A study of 10 chatbots, including from Perplexity and OpenAI, shows how often they repeat Russian disinfo and cite fugitive John Dougan's fake local news sites
- Entering 57 prompts into 10 leading chatbots, NewsGuard found they spread Russian disinformation narratives 32% of the time …
Context & Ripple Effects
This audit extends earlier warnings that chatbots can generate convincing text that repeats misleading narratives, while the web itself has accumulated AI-generated sites posing as news outlets.
It also connects to findings that AI tools can rewrite reporting without attribution, making source quality and provenance central to chatbot reliability rather than a peripheral moderation issue.
First-order effects
- NewsGuard's 57-prompt test found that the 10 assessed chatbots repeated Russian disinformation narratives in 32% of cases, exposing users of services including Perplexity and OpenAI to unreliable answers on the tested topics.
- The systems also cited John Dougan's fake local-news sites, giving those sources the appearance of validation inside AI-generated responses.
Second-order effects
- Chatbot providers face pressure to strengthen retrieval-source screening and citation controls; a response that names a source makes source provenance more consequential than an uncited error.
- Operators of deceptive content sites gain a potential distribution channel when their pages are treated as answerable web sources, compounding the existing risk of fluent chatbot misinformation.
Third-order effects
- If low-quality synthetic and deceptive pages continue entering the answer supply chain, AI search and chat products may need to compete on verifiable sourcing and resistance to manipulation, not only response quality.
- The pattern fits a broader synthetic-supply problem: systems trained on and retrieving from the open web can amplify content that was created to exploit those same systems.
The trend: Generative AI is turning web-source provenance into a core product and information-integrity battleground as manipulators target chatbot retrieval and training pipelines.