Narrative Laundering

Narrative laundering works much like money laundering: a claim too dubious to be believed outright is funneled through several intermediate stations until it appears clean and credible. It often begins with a deliberate falsehood or a heavily distorted portrayal. This is first posted on niche blogs or on social networks. Then a larger media outlet quotes it — without thoroughly checking the original source. Eventually, a reputable news portal reports on the “story” without even knowing that it originally was an unsubstantiated claim. The origin is obscured, and the story appears legitimate.

Why AI makes this process more dangerous

In the past, narrative laundering required time and human labor. Someone had to write articles, operate fake accounts, and build networks. Today, a single person using a language model — that is, an AI system that generates text — can produce hundreds of credible-sounding articles, comments, or press releases within minutes. This dramatically lowers the barrier to entry for targeted disinformation campaigns.

The second problem is the credibility of the texts themselves. AI-generated content often sounds factual, is well structured, and free of obvious errors — which is precisely what makes it hard to recognize as misinformation. When a fabricated study is phrased just like a genuine one, even experienced readers struggle to tell the difference. On top of that, many AI models were trained on text from the internet — including falsehoods that were already circulating. They can repeat these unnoticed and thereby spread them further.

The journey of a laundered story

A typical sequence looks like this: an interest group has AI phrase a falsehood as a seemingly expert analysis. The text ends up on an obscure blog with a reputable-sounding name. An automated aggregation tool — that is, a program that compiles news from many sources — picks up the article. A journalist under time pressure cites the aggregation site. The citation spreads further without anyone still checking the primary source.

What matters is this: at every stage, the likelihood that someone traces it back to the original source decreases. The term “laundering” fits well for this reason — in the end, it’s no longer possible to reconstruct where the story came from. Researchers also speak of “source evaporation”: the more often something is cited, the more its original origin fades from view.

A real-world pattern is so-called pink-slime websites — sites that look like local news portals but exclusively publish automatically generated content. In the US, hundreds of such sites were identified before elections, spreading politically slanted misinformation dressed up as respectable-looking news. With AI, such networks can be built even faster and more cheaply.

Narrative laundering in news and product debates

The term increasingly appears in debates about AI regulation. Media researchers and authorities such as the European AI Safety Office discuss how platforms could be required to label AI-generated content — precisely because narrative laundering is otherwise hardly possible to stop. The major technology companies are also being judged on whether their models can be especially easily exploited for such campaigns.

In everyday life, one encounters the phenomenon anywhere rumors suddenly seem “official”: a claim about a medication that suddenly appears on reputable health portals, or a political statement that is considered “widely reported” even though it was never substantiated. The difficulty is this: it is not enough to distrust a single source. One has to ask where the chain of citations originally began — and that requires more effort than most readers are willing to invest.

Media literacy programs in schools and universities are now explicitly addressing narrative laundering. They teach how to trace source citations backward — that is, not just to look at who is saying something, but where they originally got it from. This step is simple to describe, but unusually laborious in day-to-day news consumption.