Sid's Blog

The Same Show, Different Links: A Content Problem Worth Solving

The idea of pebbles was simple: let people recommend what they love -- shows, podcasts, music, blogs or even a TikTok without the restrictions of a platform. But the moment we let users share any link from any source, things get tricky.

How do we recognize that different URLs are actually pointing to the same media?

To us humans, it's obvious when a Netflix and a Prime Video URL refers to the same show / movie. But to a computer they're 2 completely different strings. If we don't solve this, things start to break.

This post is just my brain dump trying to solve this problem. All the messy & maybe interesting ways I come up with to solve this without relying heavily on 3rd party APIs.


The Problem

To better understand the problem let's work with an example, consider the popular TV sitcom FRIENDS:

This results in noise, and that noise has real consequences for our app. The one I dread the most is that our app becomes a messy bookmark manager instead of a social layer for recommendations

Why not use 3rd party APIs

I don't want to rely on 3rd party APIs like TMDB or Spotify, because

Things I've considered

1. Manual Mapping

This is obviously not practical in any sense, but after spending hours trying to come up with something, I hit a “F**k it” point and genuinely considered mapping everything by hand. Just brute-force my way through a JSON file of known URLs and content pairs.

Needless to say, it was a fever dream. It doesn’t scale, it’s not fun, and thankfully, I’ve moved on.

2. Fuzzy Metadata Matching

This is the most promising approach so far — and likely the one we’ll start with.

How this will work

The more structured data we can extract, the better our chances of uniquely identifying the content.

Challenges

3. AI Intervention

Not my first choice, but sometimes helpful.

When the metadata alone returns a bunch of “close enough” matches, I can feed those into something like GPT with a prompt to score and rank them. Pick the top match, discard the rest. Lightweight, fallback-only, and ideally infrequent.

4. Community Corrections

The idea here is to let users flag duplicates or incorrect merges. Like Wikipedia but for recommended content.

But this:

So yeah, this one’s for Future Me, not MVP Me.

Why am I worrying about this right now?

I’ve been asking myself this a lot lately, why am I spending hours obsessing over this instead of just hacking together an MVP and moving to launch?

So I’m writing this down as a reminder for future-me when I inevitably ask the same thing again.

Pebbles doesn’t work with a hacked MVP - not the way I want it to.

If content shows up as duplicates, or if aggregation is half-baked, the product loses its core value. What should feel like a trusted space for recommendations from your friends and family ends up feeling like an over-engineered bookmark manager.

Getting this part right, even just a rough version of it sets the tone for everything else. And it unlocks a lot of exciting possibilities down the road:

It’s a foundational decision, not a feature. And I’d rather build it once with intent than duct-tape it later.


To Sum Up

This whole problem, figuring out when two different links mean the same thing might seem small at first. But it’s shaping a lot of the core decisions behind Pebbles.

We’re not just building another place to drop links. The goal is to make recommendations feel personal, relevant, and connected and that starts with treating content with a little more respect than just a URL.

Nothing’s finalized yet. Things will change. But thinking about this now gives Pebbles a better shot at becoming something people actually enjoy using not just another feed. (Been saying that since day one.)

If you’ve made it this far, thanks for reading my little brain fart.

If any of this sparked an idea, a strong opinion, or even just the urge to tell me I’m massively overthinking it, feel free to reach out. Always happy to chat.


hello@webxsid.com Linkedin

#link-aggregation #product-design #recommendation-systems