← All insights

AI · Case note

How we turned a help centre into a self-updating video library

Yohji Alexander Kanki · · 1 min summary · 5 min full read

The problem

A fast-growing software company needed a walkthrough video for every one of its several hundred help articles.

Each one took hours to make by hand. Every product update made them outdated again. And with a small team and no video editor, they were never going to catch up.

The solution

We set up a system that makes the videos for them, and keeps making them.

  • Videos are recorded straight from the live product, with nobody at the mouse.
  • Voice, editing and captions are added automatically, in the same style every time.
  • When the product changes, an updated video is ready in about 15 minutes.
  • Anyone on the team can make one by typing a single sentence.
  • The added cost is a few dollars a month on top of tools they already had.
Read the full story ↓

The full story

Where they were stuck

Their customers didn't want to read help articles. They wanted to see how things work. Fair enough. But that meant a video for every article, and there were several hundred of them.

Making one by hand is slow. Someone clicks through the product while recording, writes and records a voiceover, edits it, then adds captions. Then the product changes, a button moves, and the video is wrong again.

For a small team without a video editor, that is a treadmill you never get off.

The shift in thinking

A help article already describes every step. If the steps are written down, AI can follow them, record them, and follow them again whenever the product changes.

So we stopped thinking about making videos, and built something that makes them.

What actually happens

It records. Claude reads the help article, opens Chrome, signs into a demo account and clicks through each step while the screen is recorded. Every frame is the real product.

It narrates. An AI voice from ElevenLabs reads one line per step, in the same voice across the whole library.

It edits. Remotion, a tool that edits video in code, adds a zoom on each click, a visible cursor, step labels and title cards, and cuts out the loading waits.

It checks itself. Whisper listens back, confirms every spoken word matches the script, and writes the captions.

One recording gives three versions: a guided one with zooms and labels, a plain full-screen one, and a quick captions-only cut for social.

The toolkit

  • Claude runs the whole process, from reading the article to driving the browser
  • Chrome is the stage where the product is recorded
  • ElevenLabs provides the narration voice
  • Remotion edits the video in code
  • Whisper checks the audio and writes captions, free and running locally

Made for the whole team, not one specialist

A tool only one expert can operate is just a new bottleneck. So we packaged it for people who don't write code. Setup happens once per computer: unzip the kit, paste one line, answer a few questions. The installer does the rest.

After that, making a video is a sentence. Open the Claude app and type something like "make a video about exporting a report". A one-page guide covers everything else.

When the product changes, they rerun the same request and have a fresh video in about 15 minutes.

What it costs

No new hires, no video editor, no production house. On top of the Claude plan they already used, the added cost is a voice subscription that starts at a few dollars a month, plus a modest editing licence once more people run it. Whisper is free.

What to take from this

The tools matter less than the pattern. Look for the work your team keeps redoing by hand, the kind that goes stale the moment something changes. That is usually where AI pays off fastest, and where a small, well-packaged setup beats a new hire.

Got work your team keeps redoing by hand?

Start a conversation