Thinktanks want AI news labels for transparency. But the real danger lies in AI’s role in shaping perception and trust before users even question accuracy.
AI tools and businesses are actively shaping how users perceive information, and that’s the real threat.
Generative AI is still sloppy at creating content that’s comparable to human creators. But it’s not as if users haven’t tried their best to rely on it anyway. The writing and designs are too discernible, and the quality too repetitive and shallow to truly match professional creatives.
However, that’s only the visible end of the problem.
AI today is not just a content generator. It is a search engine, a chatbot, and increasingly, a first point of reference. It offers answers promptly, confidently, and without friction. Technically, it’s an information exchange. But information exchange without provenance changes how authority is formed.
What happens when actors leverage that maliciously? Or subtly? Or simply at scale?
It’s something experts at The Institute for Public Policy Research (IPPR) are concerned about- first, what if AI firms steal information without compensation to publications, they’re taking data from? And second, what if they twist the data?
Both are dangerous indeed.
Even before AI flooded the internet, social platforms positioned themselves as sources of current affairs. X still does. But AI removes even more friction. You don’t need to follow anyone. You don’t need to subscribe. You don’t need to compare sources. Users get what they ask for, immediately. That’s where the problem begins.
AI models are trained on an average drawn from a limited chunk of accessible data. Meanwhile, large portions of journalism and research remain locked behind paywalls, licenses, or structural exclusion. It’s where the problem occurs-
Models don’t just hallucinate. They normalize partial truths. They sound complete even when they aren’t.
That’s precisely why IPPR has proposed a way out.
It argues that AI-generated news should carry a “nutrition label”, detailing sources, datasets, and the types of material informing the output. That label should include peer-reviewed research and credible professional news organisations.
What the proposal gets right is transparency. What it does not fully confront is power. When AI mediates perception at scale, disclosure alone cannot restore editorial judgment. It can only expose its absence.
