Summary
Most modern answer engines run two distinct steps. First they pull a candidate pool of documents into the model's context, a stage commonly called retrieval or grounding. That is being sourced. Then they decide which of those documents to credit in the visible answer. That is being cited. The exact mechanics differ by engine, and not every vendor publishes the details, but the sourced-versus-cited gap shows up consistently in the output: a page can be read by the model without being credited on screen. The gap between the two is where most AEO and GEO work happens.
The two-step pipeline behind every AI answer
Most answer engines, including Perplexity, Google AI Overviews, Bing Copilot, and the search modes of ChatGPT and Gemini, work in two phases. The retrieval phase pulls a candidate pool of documents from a live index. The generation phase reads those documents and writes an answer, choosing which of the candidates to credit on screen.
Marketers care about both ends of that chain. The retrieval phase decides which URLs become sources. The generation phase decides which of those sources show up to the user as citations.
What does "sourced" mean?
A citation is the explicit, user-facing pointer that says "this explicit part of the answer used this URL." It is what the reader sees, whether the link sits inline in the response or in a reference list at the end. Either way, it counts as a citation and delivers the click.
What does "cited" mean?
A citation is the explicit, user-facing pointer that says "this part of the answer used this URL." It is what the reader sees as an inline link inside the response.
Frequently asked questions
Is "sourced" the same as "indexed"?
No. Being indexed means a crawler has your page. Being sourced means a specific live query pulled your page into the model's candidate set for that prompt. Indexing is necessary but not sufficient.
What does it mean if I'm sourced but not cited?
It means the model pulled your page into its candidate pool, read it, and chose to credit someone else. The page cleared retrieval but failed the extraction or attribution step that follows. Common causes: claims buried in long paragraphs instead of surfaced in headers and short labeled chunks; unclear authorship, so the engine prefers a source where responsibility is obvious; or a competitor presented the same information more cleanly and took the slot.
Do all engines show citations?
No. Coverage and density vary by engine, by query type, and by surface (chat vs search-style answer). Some engines render explicit reference lists, others lean on inline links, and the same engine can return a heavily cited answer for one prompt and an uncited paraphrase for the next. Measuring both sourcing and citation across multiple engines gives a truer picture than tracking citations alone.
What is the fastest way to improve citation odds without rewriting everything?
There is no single fix that works on every page. Tools like Temso AI isolate issues inside of your content, so the optimization you make matches the actual problem instead of a generic best practice.
