Which sites does an AI actually read?
Every source GPT, Claude, Gemini and Grok attached to an answer, aggregated by domain and by type. If you want to be visible in AI answers, these are the sites your content has to be on.
- Total citations
- 1,371
- Domains cited
- 566
- Answers analysed
- 324
- Date range
- 2026-07-18 – 2026-08-01
How each engine cites
Given the same question, the engines reach for markedly different kinds of source — which is exactly why each one needs its own GEO strategy.
Searches the Bing index. Leans heavily on travel and platform-style sites.
- Official site / own domain135 · 64%
- Naver blogs (Korea's main search portal)28 · 13%
- Restaurant & travel platforms25 · 12%
- Legal platforms (Lawtalk etc.)13 · 6%
- News & media3 · 1%
Runs on Brave Search. Concentrates almost entirely on businesses' own websites.
- Official site / own domain176 · 85%
- Restaurant & travel platforms21 · 10%
- Brunch & Tistory (long-form blogs)6 · 3%
- Wikis & encyclopaedias3 · 1%
- Danggeun & neighbourhood forums2 · 1%
Google grounding. In our API measurements it often skips searching altogether; the consumer Gemini app tends to ground more aggressively.
- Official site / own domain429 · 71%
- Brunch & Tistory (long-form blogs)99 · 16%
- Social (Instagram, YouTube)39 · 6%
- News & media12 · 2%
- Restaurant & travel platforms9 · 1%
Its own search. The only engine that cites Korean local sources — Naver blogs, Instagram, Danggeun.
- Official site / own domain225 · 64%
- Naver blogs (Korea's main search portal)49 · 14%
- Brunch & Tistory (long-form blogs)30 · 9%
- Restaurant & travel platforms18 · 5%
- Social (Instagram, YouTube)18 · 5%
Cited domains, ranked
Ranked by actual citation count. Pay attention to how often individual businesses’ own websites appear — a single keyword-specific page getting cited repeatedly across engines is a pattern that recurs.
How to read this · methodology
- · Based on the most recent collection per question and engine. Duplicate URLs within a single answer count once.
- · Only source links the engine attached to its own answer are counted. Answers produced from training data without a search carry no citations — this is common in our Gemini API measurements, and the consumer Gemini app grounds more aggressively, so its real citation rate is likely higher than shown here.
- · Gemini grounding redirect URLs are resolved back to the originating domain before counting.
- · The full raw text sits in the collection log, answer by answer.
The route into an AI answer isn’t abstract SEO — it is the domain list above, literally. Your name and your evidence need to exist on the sites the engines cite in your category. Two channels come up again and again: a keyword-specific page on your own website and the density of Naver blog reviews about you. If you want this analysed for a specific brand, ask on contact — the report is free.