In 2013 I worked as an applied anthropologist at a market research firm in Mexico City. One of my jobs: when a TV network premiered a telenovela, I watched the audience, not the show.
While each episode aired, I followed what people wrote on social media in real time. How they felt about the characters and the story, and how that changed from scene to scene. The network learned how people felt while it was still happening, not weeks later in a survey.
That method has a name: netnography, ethnography of online communities. Thirteen years later, now as a frontend engineer, I used it again for a UX case study on a Mexican mobile carrier. No TV this time: 55 public comments from app stores and a Facebook community.
It turned out to be the most useful research I did in the whole project. And any developer can do it this Friday afternoon.
Why it works
Surveys ask people questions. Reviews and community posts are people talking without anyone asking. Nobody writes a one-star review to be polite. They write it at the exact moment something failed, in their own words, often with details you'd never think to ask about.
That's what I learned in focus groups too: what people say when you ask them is not always what they do. Unprompted conversation gets closer to what they actually do.
How to do it, step by step
1. Pick where people already talk. App Store and Google Play reviews, public Facebook groups, Reddit, your own support tickets. Start with public reviews: they're the easiest and raise the fewest ethical questions.
2. Collect 50 to 100 recent comments. Six months back is enough. Copy them into a spreadsheet, one per row.
3. Paraphrase, don't copy. Write what each person says in your own words, without names. That protects them, and it forces you to understand the comment before you classify it.
4. Code each comment by theme. Start with no categories; let them appear as you read. After 15 or 20 comments, patterns show up. In my case: "the app hides the basics", "I paid but it doesn't work", "support never answers", "I don't trust this".
5. Count. Counting turns impressions into evidence. "Users struggle with login" is an opinion. "In 6 of 55 comments, the registration banner hid the login button" is a finding a product manager can act on.
6. Stop at saturation. When new comments stop producing new themes, you have enough. For me that happened around comment 45.
7. Go see it yourself. Every theme is a hypothesis. Open the app and try to reproduce it. When I did, the registration banner was there, asking me to register a line I had already registered.
What I found
Here's what 55 comments looked like once coded:
| Theme | Comments |
|---|---|
| The app hides the basics (balance, usage, login) | 12 |
| Paid or completed, but not working | 8 |
| Support: slow, bots, no human | 7 |
| Trust and security | 5 |
| Unclear plan conditions | 5 |
The biggest theme wasn't a bug. It was information hierarchy: promotions came first, and what people opened the app to do came second. No crash report would ever show you that.
The limits (say them out loud)
- Reviews skew negative. Happy people rarely write. Netnography shows you where it hurts, not how many people hurt.
- It's not a replacement for talking to users. It tells you what to ask them.
- Private groups need permission. If a community is closed, ask the admins or stay with public sources.
- Never publish anything that identifies someone: no names, no screenshots with faces, no exact quotes that can be searched back to a person.
The real lesson
In 2013 I read audiences so a TV network could adjust a story. In 2026 I read reviews to figure out why people couldn't find their balance. The method didn't change. What changed is that now I can also fix what I find.
That's the case I'd make to any developer: you don't need a research team to start listening. You need a spreadsheet, an afternoon, and the discipline to count before you conclude.
You can read the full case study, with the redesign and prototypes, on my site: Activated doesn't mean working.

![Perception and Sensor Fusion — Chapter 05: IMU and Inertial Perception [KR]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fx5cdjbwd8758flk932no.png)
