AI cites a lower-ranked page over the #1 result when that lower page stated the specific answer more clearly and extractably, or matched the exact sub-question better, than the top result did. The system is optimizing for the best answer-chunk to lift and attribute, not for the page that happens to rank highest, so extractability and precision can beat position. This is observed behavior in a fast-changing space, so it is worth confirming, but the logic behind the upset is consistent.

The mechanism starts with what the system is actually doing. It is not ranking pages and citing the winner, it is looking for a passage that answers the specific thing being asked and can be quoted with confidence. The #1 result earned its rank by being the best overall page for the broad query, which is not the same as containing the cleanest answer to the narrow question in front of the model. When those two diverge, citation follows the answer, not the rank.

So a lower-ranked page wins in two common ways. First, it states the answer more cleanly: where the top result spreads its answer across paragraphs or implies it, the lower page puts it in one self-contained, liftable sentence backed by specifics. The system takes the chunk it can use. Second, it matches the sub-question more exactly: the top result may be the best page for the general topic while the lower page is squarely about the precise thing asked, so its passage fits the query better even though its overall standing is lower.

This is why the upset is logical rather than a fluke. Ranking answers “which page is best for the query,” citation answers “which passage best answers this exact question, clearly enough to quote.” A page can lose the first contest and win the second.

When a lower-ranked page gets cited instead of yours, do not just chase a higher rank. Find the specific sub-question being answered and sharpen your page’s answer-chunk so it states that answer more cleanly and precisely than the page that beat you.