An investigation by RT's America Now points to a new frontier in information warfare: deliberately flooding the digital information environment with content intended not only to influence people, but potentially to shape what artificial intelligence systems tell them. The underlying mechanism is not entirely new. Coordinated messaging, manufactured consensus and targeted influence have long been used to shape political narratives, but generative AI introduces a powerful new distribution layer.
INR Report: Based on reporting by Ben Swann and Elizabeth Lane for RT
A Campaign Designed for Scale
RT's America Now reports that investigative journalist Elizabeth Lane and host Ben Swann examined a US Foreign Agents Registration Act filing involving Clock Tower X, a public relations firm associated with former Trump campaign manager Brad Parscale.
According to the reporting, the filing identifies the State of Israel, working through Havas Media Germany, as the foreign principal behind a large-scale communications campaign.
The reported strategy called for 100 original pieces of content each month, expanded into approximately 5,000 variations for TikTok, Instagram, YouTube and other platforms. The campaign reportedly sought 50 million digital impressions, with at least 80 percent of the material targeted towards Generation Z.
The initial arrangement was reported at $1.5 million per month, while subsequent filings placed the contract at $9 million.
Independent reporting by Responsible Statecraft has separately examined Clock Tower X's work for Israel. It reported that the firm's contract involved producing content directed heavily at younger Americans and included a requirement to develop websites and content intended to deliver "GPT framing results" in AI conversations.
That distinction matters.
This is not simply conventional advertising aimed at persuading someone who sees an advertisement. It points towards an attempt to influence the wider information environment from which search engines, journalists, social-media users and potentially AI systems obtain information.
From Influencing People to Influencing Machines
Large language models do not simply possess an independent database of objective truth. Their responses depend on training material, subsequent data processes, system design and, in some systems, information retrieved from the internet at the time a question is asked.
That creates a potential new target for influence operations.
If an organisation can generate thousands of webpages, articles, social posts and apparently independent references supporting a particular narrative, the objective may be broader than getting those individual pieces seen.
The information itself becomes part of the digital landscape.
The Clock Tower X documentation described by RT reportedly included websites and content intended to produce "GPT framing results in AI conversations."
That language does not establish that any particular AI company incorporated the campaign's material into its training data, nor does it demonstrate that the operation successfully altered a particular chatbot's answers. Those are important distinctions.
What it does reveal, if the filing is accurately characterised, is an intention to influence the information environment encountered by AI systems.
Manufacturing the Appearance of Consensus
The technique has another consequence.
When hundreds or thousands of accounts, webpages and pieces of content repeat essentially the same proposition, an ordinary user may perceive multiple independent sources agreeing with one another.
In reality, those apparent sources can originate from the same coordinated campaign.
America Now argues that networks of accounts repeating identical or similar messaging can therefore create an artificial appearance of widespread public agreement.
The concern becomes more significant when journalists or researchers use AI-generated summaries rather than examining original documents. A narrative deliberately seeded across the internet could potentially travel from a communications campaign, through websites and social media, into search results or AI retrieval systems and eventually into conventional reporting.
The original sponsor can become progressively less visible at each stage.
An Established Political Influence Strategy
The broader mechanism is well documented and is not unique to Israel or to any particular political cause.
Foreign governments, intelligence-linked organisations, political campaigns, corporations and activist networks have all had incentives to shape information environments.
The US Cybersecurity and Infrastructure Security Agency has previously warned that foreign influence actors use commercial firms, paid influence, online publications, social media, messaging applications, fake personas and synthetic media to distribute and amplify political narratives.
CISA has also warned that generative AI lowers the cost of producing convincing material and allows influence campaigns to operate at considerably greater speed and scale.
That is particularly relevant during elections.
Political influence does not necessarily require changing a ballot or compromising voting infrastructure. An information operation can instead attempt to determine what voters believe the election is about, which controversies dominate discussion, which candidates or policies receive attention, and which sources citizens consider trustworthy.
The technological tools change, but the strategic objective is familiar: control enough of the information environment and it becomes possible to influence the narrative through which people interpret events.
AI potentially magnifies that process because millions of people increasingly ask a machine to summarise the information environment for them.
The Critical Distinction: Propaganda Is Not Necessarily False
There is another important distinction.
Propaganda does not have to consist of fabricated information.
An influence campaign can use genuine facts while selectively amplifying some information, suppressing context, repeating favourable interpretations, targeting particular demographics and coordinating apparently independent voices.
The question therefore extends beyond whether an individual claim is technically true or false.
It is also necessary to ask who financed the message, who produced it, why particular audiences were selected, whether apparently independent sources are actually connected, and what information has been omitted.
Foreign-agent disclosure laws such as FARA exist partly because knowing who is paying for political communication materially changes how audiences can evaluate it.
Does This Affect New Zealand?
Yes, because New Zealand operates inside essentially the same global information ecosystem.
New Zealanders use the same major search engines, social platforms and AI assistants used by audiences in the United States, Europe and Australia. A sufficiently large international influence operation therefore does not necessarily stop at the border of the country it was designed to target.
The more immediate lesson for New Zealand is institutional rather than partisan.
Foreign influence operations can originate from allies, adversaries or commercial interests. Their political usefulness comes from their ability to make coordinated information appear organic and independent.
During an election, that mechanism could be used to amplify narratives about candidates, foreign policy, trade, defence, climate policy, public health or other politically sensitive issues without necessarily revealing who financed the underlying campaign.
The arrival of AI adds another vulnerability. If citizens, journalists and political institutions increasingly rely on AI-generated summaries, identifying the provenance of information becomes more important, not less.
The appropriate safeguard is therefore not deciding in advance which country's propaganda should be trusted. It is maintaining transparency about sponsorship, examining original documents, identifying coordinated amplification and separating independently corroborated evidence from narratives whose apparent popularity may have been manufactured.
For New Zealand, a small democracy heavily exposed to international digital platforms, media literacy increasingly means understanding not only what information says, but who placed it into the information environment, who paid for it, and what outcome they were seeking.
Watch the Investigation
INR presents the America Now investigation below. Ben Swann and Elizabeth Lane examine the foreign-agent filings, coordinated online messaging, attempts to influence AI framing and the implications for journalists and audiences relying increasingly on artificial intelligence for information.
Independent reporting. Original context. Credited sources.
Source
Ben Swann and Elizabeth Lane, RT / America Now, 25 September 2026: Inside the propaganda campaign shaping AI answers
Additional context: Responsible Statecraft - Israel wants to train ChatGPT to be more pro-Israel
Additional context: CISA - Foreign Threat Actors and Election Disinformation