How do we trust anything on the internet?
, 5 min read

Back in December, Aditya Gautam and I had a conversation that’s stayed with me ever since. We were talking how deeply skeptical we’ve become of anything shared on social media or the internet at large, we are distrustful of trivial of things and impassive of grimmest of news.
That conversation led us to a pressing question: how do we trust anything on the internet?
News, after all, is a sensitive and powerful force and like all forms of mass media, it has firmly embedded itself into the fabric of the internet. This shift has brought undeniable advantages such as: real-time updates, global access, and interactive, two-way communication. Yet, this democratization of news has also opened the floodgates to agenda-driven narratives and rampant misinformation.
Think about it: how often have you come across a headline that felt too inflammatory or a reel showing a disturbing incident like a supposed food ban in schools, a politician allegedly making an outrageous comment, or a celebrity being 'canceled' only to find a completely different story unfolding in the comment section? In today’s information ecosystem, a single tweet, trending image, or viral video can shape public opinion within minutes. And yet, verifying whether any of it is real or fabricated remains a difficult and time consuming task.
Humanity's struggle with fake news isn’t new it’s centuries old. Disinformation and propaganda have long served as tools of control, from political scheming in ancient Rome to ideological campaigns of church. Yet despite the evolution of media, our primary defense remains remarkably traditional: human intervention.
This classical method—interrogating sources, verifying claims, and cross-checking with witnesses still forms the backbone of modern fact-checking. Even today, most fact-checking organizations continue to rely on these foundational practices.
But here’s the challenge: in the current digital age, this model is under immense pressure. The sheer volume of content, the lightning fast pace of information circulation, and perhaps most disruptively the rise of Generative AI has made it incredibly difficult (some might say nearly impossible) to keep up.
With all this context in mind, let’s return to the conversation that started it all:
Can technology help us solve the problem of fake news detection?
Can we build automated systems that reliably detect and flag misinformation?
At first glance, these questions might seem straightforward but they’re anything but. Solving misinformation through technology is a deeply complex challenge, not just a coding exercise.
To even begin building a solution, we need to understand the entire pipeline. We can’t write code around vague concepts. Fake news isn't just about false text; it's about nuance, context, framing, intent, and source credibility. And all of that requires a structured, multi-step approach.
So, what could the pipeline of a hypothetical fake news detection system actually look like?
Let’s think like a methodical person not a researcher, just someone who really wants to be sure something is true before believing or sharing it.
Say you see a post online that makes a big claim. Maybe it's a video of some protest or an article accusing someone of something serious. What do you do?
First, you carefully read or watch the content. You try to understand what it’s actually saying not just the headline, but the details.
Next, you pull out the main claims. Not the emotional bits, but the stuff that can actually be checked. For example: "This happened in Delhi in June 2024." That’s a claim.
Then, you go to Google (or any search engine) and look up that claim. But here’s the trick you don’t just look at the first link. You sort the results by date, from oldest to newest, to see when this story or video first showed up online.
Now, you compare the original source with the post you saw. If the earliest version of the video was posted in 2018 with a totally different context, well, that tells you something important.
Just like that, we arrived at the core operation of our detection algorithm: Reverse Search.
For instance, imagine a video claiming to show a missile strike in Iran. At first glance, it seems real. But once we perform a reverse search analyzing the video frame by frame we discover it was originally uploaded six years ago on a YouTube channel showcasing U.S. bomb testing footage.
Boom. The context changes completely.
Reverse searching isn’t just helpful it could be one of the strongest methods we have today to algorithmically verify the authenticity of online content.
And rightly so most major fact-checking organizations already spend a significant amount of time on reverse searching to trace and verify the original sources of media.
But we didn’t stop there. Since our goal is to perform this step recursively automating the search and timestamp analysis for every claim in a piece of content we can call it:
Automated Reverse Search (ARS).
ARS lies at the heart of our system. It mimics human verification but at machine speed and scale.
We kept going back and forth on this idea, debating its technical and ethical edges. What we eventually landed on was quite fascinating: a reverse search mechanism that doesn’t just stop at finding the source, but actually feeds its results into a large language model (LLM) for deeper analysis.
The LLM would then examine the context, compare content across sources, and help categorize the information as likely true or false.
In essence, it’s a layered system:
- Trace the origin with reverse search
- Contextualize the result with LLM analysis
- Classify the claim based on both evidence and nuance
So, what’s the best way for a fake news detection system to function in the real world?
It needs to operate constantly on the web ideally monitoring most (if not all) of the pages a user visits for potentially misleading or fake content. And then it hit us:
What better way to deploy this than as a browser extension?
That’s when the core structure of the project began to take shape.
More than just a fake news detector, this tool can be thought of as a cognitive companion something that gently nudges us to pause, question, and verify before we share. In a time when echo chambers amplify our beliefs and algorithms feed us what we already agree with, this research project aims to interrupt that loop. To provide not just a warning, but a window a way to step outside the bubble and confront alternative perspectives grounded in evidence.
What strategies do you use to verify what you see online?