22d • General discussion
Experiment #1: How NOT to do it
Welcome to the Lab. Here's how this place works: I run marketing experiments on my own books, in public, and I show you everything. The wins, and the ones that blow up in my face. You watch, you take what works, you skip my mistakes.
So let's start with one that tanked...bad!
THE EXPERIMENT
For the Kickstarter on book 2 of my Phillip Aisling series, "Phillip Aisling and the Lucid Multiverse," I built an AI trailer. Teaser scenes pulled straight from the book. I used AI for the visuals, the voices, the music, and the sound design. It took hours of work over weeks to make...and honestly? A lot of it came out good. The voices held up. The sound held up.
Check it out! https://www.youtube.com/watch?v=xQ5BVMaVGqc
THE DATA
The comments did not hold up.
A sample:
"don't use AI in your ads, it makes you look lazy" (13 likes)"
Using AI animation for your ad ensures it will do poorly."
"AI garbage."
"AI scum."
"Nice try clanker."
And then the one that actually mattered, the one I keep thinking about:
"AI ads always make me think the book is written by AI too."
That's the real finding. It wasn't that the trailer was bad. It's that detectable AI in the marketing made people assume AI was in the writing, and they walked before reading a single page.
So the lesson isn't "don't use AI to market." It's "don't let your AI marketing look like the kind of AI people are afraid of." That distinction is what this whole lab is going to chase.
A couple more signals:
- Platform mattered. Facebook (older crowd) was brutal. TikTok barely flinched. Same video.
- The tell was realism. The visuals reached for lifelike and hit the uncanny valley instead. That's what flipped the switch from "cool" to "gross."
WHAT I'M TESTING NEXT
A few hypotheses coming out of this:
- AI that winks at being AI and makes fun of itself tends to do fine. (think talking dogs or ridiculously obese gymnasts)
- AI pretending to be real gets punished. So: lean stylized and cartoony, kill the realism, dodge the uncanny valley entirely.
- If a character talks to camera, the lip sync has to be flawless or it reads fake instantly. When in doubt, voiceover over visuals instead of mouths moving.
- Better models than the ones I started with, which are already outdated.
- Segment by platform instead of blasting one cut everywhere.
That's experiment #1. Follow along and steal what works.
0
0 comments
Experiment #1: How NOT to do it
powered by
skool.com/author-growth-lab-6177
Self-published author documenting live experiments using AI agents to market and sell more books. Follow along, steal what works!
Suggested communities
Powered by