Imagine your social media feed as a hyper-attentive but slightly mischievous waiter at a restaurant. This waiter, let's call him "The Algorithm," is obsessed with keeping you happy in the immediate moment. If you glance at a spicy taco, he immediately brings you a mountain of tacos, then a taco volcano, then a taco parade, because you looked at the first one for a split second. He doesn't care if you actually wanted a salad later, or if you're trying to eat healthier, or if you just got distracted by the taco. He only knows what you did right now. This is how most social media works: it learns from your clicks and watch time to give you exactly what grabs your attention in the moment, often leading you down a rabbit hole of content you didn't really want to spend your time on.

But here's the twist: humans are complicated. Sometimes we want to watch funny cat videos, but other times we want to learn about history or avoid violent content. We have "reflective preferences"—what we actually want to see when we stop and think about it, rather than what we impulsively click on. The problem is that telling the waiter what you want is hard. Usually, you have to go to a back-office menu, fill out a long form, and hope the waiter remembers it next week. But by then, the waiter has already forgotten and is just bringing you more tacos because you looked at one yesterday. This paper, presented at the UIST '26 conference, explores a new way to fix this: a system called Compass that helps you guide your social media feed in real-time, turning that mischievous waiter into a helpful guide who actually listens to your deeper wishes.

The Problem: The "Taco Trap" and the Forgotten Menu

Social media platforms are designed to keep you scrolling. They use your behavior—what you click, how long you watch, what you like—to guess what you want next. This works great for keeping you engaged, but it often leads to a "Taco Trap." You might find yourself watching hours of video game content or doom-scrolling through news you regret later, simply because the algorithm is reacting to your split-second impulses rather than your long-term goals.

Researchers have tried to fix this before. Some systems let you build your own feed from scratch, but that's like trying to cook a whole meal from scratch every time you're hungry—it's too much work. Others give you a settings page where you can say, "I want more science videos" or "I want less violence." But these are like a menu you fill out once and then forget. You might set it up on Monday, but by Wednesday, the algorithm has drifted back to showing you only tacos because you clicked on one on Tuesday. You have to remember to go back to the menu and fix it, but by then, you're too busy scrolling to care.

The Solution: Compass, the "In-Situ" Guide

The authors introduce Compass, a tool designed to bridge the gap between what you do (your behavior) and what you actually want (your reflective preferences). Instead of making you go to a settings page, Compass meets you right where you are: while you're scrolling.

Think of Compass as a friendly companion walking beside you in the restaurant. Instead of waiting for you to finish your meal and fill out a complaint form, Compass whispers, "Hey, you just watched three video game clips and then looked annoyed. Do you want to tell the waiter to stop bringing those?" or "You paused on that science fact video for a second longer than usual. Want to ask for more of those?"

Compass does this through in-situ reflections. It pops up lightweight notifications right in your feed, asking simple questions like, "Do you want to see more of this?" or "Is this what you're looking for?" These interactions happen while you are browsing, so you don't have to stop and go to a separate settings page. It's like having a conversation with the waiter while you're eating, rather than writing a letter to the manager later.

How It Works: The Magic of Simulation and Direct Action

Compass doesn't just ask questions; it takes action to make your feed match your answers. It uses two clever tricks:

1. Simulating Behavior: When you tell Compass you want more science videos, it doesn't just hope the algorithm listens. It secretly "pretends" to be you. It simulates you liking, watching, and searching for science content in the background. This tricks the algorithm into thinking you are really interested in science, so it starts showing you more of it.
2. Direct Manipulation: Sometimes, the algorithm is slow to change its mind. In those cases, Compass gets a bit more hands-on. It directly edits the feed on your screen. If you said "avoid video games," Compass will hide those videos before you even see them. If you said "show me more history," it will sneak a history video right into your feed, replacing a random one.

This happens automatically. You don't have to constantly monitor your feed or manually adjust settings. Compass watches your behavior, asks for your input when it's unsure, and then quietly steers the feed back on course.

What They Found: More Reflection, Less Drift

The researchers tested Compass in a 10-day field study with 15 participants. They compared people using Compass against a group using a standard system where they had to manually set preferences and trigger changes themselves.

The results were clear:

• More Frequent Adjustments: People using Compass refined their preferences an average of 11.50 times per person. The group without Compass only did it 0.14 times on average. Basically, the Compass users were constantly tweaking their feed to match their real interests, while the others forgot to touch it.
• Better Alignment: The Compass users saw content they actually wanted to see 49.3% of the time, compared to just 20.3% for the baseline group. They also spent 58.7% of their watch time on content they liked, versus 21.0% for the others.
• No Loss of Fun: Crucially, this didn't make social media feel like a chore. Both groups rated the system as "Excellent" on usability (with scores of 84.1 for Compass and 83.2 for the baseline). The participants felt they were still browsing casually and discovering new things, but with a bit more purpose.

One participant noted that Compass gave them the "dopamine rush" of finding something good, but it was something they were actually interested in, not just random noise. Another said it felt like the system was "dramatically better" after a few sessions, learning their true interests over time.

The Takeaway: A Continuous Conversation

The paper suggests that the key to a healthy social media experience isn't a one-time settings change; it's a continuous conversation. By bringing reflection into the moment of browsing, Compass helps users stay true to their goals without fighting the algorithm. It suggests that we don't have to choose between a fun, easy feed and a feed that respects our values—we can have both if the system is designed to listen and adapt in real-time.

While the study was small (only 15 people) and done on a desktop version of YouTube Shorts, the results are promising. They show that with the right tools, we can take back control of our feeds, turning the "Taco Trap" into a dining experience that actually satisfies our hunger for the right kind of content.