The GEO Playbook: Engineering Narrative at Scale

The Hanover Institute for Public Policy is moving the goalposts on how we think about information. They’re using a commercial platform called Res to systematically shape how AI chatbots like ChatGPT and Gemini present information. This isn't just SEO; it’s "generative engine optimization" (GEO). The goal? Influence the actual output of a Large Language Model (LLM) by hitting both live search results and the underlying training data repositories, like Common Crawl.

Engineering Content for LLM Credibility

This operation centers on Piro Inc and their "AI Story Optimization." They aren't just writing for humans; they’re authoring content specifically engineered to align with how LLMs evaluate credibility. To do this, the Hanover Institute didn't just post a few articles—they published over 560,000 words in just nine days. We’re talking 124 reports, with some bursts hitting nearly 354,000 words in only 48 hours. This is a high-velocity deployment designed to "prime" the chatbot with specific narratives before the user even asks a question.

The Commercial Scale of Narrative Shaping

This isn't some grassroots effort; it’s a $1 million investment across just two work orders. By saturating the web with high-volume, high-frequency content that mimics the patterns LLMs use to establish factuality, they are creating a "default" answer. When a user asks a chatbot about a specific topic, the model regurgitates the desired framing because that's what the data says. The danger is that the user may never see the original source—they just get the polished, engineered result.

The Verification Gap in Production

From a builder’s perspective, the real story here is the commoditization of "credibility" as a measurable, gameable metric. When an organization can spend $1 million to engineer content specifically for how an LLM evaluates it, the trust model of AI shifts from "source-based" to "volume-and-pattern-based." The risk is massive: when a chatbot regurgitates these narratives without citing the source, it becomes nearly impossible for a user to fact-check the output. For those of us building on top of these models, this is a huge hurdle. As GEO becomes a standard tool for narrative shaping, the gap between "clean" training data and "messy" production influence is widening. The challenge isn't just filtering bad data; it's identifying when the "good" data has been systematically engineered to prioritize a specific narrative over an objective one.