How Does ChatGPT Decide Which Companies to Recommend?

The short answer

OpenAI has not published a formula for how ChatGPT selects which companies to name in an answer. Based on OpenAI's own documentation of how ChatGPT search retrieves and cites sources, and on patterns observed across generated answers, relevance to the query, the authority and quality of retrieved sources, corroboration across multiple sources, and clarity of the underlying content all appear to play a role.

What OpenAI has documented

OpenAI's Help Center describes ChatGPT search as rewriting a user's prompt into targeted search queries, retrieving results from the web, and generating a response grounded in those results, with citations back to sources. This confirms that retrieval and citation are real, documented mechanisms. It does not disclose a specific ranking algorithm for which retrieved sources get named directly in the generated text versus simply cited as a link.

Relevance

The most basic factor is straightforward: a company needs to have content that actually and specifically addresses the question being asked. A mortgage lender's homepage that only says "we help you find your dream home" is less useful to a retrieval system than a page that directly answers "can a self-employed borrower qualify for a conventional loan."

Authority and source quality

Retrieval systems generally weigh the credibility of a source. A page on a well-established, topically relevant domain with a track record of accurate information is more likely to be treated as trustworthy than a low-authority or newly created page making the same claim. This mirrors long-standing search engine authority signals, including backlink profiles and domain reputation.

Third-party corroboration

Observed patterns suggest that generative systems weigh whether a claim is corroborated by multiple independent sources, not just stated once by the company itself. A lender that is described consistently across its own site, industry directories, and reputable press coverage presents a more resolvable, trustworthy entity than one whose only public description exists on its own marketing pages.

Content clarity and entity recognition

Content that clearly states facts — who the company serves, what it offers, where it operates — is easier for a system to extract and represent accurately. Entity recognition refers to the system's ability to resolve a company as a single, well-defined thing with a consistent set of facts, rather than an ambiguous or inconsistently described name.

Reviews, reputation, freshness, and citation sources

Where relevant to the query, reviews and reputation signals can factor into which companies appear more favorably in a generated answer, similar to how they influence traditional local and product search. Freshness also matters: outdated information, including outdated rates, programs, or service areas, reduces the reliability of a source even if it was once accurate. The specific sources ChatGPT cites in an answer — whether news sites, review platforms, or company pages — give some insight into which types of sources the system currently treats as authoritative for a given topic.

• Relevance: does the content directly answer the specific question asked?
• Authority: is the source domain credible and topically established?
• Corroboration: do independent sources describe the company consistently?
• Clarity: are facts stated directly and unambiguously?
• Freshness: is the information current and free of outdated claims?

What is documented, observed, and Prlmnt's own analysis

It is important to distinguish three levels of confidence here. What OpenAI has documented: that ChatGPT search rewrites queries, retrieves web sources, and generates answers with citations. What is observed: patterns across many generated answers suggesting relevance, authority, corroboration, and freshness correlate with which companies appear. What is Prlmnt's own analysis: the framework above for organizing these factors into a practical checklist for companies evaluating their own AI visibility. None of this should be read as a disclosed OpenAI ranking formula, because OpenAI has not published one.