The synthetic slop wave on social feeds

New TikTok users opening the app for the first time are greeted by a flood of AI-generated content. According to a report published by video editing platform Kapwing, nearly 60% of the videos shown on the “For You” page to a brand-new account fall into the so-called AI slop category: low-effort, often disjointed or clearly fake clips produced using LLMs and image-generation tools. This figure reshapes our perception of the platform: what was once the realm of grassroots creativity risks turning into an archive of synthetic outputs, with repercussions for authenticity and trust.

The methodology: 10,000 videos under the microscope

Kapwing’s investigation analyzed 10,742 videos across 20 popular categories, supplementing the observation with a specific examination of the first 500 pieces of content served to a new account. The 60% estimate of AI-generated material is the core of the report, and while the company did not release granular details about the detection algorithms used, the sample is large enough to raise alarm for anyone operating in the digital content space. This is not just about TikTok: the dynamic fits into a broader trend where every social platform is exposed to the proliferation of synthetic media.

On-premise detection: why it becomes a competitive factor

For brands, agencies, and organizations managing large volumes of user-generated content, the wave of AI slop forces a rethink of moderation pipelines. Relying exclusively on cloud services to recognize AI-generated content can mean handing over sensitive data, increasing latency, and losing control over models and updates. Those who instead opt for an on-premise deployment of LLM-based classifiers can fine-tune detection strategies without letting data leave their perimeter, ensuring compliance with regulations like GDPR and reducing TCO in high-volume scenarios. Modern frameworks allow inference of specialized models on hardware with sufficient VRAM, potentially applying quantization to balance accuracy and energy consumption. At AI-RADAR we track these trade-offs, offering analysis on the adoption paths of self-hosted stacks.

Beyond TikTok: digital authenticity and the provenance challenge

The slop phenomenon is just the tip of the iceberg. The ease with which mass synthetic clips can be produced today foreshadows a future where verifying content origin will be an essential requirement not only for platforms but also for companies building digital experiences. Cryptographic watermarking standards, provenance systems, and increasingly sophisticated detection engines — often based on LLMs trained to distinguish patterns typical of automatic generation — will become part of security architectures. In this scenario, too, direct control over inference infrastructure allows organizations to quickly adapt models to new manipulation vectors without depending on third parties. It is time to design authentication pipelines that look beyond the feed surface.