ChatGPT and Perplexity barely cite the same sources.

When ChatGPT and Perplexity answer the same question, they tend to pull from almost completely different places. One audit found that only about 11% of the domains the two engines cite actually overlap, which means roughly nine out of every ten sources are unique to a single platform. That alone should put an end to the idea of one tidy "AI search" you can optimize for, because each engine has developed its own taste in what counts as a citable source.

It is an awkward finding, because most GEO advice still gets written as though every engine behaves the same way. They clearly do not, and the gaps are wide enough that a page racking up citations in Perplexity can stay completely invisible in ChatGPT, so it is worth understanding what separates them before you spend any effort optimizing for either.

There is no single AI search, there are several with different tastes

The cleanest way to think about it is to stop picturing a single destination and start picturing a handful of separate ones, each with its own door policy. Perplexity, ChatGPT, Google's AI Overviews, and Gemini all run their own retrieval and ranking logic, and the research that has tried to compare them keeps landing on the same conclusion, which is that citation patterns vary so much between engines that a strategy tuned for one will underperform on the next.

What makes this harder than old-school SEO is that you cannot reverse-engineer one algorithm and call it done. The overlap is small enough that winning on one platform tells you very little about how you will do on another, so the work multiplies rather than transfers. That is inconvenient, but it is also where the opportunity sits, since almost nobody is actually optimizing per engine yet.

Perplexity cites almost everything, ChatGPT is choosy

The starkest difference is simply how often each engine cites at all. Perplexity is built as an answer engine that retrieves on every query and attaches inline numbered citations to basically every factual claim it makes, averaging around 21.9 citations per response against ChatGPT's 10.4. One live 50-prompt test reported by Tech Insider pushed that contrast to its extreme, with Perplexity citing a source for 100% of factual claims, ChatGPT in browsing mode citing only 47%, and ChatGPT answering from memory alone citing nothing at all.

That last number is the one to sit with for a moment, because it explains a lot. When ChatGPT decides it already knows the answer and skips retrieval, there is no citation to win no matter how good your content is, whereas Perplexity gives you a shot on nearly every answer it produces. The two engines are not just ranking sources differently, they are deciding whether to look for sources at all on completely different terms.

What ChatGPT actually rewards when it does cite

Once ChatGPT is in browsing mode and actually choosing sources, its preferences are measurable. ZipTie's analysis weighs its selection at roughly 40% domain authority, 35% content quality, and 25% platform trust, which tilts the whole thing toward established sites in a way Perplexity does not to the same degree. Freshness matters even more than most people expect, with content updated inside the last 30 days earning about 3.2 times more citations than stale pages, so a regular update cadence on your key pages is one of the few reliably ChatGPT-friendly moves you can make.

Placement turns out to matter too, since the same analysis found that the first 30% of a page accounts for 44.2% of all the citations it earns. The practical read is that burying your strongest, most quotable answer halfway down the article is leaving citations on the table, and front-loading it gives ChatGPT the clean, authoritative passage it is looking for early.

The academic baseline still holds up

For all the platform-specific quirks, there is a stable foundation underneath all of this, and it comes from the original GEO paper out of Princeton, Georgia Tech, IIT Delhi, and the Allen Institute, presented at KDD 2024. Working across GEO-bench, a benchmark of 10,000 queries spanning eight domains, the authors showed that a handful of on-page tactics could lift a page's visibility in generative answers by up to 40%, with adding relevant statistics, citing sources, and including direct quotations driving the biggest gains.

Those tactics still work, and they are a sensible baseline whichever engine you care about, but the paper also showed the effects varied by query type and engine, which is exactly the pattern everything since has confirmed. So the foundation is real, you just cannot treat it as the whole job, because the engine-specific layer sitting on top of it is where a lot of the citations are actually won or lost.

What an engine-specific strategy looks like in practice

The shift is to stop writing for a generic "AI" and start treating each surface as its own channel. For Perplexity, the priority is making sure every meaningful claim is backed by a clear, retrievable source, because the engine cites inline and rewards content it can attribute cleanly. For ChatGPT, lean harder on authority signals and a genuine update cadence, and put your strongest answer in the first third of the page where most of its citations come from.

Underneath those per-engine moves, keep the GEO baseline running everywhere by adding statistics, quotations, and cited sources to anything you want surfaced, since that lifts you across the board. The one habit that ties it together is measuring citations per engine separately rather than as a single blended number, because an average across platforms will hide exactly the gaps you need to see.

There is no single finish line anymore, just several, each with its own rules. The advantage goes to the brands that treat ChatGPT, Perplexity, and Google's AI Overviews as separate surfaces and optimize for each on its own terms, while everyone else keeps writing for an "AI search" that does not really exist.