How do AI systems choose which sources to cite?

Why this is a working model, not a secret score

No reputable firm has the source-selection code for ChatGPT, Claude, Gemini, Grok, or Perplexity. A guaranteed “AI rank” is a guess sold as a setting. What is observable is narrower. These systems retrieve text from the web or from an index, break it into passages, and prefer passages that answer with little extra inference. Named North designs for those constraints and does not pretend to design the model.

1. Retrieval

A crawler or a search index has to fetch the page. robots.txt, a noindex tag, a login wall, or an answer that appears only after JavaScript runs can all remove the passage from the candidate set. The HTML the server sends should already contain the answer. This site is built that way. The policy is on For machines.

2. Chunking

Pages are split into passages. A passage that states the answer in the first sentence can be lifted on its own. A passage that warms up for three paragraphs cannot. Named North puts the answer first, and puts one question on one URL, so the chunk and the page agree. That unit is an extractable passage.

3. Entity resolution

Named North is a Montana firm that makes companies the named source AI systems cite. That is an entity sentence. “We help brands win the future of search” is not. Conflicting bios, and schema that disagrees with the visible text, weaken the match. The name for this consistency is entity clarity.

4. Support

Specific facts are safer to repeat than compliments. A date, a place, a definition, and a limit (“does not guarantee a model’s wording”) give the system something checkable. Unsourced claims of being number one, and testimonials with no name a person could verify, are easy to refuse. This site does not publish client results it does not have.

5. Selection

Among candidate passages, the one that answers the question with the least leftover ambiguity tends to be usable. Definitions, comparisons, and numbered methods are selected more often than manifestos. That is why this site is a record of answers rather than a slogan page.

6. Naming

Being retrieved is not the same as being named. A page can be a silent source: the model uses a fact and cites nothing, or cites a roundup instead of the company. An LLM citation means the answer names the source. A named source is a company the answer mentions by name. Named North’s work is aimed at the named citation.

What you cannot control

• You cannot force a model to mention you on a given day.
• You cannot freeze wording across model updates.
• You cannot ethically buy a sentence inside the answer.

You can control whether the public record contains a clear, consistent, retrievable sentence. That is the work.