How an AI assistant decides which businesses to name
Short answer
An assistant handling a recommendation question generally searches, retrieves a handful of pages, pulls candidate businesses out of them, filters those against the constraints in the question, and writes the answer from whatever survives. You can influence three stages: whether your page can be fetched and read at all, whether the attributes the question filters on are published as data, and whether independent sources describe you in the same terms. There is no submission form and no setting anywhere. Ranking first on Google does not carry over either, because Google renders JavaScript and leans on your Business Profile while ChatGPT's crawler reads raw HTML and grounds local answers elsewhere.
The pipeline from question to recommendation: retrieval, entity extraction, filtering and corroboration, and which stages a business can influence.
The stage that decides most answers is the one nobody can sell you, which is whether people who are not you describe your business in the same terms. Reddit made that concrete in August 2026: per ExplainX's write-up, citing data attributed to Promptwatch, a blanket robots.txt disallow took Reddit's share of ChatGPT Search citations from roughly 3.83 percent to 0.52 percent in days, while Google's AI surfaces moved far less. The off-site surface is therefore four surfaces, and any visibility resting on a single third-party property can be erased by a line in a file you do not control. NovaTechRay writes up the mechanism rather than the tactic because the tactics keep expiring and the pipeline does not. If a vendor describes this stage as something they can guarantee, they are describing something else.