source attribution

Attribution is the mechanism that turns an AI answer from a black box into something checkable. Instead of only returning prose, the engine also exposes which pages informed it, and it ties specific sentences to specific sources.

How attribution works across engines

The engines wire it up differently, but the mechanics rhyme. A retrieval step pulls candidate pages, a ranker decides which ones inform the answer, and the model writes text with citations pinned to passages. Perplexity’s own docs show this plainly: the model inserts numbered markers like [1] and [2], and each id maps to a URL returned in the search_results object. Their rule is blunt: “Never ask the model to generate source URLs,” because a citation should point at a verified retrieval, not a guessed link.

Google handles it inside Search. AI Overviews and AI Mode may use a “query fan-out” technique, issuing multiple related searches across subtopics, and eligibility is simple: a page must be indexed and eligible to be shown in Search with a snippet. There is no special markup, no AI text file, no new schema to add.

Attribution is not uniform. Citation studies show low overlap between engines, so a page ChatGPT cites often is not the one Perplexity picks for the same query. Being the answer on one engine does not guarantee it on another.

It also is not the same as being correct. A Columbia Tow Center study of 1,600 queries found answer engines regularly attach a citation to a claim the cited page does not actually support, so treat a citation as a pointer, not a proof.