Brand trust · Testimonials · Consumer apps · AI ethics
Why I never faked a testimonial, even with AI on the table
When I was leading Hallow in Brazil, we had AI available for pretty much everything. Copy, images, video, voice. If I wanted, I could have generated a smiling woman in São Paulo saying the app changed her prayer life, in perfect Portuguese, in about ten minutes. And it would have looked great.
I never did it. Not once. Every testimonial we used came from a real person, with a real story, who said yes to sharing it.
People sometimes ask me if that slowed us down. Of course it did. Collecting real stories takes time. You have to find the people, talk to them, ask permission, go back and forth on wording, and sometimes the best story you heard can't be used because the person isn't comfortable. A synthetic testimonial never says no.
But let me tell you why I'd make the same decision again tomorrow.
First, because of what we were selling. Hallow is a prayer app. If we fake a person praying to convince someone else to pray, the product is broken at the core. It doesn't matter how good the image is. I always thought about the person on the other side of the screen, maybe going through something really hard, deciding to trust us with something very intimate. That person deserves the truth.
Second, because trust doesn't scale the way content scales. You can produce a thousand fake testimonials in an afternoon. You can't produce trust in an afternoon. And when people find out something was fabricated, and they usually do, you don't lose one post. You lose the credibility of every true thing you said before.
Third, because it's also becoming a legal question. In 2024 the FTC in the US finalized a rule against fake reviews and testimonials, and it explicitly covers ones generated by AI. So this stopped being only a values conversation. If your growth plan depends on synthetic social proof, you have a risk sitting in your marketing calendar.
And here's what surprised me the most. The real stories performed better.
We ran a campaign with the Christ the Redeemer Sanctuary in Rio, projecting a message of prayer for Pope Francis on the statue when he was in the hospital. It went viral, more than 100 million views. Nothing in it was synthetic. Real people praying, a real moment, a real place that means something to every Brazilian. You can't generate that feeling. People felt it because it was true.
I see the same thing with user stories. A real testimonial has little imperfections. The person pauses, says something unexpected, uses a word you would never write in a brief. That's exactly what makes someone stop scrolling. AI is very good at producing the average testimonial, and the average testimonial is the one nobody remembers.
So where does AI fit in this? A lot of places, actually. The line for me is using it to replace the person. At Hallow and in the companies I work with now through Fitch & Co, AI can help you:
- Organize hundreds of reviews and find the themes people repeat the most
- Spot which real stories match which audience or campaign
- Transcribe and translate interviews so a story from Brazil can reach someone in the US
- Draft the outreach message asking a user if they'd share their story
All of that saves hours and keeps the human voice intact. The machine helps you find and carry the story. The person is still the story.
If you're a leader making this decision right now, my suggestion is simple. Write it down as a rule, before someone on your team is under pressure at 11pm with a campaign launching the next morning. Because that's when the shortcut looks reasonable. A clear rule protects your people from having to make that call alone.
Ours was: only authentic testimonials, from real people, nothing fabricated. It cost us some speed. It gave us something much more valuable, which was the ability to stand behind every single thing we published.
Growth matters. I've spent my whole career chasing it. But I want growth I can look at without being embarrassed. And honestly, I think our users could tell the difference.