SERP teardown: which 6 pages got their passages pulled into AI Overviews

SERP teardown: which 6 pages got their passages pulled into AI Overviews

Question: when an AI Overview cites sources, what do the cited pages share? A teardown of one query cluster.

What was done: for 30 informational queries in a finance subtopic, I logged which URLs were cited in the AI Overview and pulled the exact sentence ranges that matched the generated summary. I then compared cited vs. non-cited top-10 pages on structural features.

Outcome: cited passages were disproportionately (in 24 of 30 cases) self-contained, definition-style sentences placed directly under a question-matching heading — average 28 words, no pronoun dependency on prior text. Non-cited pages buried the same facts inside longer, reference-heavy paragraphs.

Caveat: 30 queries in one vertical is a thin sample, and "self-contained sentence under a matching heading" correlates with overall editorial clarity. So this may reflect quality broadly, not a specific extractability lever.

Method note: manual logging of 30 AI Overviews + passage diffing against page HTML. One vertical, single snapshot in time.

Confidence: low-to-medium — clear pattern, tiny and time-sensitive sample.