Do AI engines cite the same sources?
No, and the divergence is far larger than most people assume. Analysis of 680 million citations found only about 11% of domains cited by ChatGPT are also cited by Perplexity. One study found roughly 71% of all cited sources appear on a single platform only. Another found that Perplexity, ChatGPT and Gemini share zero cited domains on 35–40% of queries.
Even Google disagrees with itself: AI Overviews and AI Mode reach semantically similar conclusions about 86% of the time while citing the same URLs only around 13.7% of the time.
The practical consequence is that a single blended "AI visibility score" conceals the only information worth having.
What each engine leans on
Different studies produce different exact figures depending on query set and period, but the directional pattern is consistent across independent datasets.
| Engine | Dominant source type | Notable behaviour |
|---|---|---|
| ChatGPT | Encyclopedic and authoritative reference — Wikipedia is consistently its largest single source | Mentions brands far more often than it links to them. Very low overlap with Google's top 10. |
| Perplexity | Community and discussion — Reddit is its heaviest single source, plus news and academic | Cites far more sources per answer than any other engine. Strongly favours recent content. |
| Google AI Overviews | Mixed, with unusually heavy YouTube and Reddit presence | Highest share of brand-domain citations and the most clickable links. Built on Google's index. |
| Google AI Mode | Behaves like a standalone assistant despite being Google | Only a small fraction of citations from Google's organic top 10 — unlike AI Overviews |
| Gemini | Skews toward Google properties | Low source overlap even against Google's own AI Mode |
| Copilot | Bing-based | Fewer citations per answer than Perplexity by a wide margin |
| Claude | Reported to skew toward blogs, but large-scale data is thin | Treat as a surface to check directly rather than model from secondhand figures |
Across all engines combined, a small number of domains absorb an outsized share — Reddit, Wikipedia, YouTube and LinkedIn recur at the top of nearly every study. One synthesis put the top fifteen domains at around 68% of consolidated citation share.
A caution on all of the above. These figures come from separate studies with different query sets, date ranges and methods, and they disagree on specifics. Treat them as direction, not precision. And they move: documented cases exist of a single platform's source mix shifting by tens of percentage points within weeks of a platform change.
Why the engines diverge
Three structural reasons, and they explain more than any individual statistic.
Different indexes. Perplexity runs its own index alongside third-party search infrastructure. Copilot is Bing-based. Google's surfaces sit on Google's index. ChatGPT combines its own crawl with a search partnership. They're not reading the same web.
Different retrieval philosophies. Perplexity behaves retrieval-first, citing densely and tying sources to specific claims. ChatGPT extracts more deeply from fewer sources and compresses harder during synthesis. Google's surfaces assemble across clustered sub-intents.
Different trust weighting. ChatGPT's preference for encyclopedic sources and Perplexity's for community discussion are product decisions about what constitutes a credible answer, not accidents.
What this means for you
Pick your engines deliberately. You cannot optimise for all of them equally with a small budget. Decide which your buyers actually use. B2B software buyers behave differently from consumers researching a local service.
Measure per engine, always. Any report giving you one AI visibility number has averaged away the thing you needed to know. If you're absent from ChatGPT and strong in Perplexity, that is a completely different problem from being weak everywhere — and a blended score shows them identically.
Third-party sources do more work than your site. Every engine draws heavily on properties you don't own. Community platforms have been measured taking a larger share of citations than brand domains overall. Being accurately described on the sources your engines favour matters more than another page on your domain.
Freshness is engine-specific. Perplexity weights recency heavily. Others less so. If Perplexity matters to you, publication and update cadence matters more than it otherwise would.
Don't infer one engine from another. Strong AI Overview presence tells you close to nothing about ChatGPT, because AI Overviews inherit Google's ranking and ChatGPT largely doesn't.
The measurement problem nobody mentions
Conventional analytics cannot see most of this.
Cloudflare's network analysis found AI crawlers ingesting content at enormously disproportionate rates relative to the referral traffic they send back — in one case tens of thousands of pages crawled per user referred. Your content is being consumed and used in answers that never appear in your referral data.
The workable proxies: branded search volume, direct traffic, and manual or tool-based citation tracking. None is clean. All three together are better than assuming your analytics can see it, which they can't.
Frequently asked questions
Do ChatGPT and Perplexity cite the same sources? Rarely. Analysis of 680 million citations found only about 11% of domains are cited by both, and on a substantial share of queries the two share no cited domains at all.
Which AI engine cites Reddit most? Perplexity, by a wide margin in most studies. Reddit is also heavily represented in Google's AI surfaces, and considerably less so in ChatGPT.
Does Perplexity use Google rankings? More than ChatGPT does — its overlap with Google's top 10 has been measured around 29%, against 6–8% for ChatGPT — but the majority of its citations still come from pages outside Google's first page.
Why do AI Overviews and AI Mode cite different sources? They use different retrieval logic despite both being Google products. They agree semantically about 86% of the time but cite the same URLs only around 13.7% of the time, and AI Mode draws far less from Google's organic top 10.
Should I optimise for all AI engines? Not with a limited budget. The source pools barely overlap, so effort doesn't transfer. Identify which engines your buyers use and concentrate there.
Which AI engine sends the most traffic? Google's AI Overviews link most readily and sit inside a search environment. ChatGPT mentions brands far more than it links to them. Perplexity cites densely and tends to tie sources to specific claims.
How often do AI citation patterns change? Frequently enough that quarterly checking is a floor. Documented cases exist of a single platform's source mix shifting by tens of percentage points within a few weeks.
Measure per engine, not in aggregate
The GEO/AEO Discovery Audit reports mention rate, citation rate and share of voice broken out by engine, along with which sources each one is using instead of you — the source attribution that tells you where the work actually is.
For an ongoing baseline across Google and the AI engines, the Visibility Benchmark records the starting point so later movement is provable.