ChatGPT Search vs Perplexity

Citations Per Answer and Per-Page Influence

The Zhang study separates two things that are usually blurred together: “selection,” which is whether a page gets cited, and “absorption,” which is how much of the answer the page actually shapes. Perplexity selects generously. With more than sixteen citations per prompt on average, the odds of appearing in the source list are comparatively good, and the measured influence of any one page is small. ChatGPT selects about seven, and each carries roughly four times the influence.

Two practical points follow. A count of AI citations is not comparable across the two engines, because the same number means different amounts of exposure. And the type of page matters in both: encyclopedic pages scored 0.21 for influence against 0.07 for news pages, while question-and-answer formatting on its own made no measurable difference.

Most-Cited Domains

Ahrefs' September 2026 lists show Reddit leading in both engines, with 16.8% of the top-50 share in ChatGPT and 21.6% in Perplexity. The engines diverge on video. YouTube holds 20.8% of Perplexity's top-50 share and 2.5% of ChatGPT's. Wikipedia is close to level at 7.0% and 6.3%.

These shares need care. They are shares among the fifty most-cited domains, not of all citations. Measured against all citations, the picture is different: Otterly's September 2026 index puts YouTube at 0.196% of Perplexity citations and 0.148% of ChatGPT's, and Promptwatch data reported by Semrush (August 2026; labelled provisional) showed Reddit's share of all ChatGPT citations falling from 3.8% to 0.5% in mid-August. The denominators are not comparable, and the lists move. Goodie (September 2026) counted sixteen unannounced changes in how AI models source from social platforms over seven months.

Overlap between the engines is low. A SIGIR 2026 study found cross-platform source overlap below 0.2 Jaccard, and a small vendor-authored sample (Tannenbaum, September 2026) found 96.4% of cited URLs appeared in only one of four engines. Visibility in one does not imply visibility in the other.

Crawlers and Access

OpenAI documents three relevant agents. GPTBot collects training data and OAI-SearchBot builds the search index; both honour robots.txt. ChatGPT-User fetches pages when a person asks, and OpenAI says that because the action is user-initiated, “robots.txt rules may not apply.” Perplexity documents two: PerplexityBot, which it says surfaces and links sites in search results and “is not used to crawl content for AI foundation models,” and Perplexity-User, which “generally ignores robots.txt rules” because a user requested the fetch.

The symmetry is useful. In both systems the index crawler can be allowed or blocked through robots.txt, and blocking it removes a site from that engine's search results. In both, the user-triggered fetcher is treated as acting for a person and is not reliably governed by the file. A site that wants to appear should allow OAI-SearchBot and PerplexityBot explicitly; a site that wants to block user fetches needs server-side enforcement. Large platforms make different choices here: checked in September 2026, linkedin.com blocked GPTBot while allowing OAI-SearchBot, and tiktok.com and instagram.com disallowed both companies' bots. More detail is on the AI crawlers page.

Freshness, Leadership and Downstream Effect

Seer Interactive (July 2026) found 73% of pages cited by ChatGPT had been updated within the last year, against 65% for Perplexity. Both reward maintained pages; ChatGPT slightly more so in that sample.

On ChatGPT there is evidence about how settled categories are. Semrush and Growth Memo (July 2026; 1,094 categories) found only 15.2% had a clear “owner” and 53.7% were unsettled, though clear owners kept first place 90.4% of the time month over month. Similarweb data reported in June 2026 found people shown a brand in a ChatGPT recommendation were 2.5 times more likely to visit it within seven days, with 55.9% of those visits arriving through branded search. Comparable published studies for Perplexity were not found, so its downstream effect is less well evidenced, not necessarily smaller.