ChatGPT, Perplexity and 6 Others: Citations Per Answer Vary Up to 9x

A new study found that AI search engines vary enormously in how many sources they cite per answer. Qwairy's Luca Fancello analyzed 118,101 AI-generated answers and 669,065 citations across eight AI platforms — ChatGPT, Perplexity, Google AI Overview, Gemini, Microsoft Copilot, Claude, Grok, Mistral, and DeepSeek — from Q3 2025.

Perplexity cited the most sources per answer at 21.87, followed by Google AI Overview at 17.93, Gemini at 17.11, and ChatGPT at 7.92. Microsoft Copilot cited just 2.47 sources per answer — roughly a ninefold gap from Perplexity, the highest citer.

Citing more sources didn't necessarily mean drawing from a more diverse pool, though. Copilot used only 111 unique sources in total, while ChatGPT, despite citing less per answer, drew from 42,592 unique sources — the most diverse of any platform. Perplexity (37,399), Google AI Overview (25,785), and Gemini (20,549) followed.

Reliance on Wikipedia also split sharply by engine: 4.8% of ChatGPT's citations came from Wikipedia, while every other platform's Wikipedia share was effectively zero. The researchers concluded that "platform-specific optimization isn't optional — it's essential," since a single GEO strategy misses most of the citation landscape across other engines.

A separate analysis of 680 million citations by CiteMetrix points in the same direction. ChatGPT mentioned a brand in only 0.59% of answers, while Perplexity did so in 13.05% — 22 times more often. ChatGPT's citations correlated more strongly with domain authority (r=0.74), favoring major publishers and established outlets, while Perplexity's correlation was looser (r=0.61). The two also differ on freshness: only 55% of what ChatGPT cited was published within the last 30 days, versus 82% for Perplexity, which clearly favors recent content.

In Korea

This research covers overseas platforms and content, but the implication for Korean sites is clear. Polishing content for ChatGPT in an encyclopedia-like, reference style won't necessarily produce the same results on Perplexity or Google AI Overview. It makes more sense to first decide which AI service you most want traffic from, then optimize separately for the content format that platform actually tends to cite.