Technology
Platforms push back against AI slop
By Staff Writer | 11 August 2026
Online platforms are adding reporting controls, detection tools and distribution limits as public hostility to low-grade artificial intelligence material begins to change product decisions.
LinkedIn has added a reporting option for material that appears to be AI slop. Snapchat has barred fully generated videos from its discovery feed, while Substack has introduced detection measures. The changes point to a shift from asking whether generated material can be made to asking where it should be allowed to travel.
Other experiments have met faster resistance. Meta withdrew an Instagram deepfake feature after three days of criticism, and Google reversed a generative satellite-image editing tool. In each case, the product moved from novelty to reputational problem almost at once.
The commercial risk is trust: when users cannot tell whether an image, voice or account is authentic, the platform itself becomes less useful.
Public opinion is turning colder
A recent US survey found that 39 per cent of adults believed artificial intelligence caused more harm than good, up from 31 per cent in 2025. Only 9 per cent said it did more good than harm. That change arrived while public familiarity with the technology continued to rise.
The AI revolution has happened, and everybody hates it.
Meredith Broussard, professor of data journalism at New York University
The line is deliberately blunt, but the product reversals give it weight. Platforms that once treated generated material as a source of cheap volume are starting to recognise that volume can drive users away.
The difficulty lies in enforcement. A simple label may help when a creator declares the use of a model. It does little when an account hides that use or combines generated parts with real footage. Detection systems also make mistakes, particularly after files are compressed or edited.
Distribution becomes the pressure point
Platforms do not need to ban every generated image to change the incentives. Removing material from recommendation feeds, cutting payments and making it easier to report can reduce the audience available to mass producers.
That approach places the hardest decisions inside ranking systems, where outsiders cannot easily see them. It also raises a fairness problem for legitimate creators whose work is wrongly classified.
The next contest will be less about whether platforms can detect machine-made material and more about whether users trust the rules used to contain it.
For anyone commissioning content, the practical lesson is that disclosure is becoming a condition of distribution rather than a courtesy. Where a platform ranks undeclared generated material down, the cost of a hidden model lands on the buyer as lost reach, not on the producer as a penalty.