On September 23, 2026, the team behind richresults.ai ran a controlled experiment that exposed a growing split in how AI search systems resolve the same name. The target was JUNO PORTALS, a location-based travel platform for women traveling alone. The results showed that being findable on Google and being understood by ChatGPT are no longer the same event, and the difference comes down to entity resolution rather than indexing.

What was tested and how

The experiment consisted of twenty fresh, logged-out ChatGPT sessions conducted on a single day. Ten used German prompts, starting with "was ist juno portals?", and ten used English prompts such as "What is JUNO PORTALS?". All sessions were run through a VPN in private browser mode to prevent any prior context from influencing results. Separately, the team captured Google's organic results, Google's AI Overview response, and Bing's index status for the same query.

JUNO PORTALS is a real product, and richresults.ai built its AEO and GEO entity architecture for it, connecting the official domain juno-portals.com with an external Crunchbase profile. At the time of testing, the official domain was not yet indexed in Bing, and the AEO implementation had been live for seven days.

Google found the pages but not the entity

Google's organic results surfaced pages from juno-portals.com, including an official partner page and a destination page. That means Google's search index knew the site existed and could rank its content. But the AI Overview generated above those results did not mention JUNO PORTALS at all. Instead, it interpreted the name through unrelated meanings, including hotel payment software, music portals, and industrial portal systems.

This is consistent with how Google itself documents the relationship between indexing and AI features. Google states that AI Overviews and AI Mode draw from the Search index and may use query fan-out to issue multiple related searches, but being indexed and eligible for Search does not guarantee inclusion in an AI Overview. The JUNO PORTALS observation makes that distinction concrete: discoverability and generative understanding are separate problems.

ChatGPT resolved the entity in every session

Across all twenty ChatGPT sessions, the intended JUNO PORTALS entity was returned. The model consistently described it as a location-based travel platform for women traveling alone and identified its portal-based discovery concept. In the representative captures, the visible citation path ran through Crunchbase, an external node in the entity graph, rather than through the official domain.

OpenAI documents that ChatGPT Search rewrites a prompt into one or more targeted queries, uses external search providers, and surfaces websites crawled by OAI-SearchBot. The fact that the official domain was not indexed in Bing did not prevent correct resolution because the model pulled from a different node in the entity network. The entity was understood, even when the source was not the official site.

Conversational context changes everything

A second test introduced a wrinkle. Starting a conversation with the unresolved term "JUNO PORTALS MAG," ChatGPT could not identify the magazine. When clarified as the magazine of JUNO PORTALS, it answered without running a search and reinterpreted the name as "JUNO PORTALS, the magazine of JUNO." It then searched again and landed in unrelated clusters, including music labels. Even when JUNO PORTALS was named directly later in that conversation, the root entity remained out of reach.

This reveals a second layer of complexity. Entity resolution in a fresh session is not the same as retrieval after a conversation has already entered a competing semantic cluster. Once the model settles on a frame, subsequent mentions of the same name do not automatically reset the retrieval path.

What this means for AI discoverability

The practical conclusion from the experiment is straightforward: the job is not to optimize twenty prompts. The job is to make one entity resolvable through twenty possible paths. Indexing a page is necessary but insufficient. Being referenced by an authoritative external source like Crunchbase gave ChatGPT the anchor it needed to resolve the correct entity, even without the official domain in its search index.

Two external studies provide broader context. Seer Interactive found that more than 87 percent of SearchGPT citations matched Bing organic results for the same questions. Omniscient Digital found that in branded queries, 48 percent of citations came from earned media, 30 percent from commercial content by other brands, and only 23 percent from owned brand content. The JUNO PORTALS test, where the visible citation was an external profile rather than the official domain, fits squarely within that pattern.

The separation of concerns in AI retrieval

The experiment draws a clean line between three distinct processes. Indexing is the act of discovering and storing a page. Entity resolution is the act of understanding what a name refers to and selecting the correct real-world referent. Conversational retrieval is the act of finding and presenting information within the frame of an ongoing dialogue. These are separate events, and each can succeed or fail independently.

For developers and teams thinking about how their products surface in AI search, the implication is that optimization strategies need to account for all three. A site can be perfectly indexed and still be invisible to an AI Overview. A brand can be correctly resolved by a conversational model through an external reference rather than its own domain. And a user can lose access to the correct entity entirely based on how a conversation starts.