How Big Ideas Travel When Machines Do the Talking
At DEY, there is a simple threshold for the work we take on. The ideas have to matter. They must engage with problems larger than any single company or product, problems that shape how people live, decide, and act in the world.
That conviction comes from a basic observation. Influence does not travel through brands so much as through ideas. Ideas about how technology should be governed. Ideas about what constitutes risk, safety, or responsibility. Ideas about what is possible, and what should not be.
What has changed is not the importance of ideas, but the way they move.
For most of the internet era, influence followed distribution. Ideas traveled through search results, social feeds, conferences, and media coverage, carried forward by human readers. Today, they increasingly move through systems that summarize, recombine, and surface knowledge on demand. Large language models now act as intermediaries between ideas and the people encountering them.
Ask a chatbot a question about artificial intelligence risk, climate policy, or public health, and the answer arrives already synthesized. The user may never see the underlying sources. But the sources are there all the same, shaping what is included, emphasized, and left out.
The practical question for anyone working on consequential ideas is no longer simply how to get attention. It is how ideas become embedded in the systems that now decide what gets explained, repeated, and remembered.
Why Journalism Still Teaches AI What Matters
For all the attention paid to artificial intelligence, large language models remain surprisingly conservative about what they trust. They do not elevate ideas because they are novel, provocative, or widely shared. They repeat ideas because those ideas appear consistently and clearly in places that have already done the work of verification.
In practice, that means journalism.
Studies of training data and retrieval behavior show that LLMs rely disproportionately on professionally edited text, including major news outlets, academic publications, policy research, and other sources with institutional authority. Analyses of commonly cited material in AI-generated answers consistently show that content originating from established media and research organizations appears far more frequently than content from brand blogs, press releases, or influencer platforms.
This does not mean podcasts, newsletters, or social media are irrelevant. It means they tend to function as reinforcement rather than foundation. Journalism remains the layer that signals to machines that an idea is serious enough to repeat.
The implication for anyone working on consequential ideas is straightforward but often misunderstood. Getting an idea into a model does not begin with optimization. It begins with explanation.
1. Anchor Big Ideas in Sources LLMs Already Trust
Large language models do not decide what to repeat based on popularity or novelty. They repeat ideas that appear consistently in high-authority, professionally edited sources. These are the materials that have passed through multiple layers of human judgment, editing, and verification, making them more reliable building blocks for machine reasoning.
A single well-reported article in a respected outlet often outperforms dozens of lower-quality mentions when it comes to AI visibility. That article gets indexed, referenced, and republished across platforms. Over time, it becomes part of the informational substrate that models draw on to answer user queries. What once primarily served search engines now functions as discovery infrastructure for systems that summarize reality for millions of users each day.
In recent years, some of the world’s most respected publishers have formalized this role through partnerships with AI developers. OpenAI’s collaboration with Axel Springer, for example, allows outlets such as POLITICO and Business Insider to appear in ChatGPT responses with attribution and links to full reported articles, improving accuracy while preserving the value of editorial work. The Associated Press has similarly signed agreements with OpenAI to share select news content and technology, extending the reach of factual reporting through AI systems. Major publishers including The Wall Street Journal, Financial Times, Le Monde, and Prisa Media have also entered licensing arrangements that allow their journalism to be used for training and real-time responses.
The implication is not subtle, but it is often resisted. For organizations serious about the longevity of their ideas, traditional media cannot be treated as optional or secondary simply because it is difficult, slow, or uncertain. It has to sit closer to the center of communications strategy precisely because it operates under constraints. In a machine-mediated world, credibility accumulates where judgment, verification, and editorial standards still apply. Prioritizing earned media is not about chasing headlines. It is about committing to the long work of placing ideas in the sources that both humans and machines continue to trust.
2. Reinforce Journalism With Formats Machines Can Cross-Reference
Once an idea appears in reporting, other formats help it endure.
A traditional media placement should not be treated as the end of the process, but as the anchor. From there, the work becomes one of reinforcement. Conference panels, keynote talks, podcasts, newsletters, and public lectures often generate transcripts and secondary coverage that large language models ingest and retrieve. When these formats echo the same core idea introduced through journalism, they signal that the idea is not a one-off, but part of an ongoing conversation.
In practice, this means following a strong piece of earned media with appearances in adjacent formats whenever possible. A podcast conversation that expands on a reported article adds texture. A newsletter essay that reflects on the same idea extends its lifespan. A conference talk that repeats the framing introduced in journalism helps stabilize it across contexts. These formats do not replace reporting. They corroborate it.
Owned channels matter here as well. Updating a website’s “About” page, publishing a clear explainer, or maintaining a visible archive of talks and writing gives both humans and machines a consistent reference point. When earned media and owned content reinforce one another, ideas become easier to recognize and retrieve.
The implication is practical. Communications teams can no longer treat AI as a downstream concern. Spokespeople need to be prepared to explain ideas with clarity and consistency across settings. When an idea is framed one way in an interview, another way on a conference stage, and a third way on a website, both humans and machines struggle to track it. When it is named clearly, explained similarly, and anchored in evidence across channels, it becomes easier to summarize and repeat.
3. Put Humans at the Center of Ideas Before Machines Repeat Them
Before an idea ever reaches an AI system, it passes through a human filter. Journalists do not interview abstractions. They interview people. They look for someone who can explain an idea clearly, defend it under pressure, and situate it in the real world. That process sharpens ideas in ways machines cannot.
This human layer is essential. A credible spokesperson does more than transmit information. They humanize an idea through examples, tone, judgment, and context. They make it intelligible, memorable, and grounded. Those qualities shape not only how readers understand an idea, but how it is later summarized and repeated by AI systems.
Large language models do not handle ambiguity well. Ideas that surface reliably tend to be those articulated clearly and consistently by people who understand them well enough to explain them without drift. This is why the role of a spokesperson has become more consequential, not less, in the AI era.
In fields such as AI risk, climate science, and public health, this pattern is already visible. Organizations whose leaders explain the same core ideas across interviews, policy discussions, and long-form reporting often see those ideas reflected back in how LLMs summarize the field. The machines are not discovering these ideas independently. They are inheriting them.
The implication is practical. Media training now carries a different weight. Preparing spokespeople is no longer only about message discipline or on-camera performance. It requires an understanding of how ideas will be interpreted, compressed, and repeated by systems that privilege clarity, consistency, and evidence. Training humans to speak well has quietly become one of the most effective ways to ensure ideas survive the journey from conversation to machine summary.
At a moment when machines increasingly mediate how knowledge is encountered, the temptation is to focus on the machinery itself. But influence has always been shaped upstream. Ideas endure when they are clear enough to explain, strong enough to withstand scrutiny, and credible enough to be repeated by others.
Large language models do not create trust. They inherit it. The responsibility, then, remains where it always has: with the humans willing to articulate ideas carefully, place them in environments built for scrutiny, and stand behind them as they travel farther than ever before.