The next major battle over online information may not be fought for the attention of human readers.
It may be fought for the attention of artificial intelligence.
Researchers have warned that propaganda networks are publishing misleading material in ways designed to influence what AI chatbots retrieve, cite and repeat.
The strategy has been described as a malicious form of generative engine optimisation: creating enough apparently relevant online material that AI systems begin treating manipulated narratives as credible sources.
What Researchers Found
A report highlighted by the British think tank Demos examined whether leading chatbots surfaced narratives connected to Russian propaganda networks.
Researchers focused on content linked to the Foundation to Battle Injustice, an organisation associated with Russia and subject to Western sanctions.
The investigation found that some chatbot responses cited or repeated false pro-Russian narratives, including fabricated claims connected to Ukraine.
The findings suggest that actors are no longer producing propaganda only to persuade people directly.
They may also be attempting to influence the information environment from which AI systems construct answers.
How Generative Engine Optimisation Can Be Abused
Traditional search-engine optimisation attempts to make a webpage more visible in Google or other search results.
Generative engine optimisation attempts to make content more likely to be retrieved, cited or summarised by an AI assistant.
Legitimate businesses can use this process to publish accurate, useful and clearly structured information.
The problem begins when the same methods are used to flood the web with coordinated falsehoods.
If dozens or hundreds of connected sites repeat the same misleading claim, an AI system may interpret repetition as corroboration.
But repetition is not evidence.
Ten websites controlled by one propaganda network do not represent ten independent sources.
Why the Story Went Viral
People increasingly ask chatbots to explain breaking news, political events, wars and public controversies.
Many users assume that a fluent answer accompanied by citations has already been checked.
This story disrupts that trust.
It suggests that a citation can lead to a real webpage while the webpage itself remains deceptive, coordinated or politically manipulated.
The AI may not fabricate the source.
It may retrieve a source that was created specifically to mislead it.
What Is Confirmed and What Is Not?
Demos researchers and subsequent news reports documented examples of mainstream chatbots surfacing information linked to sanctioned or propagandistic sources.
This does not mean every answer from the named chatbots repeats propaganda.
Performance varied. Some systems rejected false narratives, while others discussed or validated parts of them.
It also does not mean generative engine optimisation is inherently dishonest.
The technique becomes dangerous when used to manufacture false authority, conceal coordinated ownership or manipulate politically sensitive answers.
Why AI Is Vulnerable
Large language models do not possess human trust in the psychological sense.
They identify patterns, retrieve information, weigh signals and construct responses.
When the online environment contains organised contamination, the system must determine which sources deserve credibility.
That becomes difficult when propaganda operations imitate the appearance of journalism, research organisations or human-rights groups.
The problem is therefore not only inside the model.
It exists across the entire information supply chain:
Who created the source?
Who funded it?
How many apparently separate sites share the same ownership?
Is the claim supported by primary evidence?
Are independent publications reporting the same facts?
Was the content designed primarily for people or for search and AI systems?
The Bigger AI Lesson
The internet is becoming the memory system of artificial intelligence.
Whoever can pollute that memory may influence future answers at enormous scale.
Social-media propaganda typically targets people one post at a time.
AI-oriented propaganda could be repeated by assistants across millions of private conversations, often with the calm language of an objective explanation.
That makes source integrity one of the defining AI-governance challenges of the coming years.
Mary Chuks’ Perspective
Artificial intelligence does not remove the need for media literacy.
It increases it.
People must learn to question not only an answer, but the information ecosystem behind the answer.
The age of AI requires a new habit: source triangulation.
Users should ask whether the cited sources are independent, whether the original evidence exists and whether a claim is being repeated because it is true or because a network deliberately made it abundant.
AI systems should also explain the quality of their sources, not merely display links.
A source label without a credibility assessment can create false confidence.
Practical Takeaways
Ask chatbots to distinguish primary sources from commentary.
Check whether several cited sites share the same owner or narrative.
Be cautious when extraordinary claims appear only in politically aligned outlets.
Use established fact-checking organisations and primary documents for sensitive claims.
Do not assume a polished AI answer represents a verified consensus.
Conclusion
The information war is evolving.
Propaganda no longer needs every human being to read the original article.
It only needs enough machines to retrieve it.
The future of trustworthy AI will therefore depend not only on smarter models, but on protecting the quality of the information those models consume.
Original Source and Further Reading
Research organisation: Demos research discussed by Carl Miller and Hannah Perry on the use of generative engine optimisation to influence AI answers, published in late July 2026.
European reporting: Euronews, “AI Chatbots Citing Russian Propaganda Sourced From EU-Sanctioned Outlet”, published 27 July 2026.
Additional context: Research on chatbot retrieval has found that retrieval failures account for a substantial portion of factual errors and that systems can become vulnerable when questions contain false premises.
Strongest lead story: the Hugging Face agent intrusion.
Most emotional story: the ChatGPT family reunion.
Most shareable light story: Jimothy.
Best creator-rights story: Instagram’s AI-remix reversal.
Best policy and psychology story: propaganda designed to influence chatbot answers.