AI Overviews are the summaries Google shows above the classic results for a growing share of queries. They are generated by Gemini models grounded in Google's own index, and they cite a set of source links. Because Google publishes little about source selection, the best evidence comes from large third-party crawls, and those crawls agree on the broad picture.
Most citations already rank, but not all
Studies from SEO data providers, including Ahrefs and seoClarity, have repeatedly found that a majority of AI Overview citations come from pages ranking in the top ten organic results for the same query, with estimates ranging from roughly half to three quarters depending on the study and the query set. The rest come from deeper in the index, which is the interesting part: a page ranking twentieth can still be cited when it contains the one passage that answers the question directly.
Passages, not pages
The Overview is assembled from passages. Google's own documentation for featured snippets and its passage ranking systems describe the same mechanics that show up in Overview citations: a self-contained paragraph under a heading that matches the question, answered in the first sentence. Long narrative pages that make the reader assemble the answer are poorly suited to this, however authoritative they are.
What Google says on the record
Google's public guidance is that there is no special markup or optimization for AI Overviews, and that the usual advice applies: helpful content, technical health, and crawlability by Googlebot. Notably, you cannot opt out of AI Overviews while staying in the classic results; the nosnippet and max-snippet controls limit both together.
What to do as a publisher
- Keep ranking: classic rankings remain the strongest predictor of being cited.
- Structure every important section to be quotable alone, one question per heading.
- Use precise, dated facts; Overviews visibly favor passages they can attribute cleanly.
- Track impressions in Search Console, which now folds AI Overview appearances into its reports.
The playbook overlaps heavily with the other engines. For the differences, compare how ChatGPT picks citations and how Perplexity does.