My last email was a bit different than usual.
That’s because you just took part in a randomized field experiment!
This is an experiment we're doing in collaboration with The Fox School of Business (Temple University) and The Wharton School (University of Pennsylvania), and it's exactly the kind of rigorous, peer-reviewed research that Science Says is built on.
Two things are happening simultaneously in video communication right now.
First, content is becoming more personal than ever. Videos that greet you by name, reference your past behavior, with recommendations tailored specifically to you.
Second, AI is becoming more human than ever. Realistic faces. Natural voices. Conversational delivery that can feel genuinely indistinguishable from a real person.
This experiment sits exactly at the intersection of those two trends.
We wanted to understand how two factors shape trust and engagement when (1) the video is personalized, and (2) when the hyper-realistic (or close enough) AI’s identity is disclosed.
This study will answer some important questions about how people respond to the next generation of digital communication and content marketing.
In the video you just watched, I share some statistics about Science Says insights. While the video looks like me talking, it is actually an AI avatar of me, trained on me and my voice (seeing myself like that still feels pretty weird, but it's not so off 😅 ).
You were randomly assigned to one of four groups. Each group received a different version of the same video. The two things we changed were:
That gave us four groups (all of them were an AI version of me):
Group 1 (Control): No personalization. No disclosure label.
Group 2: Personalized content. No disclosure label.
Group 3: No personalization. Disclosure label shown.
Group 4: Personalized content. Disclosure label shown.
You received one version only - assigned by chance, like any rigorous experiment.
We didn't rely on surveys or self-reports. We tracked real behavior:
Did you click the link at the end? This captures whether the video sparked genuine interest.
Did you watch for at least 3 seconds, and for how long? These are standard attention metrics in digital media research - they tell us whether the video held your attention at all.
Why are we giving away 5 x $100 Amazon gift cards?
The larger the sample size, the larger the chances that this experiment will succeed. If the sample size is too small (not enough people watch and engage with the video) there is the risk that even if there is an effect we would not be able to see a statistically significant difference.
So we needed to maximize the sample size. Temple University kindly funded this gift to maximize the chances that subscribers would actually click and watch rather than mark it for later and forget. Five randomly selected viewers will receive $100 each, and I'll personally email the winners.
If this were simple, it wouldn't need a study.
The existing research gives us conflicting predictions. Personalization tends to boost engagement. AI disclosure tends to reduce it. Human-like design increases warmth, but can also trigger privacy concerns. Past findings pull in different directions, and none of them were tested in a setting quite like this one.
Here's what makes hyper-realistic AI different from anything studied before.
Most research on AI disclosure looks at obvious bots like text chatbots and voice assistants where users already know they're talking to a machine. In those settings, disclosure is essentially a reminder of something people already assumed. And it consistently reduces engagement, because it weakens the sense of human connection.
But when an AI looks and sounds indistinguishable from a real person, disclosure isn't a reminder. It's a revelation.
And that revelation could go two ways:
Which effect dominates? Does knowing it's AI make personalization feel helpful or invasive? Does disclosure hurt engagement or protect it?
That's exactly what we're testing.
It has been extremely interesting to run this experiment, and I couldn’t have done it without you.
Rigorous field experiments like this one are rare - and only possible because real people like you take part in them in real conditions.
Whether you clicked, watched, or scrolled past, your behavior contributed to research that will help us understand trust, transparency, and AI in ways that actually matter.
I’m looking forward to sharing the findings with you when the study is complete.
Want to ask anything about the experiment, or share your thoughts about hyper-personalized AI-generated content generally? Send me an email at: team@sciencesays.com. I read every response. (I promise it will be the real Thomas typing back, not AI Thomas from the video!)
