How ChatGPT, Perplexity, Gemini and Copilot Actually Choose Their Sources

ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews are built on different underlying models, different retrieval infrastructure and different product goals — a shopping-focused answer, a research-focused answer and a quick-summary answer are not optimizing for the same thing.

That is why the same prompt can produce a confident citation in one system and a generic, sourceless answer in another. Treating 'AI search' as one target instead of five distinct systems is the most common reason AEO efforts show inconsistent results.

Perplexity is built around live web retrieval for most queries, and its product surfaces citations prominently as part of the answer itself. That structure rewards pages that are fresh, specific and easy to extract, since the retrieval step happens close to query time rather than from a static trained snapshot.

Because retrieval is central to the product, Perplexity tends to be one of the more responsive systems to a recent content update — a page fixed today can plausibly be retrieved and cited within the next monitoring run, faster than in systems that rely more heavily on a pretrained snapshot.

ChatGPT answers draw on both its trained model and, when browsing is invoked, live retrieval — and it does not always show its sourcing as explicitly as a retrieval-first product does. That makes ChatGPT citation behavior more variable: some answers cite openly, others answer from trained knowledge without a visible source at all.

This is also why ChatGPT citation rates differ sharply by category and by whether a query clearly needs current information. A well-established, slow-changing fact is more likely to be answered from trained knowledge; a comparison, price or recent-news question is more likely to trigger visible retrieval and citation.

Gemini and Google's AI Overviews sit closest to Google's existing web index and ranking signals, which means classic technical SEO health — crawlability, indexation, structured data, page quality — still matters directly to whether a page is even eligible to be surfaced, before AEO-specific factors come into play.

This is the clearest case where SEO and AEO are not separate disciplines: a page that is not well indexed by Google is unlikely to appear in Google's own AI surfaces regardless of how well-structured its content is for extraction.

Copilot's web-facing answers are built on Bing's index and retrieval, which historically has weighted authority, structured data and domain trust somewhat differently than Google. Pages that are strong on Bing's own search signals tend to have an easier path into Copilot citations.

Copilot is also used heavily inside productivity and enterprise contexts, where answers may lean toward established, well-known sources over newer or smaller domains — a relevant factor for niche or emerging brands trying to earn a first citation there.

Because each system retrieves, weighs and cites differently, a single 'AI visibility' number that blends all providers hides exactly the information you need to act. A page performing well in Perplexity and invisible in Gemini points to a different fix (content freshness vs. core Google indexation) than the same page struggling everywhere.

Tracking mentions, positions and citations separately by provider — and re-running the same prompts across all five systems in AI Control — is what turns 'we're not showing up in AI search' into a specific, fixable diagnosis.