Hey everyone! I'm building CreatorGrowAi, an AI tool that predicts video performance and analyzes audience sentiment before and after you publish.
While building it, I keep running into the same question: YouTube Studio (and most analytics tools) tell you what happened, but almost nothing tells you what's about to happen or what to do differently next time.
Curious what this community thinks:
- If you could know one thing about a video before publishing it, what would it be? (Title strength? Thumbnail CTR? Topic saturation?)
how does the performance prediction actually work, like is it based on similar channels in your niche or something else entirely
@ayazic10 Good question! It's a combination of a few signals, not just niche comparison:
- Your own channel's historical performance (what's worked for you specifically, since that varies a lot creator to creator)
- Title/topic similarity against high and low performers in your niche
- Comment sentiment trends from your recent uploads
So it's less "here's what similar channels do" and more "here's what your data + your niche's patterns suggest." It's directionally useful rather than a guarantee, YouTube's actual algorithm is a black box even to YouTube, so we're not claiming to predict exact view counts. More about flagging "this title pattern has underperformed for you before" or "this topic is trending in your niche right now."
Happy to go deeper on any part of that if you're curious!
Curious how accurate the performance prediction actually is in practice — like, does it base that on your own channel's history or pull from a wider dataset of similar creators?
One thing I'd love to see is a built-in thumbnail preview score that predicts click-through rate using your actual title and potential thumbnail before you publish. The prediction model is great for titles, but pairing it with visual analysis would close the loop and save so much A/B testing time.
@z_ezgi10358 There is the feature already. I will share another video dedicated to details generator.
The prediction feature actually felt useful, not just a hype machine. Ran three old video ideas through it and it called the underperformer correctly, which was a nice surprise.
The performance prediction feature is surprisingly accurate on my own channel data, gave it a few uploads I'd already posted and the scores lined up with reality. The title generator feels less generic than the usual AI stuff too.