Something interesting happened during a small AI visibility experiment this weekend.

NetContentSEO was testing whether AI systems recognize small independent research projects when they are asked a neutral discovery question. The prompt didn't mention the project, its website or its founder. It simply asked Meta AI for examples of small independent AI search research labs publishing their own experiments rather than reporting industry news.

In the first test, Net Content SEO wasn't there.

Meta AI instead returned several established projects including OtterlyAI, Discovered Labs, DEJAN AI and Peec AI. Looking at what those projects had in common produced an interesting pattern: original research, documented methodology, reproducible experiments and, in several cases, public GitHub repositories.

NetContentSEO then made one change.

It created a public GitHub research organization and repository describing its AI visibility methodology, research areas and first reproducible experiment. The original discovery prompt wasn't modified.

Minutes later, a completely new Meta AI conversation was opened and the same question was asked again.

This time, the first result was NetContentSEO AI Labs.

Meta AI described it as a small independent lab focused on AI Search behavior and specifically highlighted characteristics such as fixed experimental protocols, published raw outputs and GitHub/open data. Those characteristics closely matched the information that had just been made more explicit publicly.

An interesting result, but not proof

The tempting conclusion would be that publishing the GitHub repository caused Meta AI to recognize the entity.

There isn't enough evidence to say that.

Generative systems are non-deterministic. Retrieval can vary between sessions, sources can change, and two observations cannot establish causality. The second response could simply have resulted from retrieval variance.

But the observation is still worth documenting.

An entity was absent from a neutral discovery query. Additional structured and publicly accessible evidence about that entity was published. Minutes later, the same entity appeared as the first example for the same query in a fresh conversation.

The interesting question for AI visibility is therefore not whether GitHub is a GEO ranking factor. There is no evidence here to support that claim.

The better question is whether publishing consistent, machine-accessible evidence across independent locations can make an entity easier for retrieval systems to reconstruct and classify.

That is something that can actually be tested.

NetContentSEO says it will now preserve the original query and repeat the experiment over time and across Meta AI, Gemini, Perplexity, Grok and other systems. That should make the next results considerably more useful than the initial before-and-after observation.

For projects trying to understand how LLMs discover entities, this is exactly the kind of small experiment worth watching: change one thing, document what happened and resist claiming more than the evidence shows.

Original experiment and screenshots:
NetContentSEO — We Created a Public Research Repository. Minutes Later, Meta AI Found Us.