# The trust gap nobody is closing

> Everyone optimizes the site they own. The data says agents weight independent corroboration. That is the unclaimed move in agentic GTM.

Everyone racing to make their site "AI-ready" is perfecting the thing that matters least. The data says so, and almost nobody is acting on it.

Start with the research. [Princeton's GEO study](https://arxiv.org/abs/2311.09735) and the [GEO-16 citation work](https://arxiv.org/abs/2509.10762) landed on the same finding: generative engines weight independent earned media and routinely drop vendor-owned pages. Your perfect llms.txt can still go uncited.

The citation map shows why. Within ChatGPT's top-10 cited sources, [Wikipedia alone is roughly 48%](https://www.tryprofound.com/blog/ai-platform-citation-patterns). Agents do not roam the open web and find you. They pull from a short list of sources they already trust. As Carolyn Shelby puts it, [AI does not discover new brands, it selects from known entities](https://searchengineland.com/ai-visibility-starts-before-search-ends-with-citations-476308).

That splits agentic GTM into two jobs, and they are not the same.

- **Stock your own shelf.** llms.txt, markdown mirrors, structured docs, clean JSON-LD. Make yourself extractable.
- **Earn corroboration.** Get mentioned by the sources the model already trusts.

Everyone is pouring effort into the first. Almost no one is working the second.

**My read:** the shelf is necessary and nowhere near sufficient. If the trusted sources never mention you, "extractable" just means an agent can read a page it will never be sent to.

I should be honest about the limit of this very page. A personal site is less vendor than a corporate domain, but it is still self-published. It upgrades my named-operator surface. It is not third-party corroboration, and it does not replace a citation from Wikipedia, the press, or a peer with no stake in me.

So the unclaimed move in agentic GTM is not better on-page markup. It is earned third-party presence, run and measured like a real channel. That is the gap I am watching.

---

**Sources**

- Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), KDD 2024. The empirical bedrock.
- [Bringing the GEO-16 Framework to B2B SaaS](https://arxiv.org/abs/2509.10762): AI answer-engine citation behavior.
- Profound, [AI Platform Citation Patterns](https://www.tryprofound.com/blog/ai-platform-citation-patterns): Wikipedia is 47.9% of ChatGPT's top-10 cited sources.
- Carolyn Shelby, [Why AI visibility starts before search and ends with citations](https://searchengineland.com/ai-visibility-starts-before-search-ends-with-citations-476308), Search Engine Land.