Thumbnails Are Good. Video Thumbnail Summaries Could Be Better
I've had a somewhat unconventional idea about taking a new step in driving engagement and content discovery.
Thumbnail images are good. They signpost your content, and if done well, give you an idea of what to expect.
However, more often than not, you're somewhat limited - showing only one idea or a brief teaser, typically in a strapline.
The point of trying this is not to see if we can do it, the main CVO here is will it improve engagement in the form of improved click-through rates. That is the fundamental objective.
The Thumbnail Limitation and The What-If Experiment
What if you could use AI to create a short summary video of your textual article?
Someone's looking at your website - desktop, mobile, whatever. They hover over the thumbnail image, and it plays a video that provides an executive summary or a more engaging depiction of the article.
Content Is Going Multimodal
One of my observations around AI: content is just going to become increasingly multimodal.
The separation between text-to-text, audio-to-audio, and video-to-video is going to get eroded. Those lines aren't just going to get blurred. It's all just gonna morph together.
And I think that would be genuinely useful.
If you're creating content, you want people to view it. To convince someone to give you their time - to actually engage with what you've created - it may be helpful to have something that uses that multimodal cue to provide enough information to validate that this content is worth their time.
I'm going to try to build this.
Be crazy if they all autoplayed - your website would look nuts. But to allow someone to swipe on an article if they're on the phone and it plays?
For a start, it'd be a very interesting exercise. Super interesting to see how quickly you could get there. And it would be even more interesting to see what it actually drives in terms of engagement, so we will need to measure that.
Testing The Impact With Real Users
Here's the thing, though: I can build the technical system, but I will also need to put in place robust measurements to validate whether it actually delivers value. The CVO (Customer Value Outcome) is will this improve the click-through rate from lists?
The real test is whether a 60-90 second summary video actually helps them decide what's worth reading.
Are they more likely to engage with the full article after watching the summary? Or is the summary itself enough? What information density actually works?
That's human-in-the-loop methodology applied to feature development: technical capability is necessary, but user validation determines whether you've solved a real problem or just built a clever system.
Where to start
While the basic idea is doable, the real intent here is to improve the click-through rate by providing more information and greater confidence in the article. The current set-up of thumbnail + title does some of this, but it may not be enough. A tiny 3–6 second micro-video on hover adds that "Oh, this article will help with X" moment.
However, before we invest too much time, we need to validate that we'll actually get an impact. For that reason, to start, we will:
- Use the existing thumbnail (so we're not creating extra work)
- Have AI polish the current exec summary (making it punchier and easier to scan)
- Render the video with the thumbnail as background and have the exec summary fade through
- Use Make.com to coordinate the automation and have an automated video generator render everything
- Tag everything up using Google Tag Manager and see if we get an uptick in click-through rate
If we receive a positive signal, tweaking the Make.com flows should be straightforward to test whether fancier motion graphics can further boost the results.
We'll store the videos in Google Cloud Storage and link them in our content management system.
Think of it like those recipe videos that show you the finished dish before you commit to reading the full recipe; you know immediately if it's what you're after. Success here means more people clicking through to articles that genuinely help them, and fewer people wasting time on content that's not quite right.
We should know pretty quickly if this makes a difference. Other than time, the direct cost of the experiment should come in under £5 easily. What's not to like?
Yet Another HITL Experiment
That is yet another human-in-the-loop experiment as the lines between content formats are dissolving. It's best to experiment with what that actually means in practice.
The next step, if the signals are positive, is to level up the video thumbnail and have it tell a story in motion rather than just the text on thumbnail video described here.
