HCI Research Methods

How We Built aResearch Questionfrom Scratch

On literature dives, dead ends, and learning to ask the right thing.

For one of my HCI courses, my partner Ivy and I were given a deceptively simple task: come up with a UX research proposal. No dataset. No predefined topic. Just, find something worth studying.

What sounds simple is, in practice, one of the hardest things I've done in grad school. Because before you can design a study, you have to earn your question.

01, Starting With a Hunch

Teens, TikTok, and food.

We both kept circling back to teenagers and food. Not because it was an obvious HCI topic, it isn't, really, but because something felt genuinely unexamined there. TikTok had become the dominant way adolescents discover food trends, recipes, diet ideas, and health information. Around 14.25 million American teens are on the platform almost every day.

And yet most research on social media and teen health was either focused on mental health broadly, or on body image in isolation. The food angle felt specific enough to be tractable, and wide enough to be interesting.

But "teens and TikTok food content" is an observation, not a research question. We had a lot of work to do.

✦   ✦   ✦
02, The Literature Deep Dive

Three themes that shaped everything.

Before we could ask anything, we needed to understand what was already known. We spent a significant chunk of the project just reading, and what we found shaped everything that came after.

📱

TikTok as an educational tool

Short-form video is genuinely effective for information transfer, more engaging than text, more immediate than YouTube. But the format that makes information sticky also makes misinformation sticky.

⚙️

The power of algorithms

TikTok's recommendation system isn't neutral, it amplifies content based on engagement. For teens navigating food allergies or disordered eating, that's dangerous. But it also cultivates identity and creates real communities.

🍳

How teens actually engage with food content

Teens follow a progression: exposure → planning (saving videos, checking ingredients) → execution (trying the recipe) → habit formation or abandonment. Family dynamics, peer conversations, and cost all filter which trends make it into real life.

Everyone was studying what teens watched. Nobody was studying what happened next, over time.

✦   ✦   ✦
03, Getting to the Question

The gap was longitudinal.

Almost all existing research was cross-sectional. A snapshot. But food habits don't form in a snapshot. They form slowly, through repetition and reinforcement and negotiation with the people around you.

We kept asking ourselves: what would it actually mean to understand TikTok's influence on teen eating, not just in the moment, but longitudinally?

RQ 1, Health

How do adolescents' interactions with TikTok food content shape their understanding of "healthy eating" and food habits over time?

RQ 2, Design

What platform features do adolescents interact with to contribute towards establishing long-term healthy eating habits?

The second question felt distinctly HCI. It wasn't just asking whether TikTok is good or bad for teens, it was asking what the design is doing. Which features, specifically, are teens using in ways that build lasting behaviors? That's a question with actionable implications for the platform itself.

Getting to these two questions took a lot of iteration. Earlier versions were too broad, too narrow, or too leading. We kept stress-testing drafts against the literature, does this question actually address the gap? Can it be studied? Does it matter?

✦   ✦   ✦
04, Designing the Future Study

A year-long, three-phase plan

Once we had our question, we mapped out how we'd actually study it, if we had the time, resources, and IRB approval to do so.

Month 0

Baseline Interview

A 45–60 minute interview capturing each participant's starting point, their current food habits, how they use TikTok, what "healthy eating" means to them.

Months 1–9

Monthly Micro Check-ins

Short, low-burden email prompts asking participants to share 3 food videos that stuck with them and reflect on how they influenced their habits that month.

Month 12

Follow-up Interview

A closing interview mirroring the baseline, to see what had actually shifted over a year of engagement with TikTok food content.

We targeted 15–20 participants, ages 13–19, who actively consumed TikTok food content, recruiting through schools, community centers, Reddit, and TikTok itself, with a deliberate focus on diversity across gender, race, and socioeconomic background. The algorithm doesn't behave the same way for everyone, so a homogeneous sample would have given us a skewed picture.

We also knew we'd need both youth assent and parental consent. Asking minors to share a year of their media habits and health beliefs is intimate. That ethical weight shaped how we thought about the whole design.

✦   ✦   ✦
05, What We'd Do Differently

Honest about the weaknesses.

📊

More structured feature logging

RQ2 was the most interesting, and the hardest to capture with interviews alone. I'd want systematic behavioral data: saves, duets, follows, search behavior.

📅

Timeline questions

One year might not be enough. Meaningful behavioral change often takes longer to stabilize. A two-year window or more intensive 6-month design might work better.

🛡️

Harm protocols needed

Participants might deal with eating disorders or body image struggles. We needed clear protocols for concerning disclosures mid-interview.

💬

Talk to teens first

We built questions entirely from the literature. Even 3–4 informal conversations with actual teenagers early on would have pressure-tested our framing.

06, What I Took Away

The question is the product.

  • Getting the question right means understanding a field deeply enough to see its gaps, caring enough about a problem to close them, and being rigorous enough to keep asking: is this actually the right thing to ask?
  • Teens aren't passive consumers of an algorithm. They're active, thoughtful, sometimes contradictory users, learning about food, forming identities, negotiating with families and peers, on a platform that is simultaneously useful and risky.
  • Research that flattens that complexity doesn't do them justice. For now, I'm proud of the question we built. And I genuinely hope someone runs the study one day.

This proposal was developed with Ivy Tang as part of an HCI research methods course.