CASE STUDY: GEOforge

June 26, 2026 | Insights

GEOforge addresses the shift from traditional SEO to AI-driven discovery by helping government websites communicate clearly with large language models. It enables sites to define how AI understands their content, ensuring accuracy and intent rather than leaving interpretation to inference.

Your .gov Has a Story for Humans. Does It Have One for AI? Try GEOforge.

By Calvin Kwon, IT Consulting Intern

Suppose you run a federal website. Its mission, like most US government websites, is to deliver authoritative information to citizens (e.g. nutrition guidance, agriculture data, anything). This mission only works if the information reaches people.

For years, websites have been built with search priority in mind. Traditional SEO (Search Engine Optimization) was about visibility, ranking, and discovery by human search. It helped pages gain internet exposure, retention, and helped certain internet content to surface at the right times.

Now suppose an LLM (Large Language Model, the AI behind tools like ChatGPT and Claude) is trying to understand your site. Unlike traditional search crawlers, the way AI processes and retains data is fundamentally different from how search indexers do. For example, a traditional crawler looks for keywords and links, whereas an LLM builds a representation, or an approximation of what your site is, what it does, and how it relates to everything else it ‘knows’.

The web is changing, and most either know or use AI systems as an integral method of search, summarization, and information consumption. This poses a new question for website owners: what should it look like for my site to communicate with AI systems as well? GEOforge was built to answer exactly this question.

Initially created for the US Department of Agriculture, GEOforge is an open source toolkit built around GEO (Generative Engine Optimization), designed for developers who build and maintain .gov sites. Unlike traditional internet crawlers, which scan for keywords and links, GEOforge uses multiple AI crawlers and fetchers that read a site the way an LLM would to build a semantic image of a site's content. That image is then used to generate machine-readable files that define how AI systems should interpret your site: robots.txt, sitemap.xml, ads.txt, well-known/ai.txt, and a geoforge.json summary. These files help AI visitors contextualize the website’s purpose, structure, and policy. GEOforge also supports both browser and CLI input, with profile presets such as strict-privacy, balanced, and open-discovery so users can choose how openly they would like their site to be interpreted by AI models.

There is, in fact, a meaningful difference between a model that has visited your site and one that has understood it. An LLM will build a representation of your site whether you shaped it or not (your intent is either expressed or inferred). In a sense, GEOforge is how site owners stay present in that process. It doesn't touch your content; it adds a second layer written in the language AI systems read, so the model's understanding reflects what you actually meant.

This distinction matters more now than it did two years ago, and it will matter more in two years than it does today. The share of web traffic that arrives through AI mediated discovery (agentic browsing, for example) is growing rapidly, and the web is only just beginning to feel this difference.

If SEO was about being found, GEO is about being understood. Thus, sites that have defined themselves for these systems will be understood. Sites that haven’t will be inferred, and inference holds almost no accountability over your intent. GEOforge is open source and can be access online at https://geoforge.dev or via CLI with: npm install -g geoforge-cli

With GEOforge, the surface area of your .gov website (how AI reads, cites, and decides what to do with it) is yours to define. The question now is whether you define it, or let a large language model decide for you.