Spotify just confirmed something a lot of people had already suspected. Over the past year, the platform pulled down more than 75 million spam tracks, the vast majority churned out with AI tools and uploaded for one reason: to skim royalties away from real artists. Spotify's Sam Duboff estimated that around 100,000 new songs are uploaded every single day, and industry figures suggest something close to 44 percent of everything landing on streaming platforms now carries some kind of AI fingerprint. To be clear, Spotify isn't banning AI music. It's going after the specific slice built to game the system rather than to actually be heard.

That number, 75 million, is really just a way into a much bigger story. Music happens to be the front page today, but the same flood is hitting text, books, images, and video at the same time, and once you see the pattern repeat across every medium, it gets a lot easier to guess what comes next.

Just how big is this flood, medium by medium?

Text and the open web. A joint study from Stanford, Imperial College London, and the Internet Archive found that by mid-2025, 35.3 percent of newly published websites were AI-generated or AI-assisted, with 17.6 percent fully synthetic, no human hand in it at all. Separately, both Cloudflare and the security firm Imperva have reported that automated bot traffic now outweighs human traffic across the web: Imperva puts it at roughly 65 percent of all requests in 2025. Meanwhile, Google says publisher referral traffic has fallen by about a third over the same stretch, which points to something uncomfortable: AI summaries in search and AI-written pages are cannibalizing the same ecosystem from opposite ends.

Books. Amazon's Kindle Direct Publishing has been flooded with AI-generated titles for long enough that Amazon finally capped uploads at three books a day per account and now requires AI disclosure. Even with that cap in place, estimates put AI-authored ebooks at 10,000 to 40,000 new titles a month. It got serious enough that AI-written mushroom foraging guides, some of them listing poisonous species as edible, made it onto the platform before being pulled. In response, the Authors Guild launched a "Human Authored" certification so legitimate writers have some way to stand out from the noise.

Video. YouTube spent 2026 tightening its rules around what it calls inauthentic content: generic, repetitive uploads, emotionally manipulative videos, and synthetic AI personas pretending to be real people. In one enforcement wave, the platform terminated 16 channels with a combined 35 million subscribers. CEO Neal Mohan named the fight against AI slop a top priority for the year, and YouTube now looks at patterns across an entire channel rather than judging each video on its own.

Social platforms. Some estimates put AI-run accounts at 30 to 45 percent of active profiles on X, Instagram, and TikTok. One observer of this trend even launched Moltbook this year, a social network built exclusively for AI agents, no humans allowed. It sounds like a joke until you realize it's more of an honest admission: a slice of the internet has already gotten there.

Why is this even happening?

None of this is really about AI being malicious. It comes down to incentives. Every one of these platforms rewards volume in some form, whether through ad revenue, royalty pools, algorithmic reach, or search visibility, and generative tools collapsed the cost of producing that volume to nearly zero. A song that once needed a studio session now takes a text prompt. A travel guide that once took weeks of research now takes an afternoon. Once production gets that cheap, flooding the system with low-effort content becomes a rational, if cynical, move for anyone chasing a slice of a finite payout, whether that's streaming royalties, ad dollars, or a spot on a bestseller list.

So what are platforms actually doing about it?

Two responses seem to be converging across the industry. The first is detection and removal, which is what Spotify, YouTube, and Amazon are all doing, each in their own way: dedicated teams and models built to spot low-effort, mass-produced content and cut it before it dilutes payouts or trust.

The second, newer response is provenance and labelling rather than outright removal. The Coalition for Content Provenance and Authenticity (C2PA) now counts more than 6,000 member organizations, including Google, Microsoft, Adobe, Meta, OpenAI, Sony, and the BBC, all embedding signed metadata into content to show where it came from and whether AI touched it. Google's SynthID has already watermarked over 20 billion images, and TikTok says it has labelled more than 1.3 billion videos with AI provenance data. Regulation is catching up too: the EU AI Act's Article 50 disclosure rules begin enforcement in August 2026, and California's SB 942 is already in effect, both pushing for clearer labelling of AI-generated content.

Music's own trade bodies, including the IFPI, the RIAA, and the Recording Academy, have rolled out a voluntary labelling system that distinguishes fully AI-generated tracks from those where AI simply assisted a human artist, echoing the same disclosure logic showing up everywhere else.

A third response just showed up this week, and it might be the most interesting one yet. Instead of platforms removing content or forcing disclosure at upload, Substack announced it's putting detection directly into readers' hands. Working with the AI-detection company Pangram, the new feature lets anyone scan a post, note, or comment over 100 words and see an estimate of how much was likely written by a person versus generated by AI. Substack even coined a term for the specific harm it's trying to prevent: "Claudefishing," the experience of investing your attention and trust in something you believe came from a person, only to find nobody was actually on the other end. Substack has been careful to frame this as a transparency tool, not a ban. Writers can add a "How I Make This" statement explaining their process, run the scanner on their own drafts before publishing, and dispute scans they think got it wrong. The company says it isn't opposed to AI-assisted writing at all, only to readers being misled about what they're actually getting.

That distinction, between banning AI content and simply making its origin visible, is probably going to matter more than any single platform's detection numbers. It shifts the decision about what counts as acceptable AI use away from the platform and onto the individual reader or community, which is a meaningfully different posture than Spotify or YouTube unilaterally deciding what gets removed or demonetized.

Where is this all heading?

A few patterns are already clear enough to call.

Detection and removal is becoming a permanent cost centre, not a one-time cleanup job. Spotify's 75 million is a floor, not a ceiling. Every platform now needs standing enforcement teams the way banks run fraud detection, because the moment enforcement pauses, the flood comes right back.

Provenance labelling is on track to become the default rather than the exception. With C2PA adoption already spanning thousands of companies, and regulation forcing disclosure across the EU and parts of the US, expect labels like "AI-generated," "AI-assisted," and "human-made" to become as ordinary as nutrition labels on packaged food within the next two or three years.

A two-tier content economy is taking shape. Mass-market, routine, templated content, generic listicles, stock background music, filler video, will keep getting produced at scale by AI and largely ignored by serious audiences. Meanwhile, verified human-made work, original reporting, distinctive creative voices, credentialed expertise, is becoming the premium tier that platforms, advertisers, and audiences actively pay to distinguish and protect. Some analysts are already calling this the "human factor" becoming a scarce resource, rather than something you can just assume by default.

Discoverability, not production, is becoming the real battleground. When anyone can produce infinite content, the constraint shifts from making things to being found. Expect search engines, streaming platforms, and marketplaces to keep tightening ranking signals around originality, verified authorship, and engagement quality, punishing templated volume even when it isn't removed outright.

Smaller platforms and open ecosystems will feel the sharpest version of this problem. Big players like Spotify, YouTube, and Amazon can afford dedicated detection teams. Newer, smaller, or open platforms, including community blogs and niche marketplaces across Africa's growing digital media space, will struggle to keep pace unless they adopt the same provenance tools the big players are building, or lean on shared industry standards like C2PA instead of trying to solve detection alone.

Reader-facing detection is likely to become a standard feature rather than a novelty. Substack putting a scan button directly into readers' hands is probably just the first of many. Expect more publishing and social platforms to add in-app AI-likelihood scores over the next year, shifting some of the trust burden from the platform's enforcement team onto the reader's own judgment, alongside the writer's own disclosure statement. Vocabulary is already forming around this specific harm. Terms like "Claudefishing" and "AI slop" are catching on fast enough that they're likely to stick around the way "clickbait" did a decade ago.

What this means for African creators and publishers

For anyone building a media platform, a music catalogue, or a self-publishing business on this continent, the real lesson from Spotify's cleanup isn't about music at all. It's about what happens to any open platform that doesn't build enforcement and provenance into its foundations early. The flood doesn't announce itself. It shows up 100,000 uploads at a time, and by the time a platform notices, the cleanup bill is enormous. The smarter move, for a new platform or a small publisher alike, is to start building disclosure norms and quality signals into the product now, while the volume is still manageable, instead of waiting for a crisis to force a 75-million-item purge later. Substack's "How I Make This" statement is a good, low-cost example to study: a simple, voluntary disclosure habit built in early, before trust becomes the thing you're scrambling to repair.