Ghost Frequencies: How Noise Collectives Are Haunting the Algorithm Without Getting Caught

The conventional wisdom about TikTok and experimental music goes something like this: the platform rewards the immediate, the digestible, and the familiar. Noise music doesn't trend. Drone doesn't go viral. Anything that requires more than four seconds of attention before delivering a dopamine hit is dead on arrival. The algorithm eats pop and spits out everything else.

Some people in the experimental underground looked at that conventional wisdom and decided to treat it like a puzzle rather than a verdict.

Over the last two years, a loose network of noise artists, drone producers, and experimental collectives across the US has been quietly reverse-engineering TikTok's recommendation engine — not to beat it at its own game, but to exploit its blind spots. The goal isn't virality. It's something more interesting: building genuine underground communities that the algorithm actively can't understand, commodify, or replicate.

The Algorithm Has Ears, But It's Half Deaf

To understand the strategy, you have to understand what TikTok's recommendation system is actually listening for. The platform's audio fingerprinting technology is sophisticated at identifying known sounds — it can match a song to its Spotify metadata in milliseconds and use that association to categorize content and predict audience behavior. What it's less good at is handling audio it doesn't recognize, sounds that don't resolve into clean spectral patterns, or content that behaves unpredictably in ways that confuse engagement metrics.

A collective based out of Chicago called Null Terrain figured this out more or less by accident in 2022. They'd been posting raw contact microphone recordings — the kind of harsh, unpredictable textures that sound like industrial machinery arguing with itself — and noticed something strange. Their content wasn't getting suppressed, exactly, but it also wasn't being categorized. The algorithm was passing it to small clusters of users seemingly at random, and those users, whoever they were, were watching the entire video. Completion rate was high. Shares were cryptic — people reposting with no caption, or with single characters, or with coordinates.

They leaned into it.

Intentional Artifacts as Anti-Commodity Armor

What emerged from that accident became a deliberate methodology that's since spread through several experimental collectives. The core principle: make content that is genuinely difficult for the platform's systems to process, while remaining compelling to a specific human audience.

This takes a few forms. Audio artifacts — intentional clipping, DC offset, subsonic content that creates visual distortion in the waveform display — confuse audio fingerprinting and prevent clean categorical assignment. A post that the algorithm can't file into "ambient," "experimental," or any other recognizable bucket gets treated differently than one it can sort. It floats. It gets tested against unusual audience clusters. Sometimes it finds people it would never have found through conventional tagging.

Metadata manipulation is another tool. Experimental producers have been using deliberately incoherent or contradictory tags — pairing genre descriptors that don't coexist, using misspellings that are common enough to exist in search but rare enough to indicate insider knowledge, burying content under tags from unrelated communities and then watching which ones actually surface engaged viewers. It's a form of A/B testing that the platform itself isn't running.

A Brooklyn-based noise collective called Terminal Drift took this further by developing what they describe as a "cryptic breadcrumb" system: each post contains a single piece of information — a frequency, a date, a partial address — that only makes sense in context of previous posts. The algorithm sees disconnected content with inconsistent tagging. The audience sees an unfolding narrative. Engagement from the right people is extremely high; engagement from everyone else is zero. The algorithm, optimizing for broad reach, keeps distributing the content to fringe clusters, which is exactly where Terminal Drift wants to be.

From Zero to Underground Movement

The results of these approaches are genuinely interesting, even by conventional metrics. Null Terrain went from a few hundred followers to a devoted following of around 40,000 over 18 months — modest by pop standards, enormous for a project built around contact microphone recordings and feedback loops. More importantly, they describe their audience as unusually engaged: people who show up to live events, who share content through encrypted channels, who reach out with their own recordings.

A noise project called Ferric Oxide, operating out of Portland, used a variation on the mistagging approach to build a following that skewed heavily toward visual artists, architects, and sound designers — an audience they'd never have reached through conventional experimental music channels. The algorithm, trying to figure out what Ferric Oxide was, kept testing it against unusual demographics. Some of those demographics stuck.

"We're not trying to hack the algorithm in the sense of gaming it for numbers," one member of Ferric Oxide explained in a voice message (they declined to do a recorded interview). "We're trying to make it confused enough that it puts us in rooms we couldn't get into any other way."

The Ethics of Algorithmic Judo

Not everyone in the experimental community is comfortable with this. There's a legitimate critique that engaging with TikTok at all — even adversarially — feeds a platform with genuinely troubling labor practices, data policies, and cultural effects. Some producers refuse to post there on principle, building their communities entirely through Bandcamp, encrypted group chats, and physical distribution.

The collectives using anti-algorithm tactics generally acknowledge the tension without resolving it cleanly. The platform is where a lot of young, curious people are. Refusing to engage doesn't make the algorithm better; it just cedes that space to content that's more amenable to commodification. Working against the platform's logic from inside it, they argue, is at least more interesting than boycotting from outside.

What's undeniable is that these tactics are producing something the algorithm wasn't designed to produce: real underground communities with genuine subcultural coherence. People who find Null Terrain or Terminal Drift through TikTok's fractured recommendation logic tend to be people who were already looking for something they couldn't name. The algorithm, broken in just the right places, is accidentally doing what good record store clerks used to do.

That's not nothing. In 2024, it might even be enough.