ChatGPT citations

ChatGPT does not cite sources on every answer. The model decides per turn whether to invoke its web-search tool. It usually invokes search when the question contains time-sensitive language (today, latest, this year), names a specific company or product, asks for recommendations, or covers a topic the model is uncertain about. For straightforward general-knowledge questions it answers from training data without citations.

When search runs, ChatGPT typically retrieves 5 to 15 pages, reads them, and renders an answer with 3 to 8 citation links pointing to the pages it used. The exact citation count varies with answer length and topic depth.

ChatGPT uses Bing for the underlying web index, then re-ranks results with its own model before deciding which pages to actually read. The re-ranker favours: pages whose visible text answers the user's question directly (extractive content beats marketing fluff), pages with clear authorship and date signals (schema markup helps), pages on domains the model has prior signal for in this category, and pages that load fast enough for the headless fetch to complete.

Domains that appear most often in ChatGPT citations follow predictable patterns: Wikipedia, GitHub, Stack Overflow, Reddit, government and academic sites, plus the top trade publications and review sites per category (G2, Capterra, NerdWallet, Healthline, and so on). For commercial categories it also pulls heavily from individual brand domains when the brand is the most direct source for the answer.

Two different signals. Being cited means your URL is in the source list, the AI used your content as evidence. Being named means your brand name appears in the answer text, the AI is actively recommending you. The two correlate but are not the same. A comparison page might cite ten brands' sites while only naming three in the lead paragraph. A well-known brand might be named from training-data memory without any citation.

For commercial intent the named signal usually drives more attributable traffic. For technical or research queries citations drive most clicks because users follow the links rather than relying on the summary.

Three options in order of effort. (1) Manual spot checks: open ChatGPT in a private window, run 10 to 20 queries a buyer in your category would actually ask, screenshot the citations and tally how often your domain appears. Cheap, repeatable, no setup. (2) Browser extensions that log ChatGPT citations into a CSV. (3) Run a programmatic audit that scripts the same queries through ChatGPT's API and aggregates the citation lists across hundreds of runs, which is what monitoraeo's full audit does. The full audit also tracks citation rate over time so you can see whether changes you ship are actually moving the number.