How does ChatGPT choose which sources to cite?
OpenAI's search guide says ChatGPT uses several ranking factors to seek relevant, reliable information; it does not disclose a complete citation formula. In our 22 September 2026 construction capture, fieldwire.com appeared as a cited host in 52 of 150 answers across five engines. ChatGPT itself named Fieldwire in 9 of 30 responses. Those are observed outputs you can inspect, rather than weights explaining the source choice.
What we measured
Reported. OpenAI's web-search guide, read on 29 October 2026, describes rewriting a question into targeted queries sent to search providers. It says search responses may contain citations and that a Sources view can also contain other relevant links. The guide warns that results and citations can be incomplete, outdated or incorrect.
Measured. The September Index counted fieldwire.com as a cited host in 52 of 150 valid construction answers, once per answer. This figure pools five engines. Separately, the ChatGPT engine-presence cells total 9 Fieldwire naming answers across its 30 valid responses.
These observations cannot reveal why ChatGPT selected a particular cited page.
What this means for you
Reasoned. Read the two Fieldwire counts. The two Fieldwire figures require their scope labels. The 52 cited-host answers do not give a ChatGPT-only citation rate. The 9 naming answers do not establish that ChatGPT cited Fieldwire's domain in those responses. Keeping those events separate prevents a vendor-presence chart from becoming an invented account of source selection.
The question rewrite also matters when you compare receipts. OpenAI describes sending one or more targeted searches, potentially followed by more specific queries. An original buyer question and the eventual web queries therefore need not have identical wording. The public guide describes that possibility without giving a receipt for the exact searches behind our September answers.
Location and saved context are further conditions named by the guide. If you compare a local product-search answer with a colleague's result, preserve the stated location and relevant context beside the capture. A different source list alone cannot isolate which of those conditions changed the result.
Also asked as
How does ChatGPT choose which sources to cite?
Related
- OpenAI search and citation guidance
- Construction source observations
- Fieldwire public record
- Checking a cited claim
Next step
Use Free 10-question check (one engine) to inspect a defined set of buyer questions. No email is needed; runs are limited per network and per day, and by the site's daily allowance.
For one documented gap, Fix one answer, $490. Keep the question and source receipt with the requested change.
Claiming or correcting a vendor record is free and never depends on a purchase.
Every figure comes from the published method, v1.1.
Prepared by the Broadcastwell team.
Page owner: Vikram Patel, AI Architect. Corrections: hello@broadcastwell.com