Why ChatGPT and Perplexity cite different sources (only 11% overlap)

ChatGPT and Perplexity cite the same domains only about 11% of the time, and the reason isn't taste — it's architecture. As of 2026, cross-platform studies from Profound and Leapd put the overlap in cited domains near 11%, because ChatGPT answers from a curated, authority-weighted index where Wikipedia is the most-cited domain, while Perplexity retrieves the live web on every query and leans on fresh, discussion-heavy pages, with Reddit at the top. A page tuned to win one is usually not the page that wins the other.

Store owners come to us holding a single "AI visibility score" and expect it to move every assistant at once — and that flat 11% is exactly why it can't. We built Contexta partly because the one-number view hides the thing that decides everything: which engine is actually reading you and sending anything back. What follows is the per-engine logic behind the 11% — what each system rewards, the single layer of work that genuinely carries across all of them, and how to decide where the rest of your effort goes instead of spreading it evenly and thin.

Why do ChatGPT and Perplexity cite different sources?

They cite different sources because each assembles its answer from a different pool of pages, and those pools barely intersect: overlap in cited domains sits around 11%, and roughly 71% of cited sources appear on only one platform. This isn't two engines ranking the same shortlist in a different order — much of the time they're reading different shortlists entirely.

~11%

of cited domains overlap between ChatGPT and Perplexity

71%

of cited sources appear on only one AI platform

21.9 vs 10.4

citations per answer — Perplexity cites more than twice as many as ChatGPT

615x

swing in citations for one brand across platforms (Superlines, March 2026)

The fragmentation isn't limited to rival companies, either. Google's own AI Overviews and AI Mode cite the same URL only about 13.7% of the time, so "appear in Google's AI" isn't even a single target — a gap we unpack in how AI Mode and AI Overviews differ. And it's widening rather than converging: Superlines' March 2026 analysis found citation volume for the same brand swinging by up to 615x from one platform to the next.

If AI crawlers can already reach your page, why isn't that enough to get cited?

Because being reachable makes you eligible, not chosen. Clearing the crawl-and-parse floor — the bot fetches you, no index rule blocks you, your text survives without JavaScript, a clean passage can be lifted — puts you in every engine's candidate pool; it doesn't decide which engine then picks you out of it. That floor is the one part of AI visibility you genuinely optimize once, and the five reasons AI skips a page entirely all live there, beneath any per-engine tuning.

Everything above the floor is where the 11% comes from. Two pages can both be perfectly crawlable, well-structured and factual and still land in different engines' answers, because the rankers sitting on top weigh authority, freshness and source type differently. So the honest mental model is two layers: a shared floor you build once, and a per-engine layer you can't.

What does ChatGPT reward when it picks a source?

ChatGPT rewards established authority and stable reference pages, which is why Wikipedia is its single most-cited domain — reported near 47.9% of its top citations in 2026. It also cites fewer sources per answer than Perplexity, about 10.4 on average, so the bar to be one of the chosen few is higher, and it tends to clear in favor of pages that other trusted sources already corroborate.

The mechanism behind the preference is retrieval plus training. ChatGPT's search index is built by OAI-SearchBot and weighted toward durable, well-linked pages, while the model's own priors lean toward canonical references it saw repeatedly during training. For a store or blog, that means the work that moves ChatGPT is the slow kind — clear entity identity, consistent naming, and third-party corroboration, the same experience and authority signals E-E-A-T describes — not simply publishing more often.

What does Perplexity reward instead?

Perplexity rewards freshness and discussion, almost the mirror image of ChatGPT. It retrieves the live web on every query rather than reading a pre-built index, cites far more sources per answer — about 21.9 on average — and heavily favors recent and forum content: Reddit is its top domain at roughly 46.7%, and pages under 30 days old are reportedly cited around 3.2x more than older ones.

ChatGPT — rewards authority

• Answers from a curated index built by OAI-SearchBot, not the live web per query
• Wikipedia is its single most-cited domain (~47.9% of top citations, 2026)
• Cites fewer sources per answer (~10.4), so the bar to make the list is higher
• Favors stable, corroborated reference pages over recent ones

vs

Perplexity — rewards freshness

• Retrieves the live web on every query through PerplexityBot
• Reddit is its top domain (~46.7%); pages under 30 days old cited ~3.2x more
• Cites far more sources per answer (~21.9), widening the field it draws from
• An authoritative but stale page underperforms a fresh, specific one

The engine of that preference is live retrieval: PerplexityBot crawls in near-real-time and the ranker weights recency, so an authoritative-but-stale page loses to a fresh, specific one on the same topic. Recency isn't a per-page setting either — how many of your pages can be fresh at once is capped by your publishing cadence, which is the 13-week window argument in full. It also means reachability by that specific crawler is non-negotiable — allowing "AI bots" in general is not the same as allowing PerplexityBot by name, a distinction that trips up more sites than it should and one we lay out in which AI crawlers to allow.

Where should you spend if you can't optimize once for both?

Build the shared floor once, then let measured traffic — not guesswork — decide the rest, because no single page tops ChatGPT's authority ranking and Perplexity's freshness ranking with identical work. The sequence that holds up: fix the floor so you're eligible everywhere; find which engines actually cite you and send visitors; then tune for the one or two that matter to your store, not all four in parallel.

That middle step is the one most sites skip, and it's where Contexta's AI Traffic tracking earns its place — it counts the real visitors arriving from ChatGPT, Perplexity, Gemini and Copilot separately, by referrer and utm_source, so the decision about where to spend rests on who is actually sending buyers rather than on a generic "optimize for AI" checklist. For most stores one or two engines dominate their real referrals; pouring effort into a third that sends no one is motion, not progress. And once you know which engine to chase, what to actually build differs by where each one pulls its sources from — consensus for ChatGPT, your own site for Gemini, reviews for Perplexity. The 11% overlap isn't a problem to solve — it's a map telling you your effort has to be allocated, not duplicated.