Do AI Engines Cite the Same Sources?

Do ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews cite the same sources?

No. Foglift’s frozen Q3 2026 panel found a mean pairwise Jaccard score of 0.094 across the five engines. Of the 92 domains in their combined top-25 lists, 69 appeared in only one engine's top 25 and none appeared in all five. A citation win in one engine therefore does not establish visibility in the other four.

Headline findings

AI citation visibility remained an engine-by-engine problem in Q3. The combined top-25 lists contained 92 domains. Of those, 69 appeared in only one engine's top 25, and none appeared in all five. A single-engine monitor therefore misses most of the sources shaping buyer answers elsewhere.

What changed since Q2

The frozen panel cited 1,510 distinct domains in Q3, compared with 1,119 in the Q2 benchmark. Mean pairwise domain overlap moved from 0.180 to 0.094, a 47.8% decrease. YouTube remained first while Reddit moved from eleventh to second and G2 moved from twenty-fourth to fifth. These changes are observational. They do not prove a durable market trend because the provider models and indexes also changed. They show why a current, multi-engine baseline matters: an authority map captured in May did not describe the citation surface measured in August.

Which leading domains moved?

The table compares response-level citation share for domains that made both quarters' overall top 25. A positive change means the domain appeared in a larger share of the same 375-response panel.

How did the top-25 source mix change?

This composition view counts domains, not citation occurrences. Vendor first-party sites occupied five of Q3's overall top-25 slots, up from two in Q2. Niche publisher hubs moved from eight slots to four, while social and search-cache domains entered the top 25. The classification is a descriptive annotation of the observed leaders.

AI engines cite pages, not homepages

Only 1.5% of the 3,412 cited URLs were homepages. 74.7% were at least two path segments deep. Teams need page-level citation tracking and crawl evidence to see which specific guide, comparison, documentation page, or discussion earned the answer. Domain-level mention counts alone hide that work.

Overall top 25 cited domains

Each domain is counted once per response, even when an answer cites multiple URLs from that domain. Shares use all 375 responses as the denominator.

Top 10 cited domains by engine

Every engine answered the same 75 prompts. The different leaders below are the practical reason to inspect each engine separately.

How many engines share the same leading sources?

Interpretation limits

This is a fixed buyer-question panel, not a census of every AI answer. Rankings describe the 25 sampled verticals and the provider behavior observed during the collection window. Structural domain categories are a secondary annotation: 1,508 of 3,030 domain-citation occurrences were classified, so category-level conclusions should use that documented subset. The domain rankings, engine overlap, URL depth, and downloadable CSV do not depend on category completeness.

The operational conclusion is narrower and stronger. Citation sources differ materially by engine and change over time. Foglift tracks all five engines, the cited pages behind each answer, and the crawler and referral evidence needed to connect visibility with actual site activity.