro.
- Click AI data at the top of the rail, then Segments. There is one segment per step.
- Click a segment and check the Task name and the times. Use Start at playhead and End at playhead to adjust.
- Write the Answer-style summary: one or two sentences that answer "how do I..." on their own. Or click Draft from the narration to start from what you said.
- Fill in Actions, one per line, such as "Click Add line item".
- Add Tags, and under Questions this step answers, click + Add a question for the questions people actually ask.
- Set the step's picture: move the playhead to the right moment and click Use the playhead, or leave it on Automatic.
With Pro, Draft N steps with AI fills empty summaries, actions, tags and questions for every step at once. It sends titles, bullets, the redacted narration and screen labels as text, with no video or pictures, and it never replaces text you typed. Text written by AI is marked so you can read it before you export.
Add the context an SOP needs
A video can get away with jumping straight into the task. A written procedure can't. Open AI data > Video details and fill in what helps:
- Before you start: prerequisites, such as the role or permissions needed, and what must already exist (for example "a quote in Draft status").
- Result: what the reader should see when they're done.
- Who it's for, System and Version, so readers know whether it applies to them.
- Owner and Review by, so someone is responsible for keeping it current.
- Video link, which adds a "Watch this step" link to each step.
Export the bundle
Open AI data > Export, fix anything listed under Before you can export (the video must be prepared, have step cards, and every step needs a title and summary), then click Export bundle (.zip). The zip is saved on your computer and holds:
- guide.md: the written procedure, one heading per step, with its questions and a link or time.
- segments.jsonl: one record per step, with the summary, actions, tags, Q&A, narration and times.
- screens/: one picture per step, blurred like the safe copy. You can choose to Draw highlight boxes on the step pictures.
- captions.vtt: the redacted narration, as a separate file. (For captions inside the finished video, turn on Captions from the narration in Render.)
- manifest.json: the source, policy, rights and how many items were blurred.
Every text field goes through the redactor at export time, and the step times match the finished video, with trims and cuts taken out.
Review it like any other document
Treat the export as a strong first draft, not a finished SOP:
- Read every step against the software. Does each action match what's on screen? Did a summary drafted from your narration pick up an aside you didn't mean as an instruction?
- Check the pictures. Is each one the right moment, and is everything sensitive covered?
- Check the masked text. The narration and other text are masked where the scan found something, but names are only masked when you listed them, and the scan can miss small or fast-moving text.
- Fill the gaps a video hides. Warnings, exceptions ("if the product isn't listed, ask an admin") and what to do when something fails often live only in your head.
Then paste guide.md into your help center or wiki and adjust the formatting.
A written guide is not "training an AI"
It's worth being clear about what this export is. It is a set of files: a readable guide, structured step records and pictures. Your team can read it, publish it, or load it into tools you already use, such as a help center, a wiki or an internal assistant your company runs.
GlideClip does not train AI models, and it does not build or host an assistant for you. It doesn't connect to other systems either; you decide where the files go. The structured records and Q&A simply make the content easier for your own tools to use.
When this works best
- Repeatable, click-by-click tasks in business software: creating records, approving requests, running reports.
- Tasks that change occasionally. Re-record and re-export to keep both in step.
For judgment-heavy work where the "why" matters more than the clicks, write that part yourself.
Try it on a sample
You can start with the video part on the free plan. GlideClip opens a 38-second sample recording of a made-up CRM, so you can see how step cards come together before you record your own task.
Your recording is processed in your browser and never uploaded. Optional AI features work from text only, with detected details masked.