Korea AI Pulse
Measured statistics

Four engines, in numbers

Given an identical question, the engines differ completely on whether they search, whether they cite, and whether they name anyone. Aggregated across everything collected from 2026-07-18 to 2026-08-01.

Tracked questions
81
Parsed answers
324
Collections to date
714
Citations captured
1,371

What this data says

Four things fall straight out of the measurements. These figures move as more collections land.

100% vs 72%
One engine searches, another answers from memory

Grok runs a real web search on 100% of its answers; Claude manages only 72%. When an engine doesn't search, only brands already baked into its training data can appear.

47%
Korea's local web belongs to Grok

Grok produces 47% of all citations to Korean local sources — Naver blogs, Instagram, community forums. The other engines stick to official sites and platform profiles, so local review marketing effectively reaches one engine out of four.

5.9
The engine that names the most businesses

Gemini names an average of 5.9 businesses or services per answer. It is the engine where competition for a named slot is fiercest.

75%
The most open category — Law firms

Across the four engines, Law firms questions have the lowest rate of naming anyone at all, at 75%. Nobody has locked this category down yet, which makes it the easiest to enter.

How the engines behave

GPTClaudeGeminiGrok

81 questions × 4 engines, latest collection. The reason GEO strategy has to differ per engine is contained in these six numbers.

Web search rate

Share of answers where the engine ran at least one web search first

GPT
91%
Claude
72%
Gemini
85%
Grok
100%
Citation rate

Share of answers with at least one source link attached

GPT
84%
Claude
72%
Gemini
85%
Grok
100%
Brand-naming rate

Share of answers naming at least one business or service

GPT
83%
Claude
72%
Gemini
100%
Grok
100%
Businesses named per answer

Average number of businesses or services named in a single answer

GPT
3.3
Claude
2.6
Gemini
5.9
Grok
5.9
Citations per answer

Average number of source links attached to a single answer

GPT
3.8
Claude
3.8
Gemini
7.7
Grok
5.1
Average answer length

Average character count of the latest collected answers

GPT
1,194 chars
Claude
595 chars
Gemini
1,769 chars
Grok
1,423 chars
View as a table
EngineSearch rateCitation rateNaming rateNamed / answerCitations / answerAvg length
GPT91%84%83%3.33.81,194
Claude72%72%72%2.63.8595
Gemini85%85%100%5.97.71,769
Grok100%100%100%5.95.11,423

Where the engines will name names

Share of answers naming at least one business, by vertical and engine. Darker means more willing to name — and every pale cell is a slot still open.

VerticalGPTClaudeGeminiGrok
Vet clinics (22)
68%
68%
100%
100%
Law firms (13)
62%
38%
100%
100%
Restaurants (14)
100%
79%
100%
100%
Tech & fintech (13)
100%
85%
100%
100%
Insurance (9)
89%
100%
100%
100%
Fortune telling (5)
80%
60%
100%
100%
GEO & AI marketing (5)
100%
80%
100%
100%

The 15 brands AI recommended most

Ranked by how many times the four engines named them across all verticals. The coloured dots show which engines named each one — a brand hitting all four owns that part of the answer space. Names are left in Korean, as returned.

  1. 1법무법인 대륜Law firms · 12 questions23×
  2. 2카카오페이Tech & fintech · 8 questions16×
  3. 3뱅크샐러드Tech & fintech · 11 questions15×
  4. 4현대해상Insurance · 7 questions12×
  5. 5KB손해보험Insurance · 5 questions12×
  6. 6DB손해보험Insurance · 6 questions9×
  7. 7토스Tech & fintech · 4 questions8×
  8. 8점신Fortune telling · 4 questions8×
  9. 9법무법인 오현Law firms · 6 questions8×
  10. 10보맵Tech & fintech · 3 questions7×
  11. 11삼성화재Insurance · 4 questions7×
  12. 12토스 (비바리퍼블리카)Tech & fintech · 5 questions7×
  13. 13법무법인 YKLaw firms · 6 questions7×
  14. 14예은동물의료센터Vet clinics · 3 questions6×
  15. 15카카오뱅크Tech & fintech · 4 questions6×

What the answers are built on — source types

The 1,371 measured citations, broken down by type. Colour shows which engine produced each citation — the Naver blog and social segments are effectively Grok on its own. Domain-level detail lives on source analysis.

Official site / own domain965
Brunch & Tistory (long-form blogs)137
Naver blogs (Korea's main search portal)77
Restaurant & travel platforms73
Social (Instagram, YouTube)57
News & media20
Legal platforms (Lawtalk etc.)17
GPTClaudeGeminiGrok

Slots still empty — where to get in

Question × engine combinations where no brand has been named yet. A filled dot means that engine has recommended nobody for this question — publish the right content and you take the slot.

Method: search rate, citation rate, citations per answer and average length come from the latest collection per question × engine; naming rate, businesses named per answer and the open slots come from the parsed answer entries. Answers that named nobody are counted as measured, not discarded. Every timestamp and raw answer is in the collection log.