Observation · Microsoft Copilot

How do AI search engines choose which sources to cite?

Copilot described retrieval and grounding as the common sequence: interpret the query, retrieve candidate passages, filter for relevance and trust, generate an answer, then attach supporting citations.

15 November 2026, 18:29 CSTen-US (English answer; zh-CN interface)Run 2 of 3Method v0.1
Question

Recorded prompt

“How do AI search engines choose which sources to cite?”
Product
Microsoft Copilot
Surface
Microsoft Copilot web chat with web search
Model
Copilot Auto (model version not exposed)
Locale
en-US (English answer; zh-CN interface)
Observed answer
Open session ↗
Observed
15 November 2026, 18:29 CST
Fresh conversation
Yes
Answer snapshot

What the answer said

Copilot said the citations usually come from material retrieved at query time rather than model memory. It outlined five stages: analyze intent and generate internal queries, retrieve candidate sources from an index or corpus, rank and filter at passage level for relevance, authority and freshness, generate the answer from the retrieved context, then map claims back to documents and display supporting links. It said direct answers, clear factual passages, trust and extractability make citation more likely, and that strong traditional ranking does not guarantee AI citation.

Linked citations

6 recorded sources

Limitations

What this observation cannot show

Authenticated Microsoft Copilot web-chat run in a fresh conversation. Copilot Auto did not expose model or retrieval-index details, and the source panel exposed domains and titles but not stable destination URLs.