Recording software demos is a special kind of torture. You stumble over your words, the mouse cursor disappears, and you’re never quite sure if you’re focusing on the features that actually matter to the user.
Usually, getting feedback means pestering a coworker, your neighbor at coworking, or paying a consultant. But there’s a better way. You can use Google AI Studio to analyze your screen recordings.
The Gemini models have the ability to “watch” video, “hear” audio, and reason across massive context window. It’s surprisingly good at acting like a grumpy creative director and a marketing intern rolled into one.
Here is exactly what you need to do to get a full critique (and your YouTube metadata) in under two minutes.
The Workflow
Step 1: Get into the Studio – Head over to aistudio.google.com .
Step 2: Select the Heavy Hitter – On the right-hand panel, under the “Model” dropdown, ensure you select Gemini 3 Pro (models/gemini-3-pro-preview). You want the smartest multimodal model available for video analysis.
Step 3: Upload your Video or Provide a YouTube URL – Click the + button (Add to prompt) and select Upload to drop in an MP4 of your app walkthrough, or paste a YouTube URL directly. Don’t worry about file size or length because Gemini 3 Pro has a 1M token context window.
Step 4: The Prompt – Add prompt this into the text box and hit Run:
- Give me suggestions on how to improve showing this app
- Give me a video summary
Optional, add - suggest improvements to the app
The Result
I tried this with a raw, unedited recording of a construction schedule analysis app which is published at https://www.youtube.com/watch?v=3k-5zJbVJso . The video was rough, with some silence while loading, small text, and a wandering mouse cursor.
Here is what Gemini 3 Pro gave me back:
Gemini 3 Pro:
Suggestions for Improvement:
- Edit Out “Dead Air”: Around the 4:50 mark, there is a significant pause while the system generates the report. You should cut this or speed it up (2x) in post-production.
- Zoom and Focus: The text on the sidebar settings is quite small. When you are explaining the “Strict Mode” vs “Inference Mode” toggle, zoom in on that section so mobile viewers can read it.
- The “Hook”: You started with a dry CSV file. Try starting with the final result (the Word Document report) to grab attention immediately, then rewind to show the process.
- Mouse Visibility: Your cursor gets lost against the white background. Use a highlighter tool or click effects to make navigation easier to follow.
Video Summary: Title: BSE Schedule Analysis Platform Walkthrough Duration: 6:15
Overview: A walkthrough of the BSE platform, demonstrating how to use AI to analyze complex construction CSV data and generate inference-based reports.
Chapters:
- 0:00 – The Problem: The user shows a raw CSV construction schedule, illustrating the difficulty of manual data entry.
- 1:45 – Analysis Mode: The user demonstrates the “Inference Mode” using a paid model, allowing the AI to deduct manpower needs from activity descriptions.
- 3:20 – The Stress Test: A comparison is shown where “Strict Mode” fails to find manpower data, but “Inference Mode” successfully generates a trade-by-trade breakdown.
- 5:00 – Export: The final analysis is downloaded as a clean Microsoft Word document.
Why do this?
First, you get an objective critique. Gemini didn’t just transcribe; it saw that the text was small and felt that the pause at 4:50 was awkward. It helps you fix pacing issues before you ever hit “Publish.”
Bonus productivity hack:
Take that Video Summary response and copy-paste it straight into your YouTube description. It’s already formatted with an overview and timestamps for chapters. You just saved yourself 15 minutes of writing and optimized your video for SEO. Plus, it adds chapter headings to the video slider under the video.
If you are building in public or recording demos for clients, this workflow is a superpower.
Leave a Reply