What Source Selection means
Retrieval produces candidates; selection narrows them to the handful that appear. Being retrieved is necessary but not sufficient.
Selection weighs relevance to the specific sub-question, extractability of the passage, credibility of the source, and how well the claim is corroborated elsewhere.
Why it matters for AI search
This is where most AI SEO effort should concentrate after retrieval is fixed. Plenty of sites are retrieved constantly and cited rarely.
It also explains why a lower-ranked page can be cited over a higher-ranked one — selection isn't ranking.
See it in action
Forty candidates, three citations. What separated them.
High authority, vague page
Retrieved. Nothing specific enough to quote — no numbers, no method.
Mid authority, specific page
Retrieved. States pricing ranges and methodology explicitly. Corroborated on two platforms.
High authority, gated content
Retrieved the landing page. The substance is behind a form.
Authority lost to specificity twice.
Selection rewards checkable, extractable substance. A page that states real numbers beats a more authoritative page that states nothing a model can quote.
How to get it right
Improving source selection odds
- Make sure passages are self-contained and quotable, not just present
- State specifics — real numbers, documented methodology, named outcomes
- Build corroboration so claims aren't single-sourced to your own domain
- Ungate information you want cited; a model can't quote what's behind a form
- Keep facts current, since stale sources get deprioritised in selection
Common questions
Why are we retrieved but not cited?
Usually because no passage is cleanly extractable, or because your claims aren't corroborated elsewhere so the model treats them cautiously.
Does domain authority decide selection?
It contributes but doesn't decide. Specific, extractable, corroborated content from a mid-authority site regularly beats vague content from a stronger one.
Can we see why we weren't selected?
Not directly. Perplexity shows its source list, which lets you see who won and inspect what they did differently — that's the closest available diagnostic.
Related concepts
These come up alongside Source Selection constantly.