How Often Does AI Cite
a Brand’s Official Site?
A Study of 1.04 Million Answers
From October 19 to November 15, 2026, Anymorph analyzed 1,042,348 answers from ChatGPT, Gemini, Google AI Overviews, and Naver AI Tab, comparing in Korean and English how often AI cited a brand’s official site as evidence when recommending it, and how that related to being the first recommendation.

Key takeaways
- When AI search mentioned a brand, it cited that brand’s official site as evidence only about one or two times out of ten. This was 18.0% for ChatGPT, 18.1% for Google AI Overviews, 13.9% for Gemini, and 5.2% for Naver AI Tab in Korean answers, and 10.1–12.5% in US English answers. Even when looking only at projects measured in both languages, Korean was higher across all three engines.
- Brands whose official sites were cited as evidence were 1.2–2.0 times more likely to be recommended first. In Korean ChatGPT answers, the figures were 55.5% and 35.1%. The difference remained when comparing the same brands: 44.1% and 29.0%. For eight out of ten brands, the first-recommendation rate was higher when the official site was cited.
- Guides and comparison articles were the most commonly cited official-site pages. These accounted for 69.2% of cited pages in Korean Gemini answers, 48.4% in Google AI Overviews, and 25.8% in ChatGPT. Outside official sites, blogs, review and comparison platforms, and videos were prominent sources in Korean answers; media, retail, and communities were prominent in US English answers.
When people ask AI search for “recommended brands” or “which products are good,” AI mentions brand names and includes the pages it used as evidence alongside them. That evidence may come from the brand’s own site, a blog or review platform, or an article. From October 19 to November 15, 2026, Anymorph asked the questions from 42 measurement projects every day on ChatGPT, Gemini, Google AI Overviews (AI Overviews), and Naver AI Tab, and analyzed 1,042,348 answers. For each of the 1,177 brands mentioned in the answers, we checked whether that same answer cited the brand’s official site and compared the recommendation order of brands whose official sites were cited in Korean and English.
Research findings at a glance
| Item | Korea (Korean) | US (English) | What it means |
|---|---|---|---|
| Percentage of brand mentions citing that brand’s official site: ChatGPT / Gemini / Google AI Overviews | 18.0% / 13.9% / 18.1% | 12.5% / 10.1% / 12.5% | Most evidence comes from outside official sites |
| Same percentage, Naver AI Tab | 5.2% | Not applicable | Naver Shopping product pages are the main source of evidence |
| Same percentage in projects measured in both languages (ChatGPT) | 13.8% | 9.8% | Korean is higher even within the same projects |
| Rate recommended first (ChatGPT): official site cited / not cited | 55.5% / 35.1% | 49.6% / 32.4% | 1.58 times / 1.53 times |
| Same rate, Naver AI Tab | 67.1% / 33.8% | Not applicable | 1.98 times |
| First-recommendation rate compared within the same brands (ChatGPT) | 44.1% / 29.0% | 44.9% / 29.5% | Higher when cited for 81.3% / 80.6% of brands |
| Guides, comparison articles, and blog posts among cited official pages: ChatGPT / Gemini / Google AI Overviews | 25.8% / 69.2% / 48.4% | 47.8% / 57.9% / 48.1% | The most commonly cited page type across all engines |
| Prominent sources outside official sites (Google AI Overviews) | Blogs 25.2%, videos 14.9% | Communities and social media 20.1%, retail 18.1% | Different channels matter in different countries |
Research methodology
Scope
| Item | Details |
|---|---|
| Period | October 19–November 15, 2026 (4 weeks) |
| Engines | ChatGPT (web search enabled), Gemini, Google AI Overviews, Naver AI Tab (Korean only) |
| Country and language | Korea (Korean), US (English) |
| Measurement projects | 42. Primarily commerce projects, including software, education, and healthcare services |
| Brands | The project’s own and competitor brands tracked by each measurement project that had a registered official domain. A total of 1,177 brands appeared in the answers |
| Answers | 1,042,348 (499,159 in Korean, 543,189 in English). Counted only answers that mentioned at least one tracked brand and included citations |
| Brand mentions | 2,811,717 (1,271,541 in Korean, 1,540,176 in English). If an answer mentioned three brands, that counted as three mentions |
| Citations | 10,165,503 |
All figures combine multiple measurement projects; results for any specific company are not included. The data consists of AI search answers collected by Anymorph.
Measurement methods
- Official-site citation: For each brand, we counted a mention as “official site cited” if its registered official domain (including subdomains) appeared among the citations in the same answer. Official accounts and stores on Naver, social media, and blog services were not counted as official sites.
- Citations: For ChatGPT, we counted citations in the answer text. For Gemini, Google AI Overviews, and Naver AI Tab, we counted the source lists attached to answers.
- Recommendation order: The order in which tracked brands were first mentioned in an answer. We treated the first brand mentioned as the “brand recommended first.”
- Same-brand comparison: To reduce differences in brand recognition, we selected only brands with at least 5 mentions each with and without an official-site citation over the 4 weeks. We calculated the first-recommendation rate for both cases for each brand, then averaged the rates.
- Source types: We categorized cited domains as brand and company sites, blogs, review and comparison platforms, videos, communities and social media, media and news, retail and commerce, and other types. We categorized the most frequently cited domains for each engine and language first.
- Official page types: We categorized 192,895 official-site citations with verified URLs from November 2–15 by URL path. Categories included guides, comparison articles and blog posts; product and service details; homepages; customer support, FAQs and documentation; pricing and plans; and company and brand information.
Only one or two out of ten brand mentions cited that brand’s official site as evidence
| Engine | Korea (Korean) | US (English) |
|---|---|---|
| ChatGPT | 18.0% | 12.5% |
| Gemini | 13.9% | 10.1% |
| Google AI Overviews | 18.1% | 12.5% |
| Naver AI Tab | 5.2% | Not applicable |
Among answers in which AI search mentioned a brand, the official site was also cited in just over 18% of Korean ChatGPT and Google AI Overviews answers, and 13.9% of Gemini answers. In US English answers, the figure ranged from 10.1–12.5% across all three engines. The remaining 80% or more of mentions relied on pages from elsewhere, such as blogs, review platforms, and articles.
Over the 4 weeks, official sites were cited as evidence at least once for 74.3% of brands in Korean ChatGPT, 74.5% in Google AI Overviews, and 52.8% in Gemini. Most brands had an opportunity to have their official site cited, but it was not cited every time they were mentioned.
Korean rates were higher even when comparing the same projects
| Engine | Measurement projects | Korea (Korean) | US (English) |
|---|---|---|---|
| ChatGPT | 8 | 13.8% | 9.8% |
| Gemini | 7 | 8.0% | 7.1% |
| Google AI Overviews | 7 | 11.8% | 10.9% |
The Korean and English projects differed in their industry mix. To reduce this difference, we looked separately at projects that measured the same brands in both Korean and English. The gap narrowed, but official sites were still cited more often in Korean answers across all three engines.
Brands with cited official sites were 1.2 to 2.0 times more likely to be recommended first
| Engine | Language | Official site cited | Official site not cited | Multiple |
|---|---|---|---|---|
| ChatGPT | Korean | 55.5% | 35.1% | 1.58 times |
| ChatGPT | English | 49.6% | 32.4% | 1.53 times |
| Gemini | Korean | 52.7% | 30.3% | 1.74 times |
| Gemini | English | 48.8% | 30.0% | 1.63 times |
| Google AI Overviews | Korean | 52.2% | 35.4% | 1.47 times |
| Google AI Overviews | English | 46.9% | 37.9% | 1.24 times |
| Naver AI Tab | Korean | 67.1% | 33.8% | 1.98 times |
Across every engine and language, brands whose official sites were cited as evidence appeared first in answers more often. In Korean ChatGPT answers, 55.5% of brands with an official-site citation were mentioned first, compared with 35.1% of brands without one. Their average recommendation order was 1.76 and 2.27, respectively. Official sites were rarely cited in Naver AI Tab, but when they were, the brand appeared first 67.1% of the time—the highest rate.
The difference went in the same direction in all four weeks. Looking at Korean ChatGPT by week, the first-recommendation rate was 53.7–58.0% when an official site was cited, compared with 33.6–37.1% when it was not.
The difference remained when comparing the same brands
| Engine | Language | Brands compared | Official site cited | Not cited | Brands with higher rate when cited |
|---|---|---|---|---|---|
| ChatGPT | Korean | 444 | 44.1% | 29.0% | 81.3% |
| ChatGPT | English | 376 | 44.9% | 29.5% | 80.6% |
| Gemini | Korean | 238 | 41.4% | 26.4% | 79.0% |
| Gemini | English | 335 | 43.4% | 27.5% | 82.7% |
| Google AI Overviews | Korean | 377 | 38.8% | 30.0% | 72.4% |
| Google AI Overviews | English | 351 | 43.8% | 34.1% | 78.3% |
| Naver AI Tab | Korean | 65 | 53.5% | 36.2% | 80.0% |
Well-known brands may have their official sites cited more often and also be recommended first more often. So we compared mentions with and without an official-site citation within the same brand. In Korean ChatGPT answers, the average first-recommendation rate across 444 brands was 44.1% when the official site was cited, compared with 29.0% when it was not. Depending on the engine and language, the first-recommendation rate was higher when the official site was cited for 72.4–82.7% of brands.
The pattern also held in answers mentioning multiple brands
| Engine | Korean: cited / not cited | English: cited / not cited |
|---|---|---|
| ChatGPT | 40.3% / 24.2% | 39.2% / 24.7% |
| Gemini | 37.4% / 24.1% | 39.4% / 23.9% |
| Google AI Overviews | 37.8% / 24.4% | 33.7% / 25.8% |
| Naver AI Tab | 36.3% / 23.0% | Not applicable |
In an answer discussing only one brand, that brand is necessarily first, and it is also easy to cite its site. Even when we looked only at answers mentioning at least three tracked brands, the first-recommendation rate for brands with an official-site citation was 7.9–16.1 percentage points higher than for brands without one.
Guides and comparison articles were the most commonly cited official pages
| Official-site page type | ChatGPT Korean | Gemini Korean | Google AI Overviews Korean | ChatGPT English | Gemini English | Google AI Overviews English |
|---|---|---|---|---|---|---|
| Guides, comparison articles, and blog posts | 25.8% | 69.2% | 48.4% | 47.8% | 57.9% | 48.1% |
| Product and service details | 18.9% | 11.1% | 20.8% | 23.7% | 30.1% | 29.0% |
| Homepage | 14.2% | 2.5% | 6.2% | 1.8% | 1.0% | 0.4% |
| Customer support, FAQs, and documentation | 12.2% | 3.3% | 3.4% | 5.9% | 0.7% | 1.3% |
| Pricing and plans | 0.6% | 0.1% | 0.1% | 6.0% | 0.9% | 3.6% |
| Company and brand information | 3.8% | 0.9% | 2.3% | 1.3% | 0.2% | 0.3% |
| Other pages | 24.5% | 12.9% | 18.8% | 13.5% | 9.2% | 17.3% |
When AI search used official sites as evidence, the pages it cited most often were articles about how to choose, comparisons, and recommendations. Such articles made up 69.2% of cited official pages in Korean Gemini answers and 48.4% in Google AI Overviews; in English answers, the figure ranged from 47.8–57.9% across all three engines. Even when we excluded the one project where such articles were cited most often, the rates were 63.7% for Korean Gemini, 49.0% for Google AI Overviews, and 39.9–43.4% for English answers. Product and service detail pages were the next most common.
Korean ChatGPT differed from the other engines. After guides and comparison articles (25.8%) and product and service details (18.9%), it frequently used homepages (14.2%) and customer support, FAQs, and documentation (12.2%) as evidence. In English ChatGPT, homepages accounted for 1.8%, customer support, FAQs, and documentation for 5.9%, and pricing and plans pages for 6.0%.
Sources outside official sites differed between Korea and the US
| Source type | ChatGPT Korean | Gemini Korean | Google AI Overviews Korean | ChatGPT English | Gemini English | Google AI Overviews English |
|---|---|---|---|---|---|---|
| Brand and company sites | 33.1% | 25.9% | 18.5% | 17.8% | 20.0% | 13.9% |
| Blogs | 4.0% | 15.9% | 25.2% | 0.9% | 2.5% | 0.7% |
| Review and comparison platforms | 9.6% | 15.6% | 5.6% | 3.6% | 1.6% | 0.6% |
| Videos | 0.1% | 3.2% | 14.9% | 0.0% | 3.0% | 11.4% |
| Communities and social media | 0.4% | 2.3% | 7.5% | 0.2% | 3.2% | 20.1% |
| Media and news | 6.4% | 7.2% | 4.5% | 22.7% | 14.6% | 8.0% |
| Retail and commerce | 7.1% | 8.5% | 7.9% | 19.6% | 20.2% | 18.1% |
| Other sites | 39.3% | 21.4% | 15.9% | 35.2% | 34.9% | 27.2% |
We categorized all citations attached to answers mentioning a brand by source type. Brand and company sites include both the official sites of brands recommended in the answer and other company sites.
In Korean answers, blogs, review and comparison platforms, and videos were major sources outside official sites. For Google AI Overviews, 25.2% of citations were from blogs and 14.9% from videos. For Gemini, 15.9% were from blogs and 15.6% from review and comparison platforms. For ChatGPT, brand and company sites had the largest share at 33.1%, followed by review and comparison platforms at 9.6%.
In US English answers, media, retail, and communities had a large share. For ChatGPT, media and news accounted for 22.7% and retail and commerce for 19.6%; for Google AI Overviews, communities and social media such as Reddit accounted for 20.1%. Blogs accounted for less than 3% in all three engines’ US English answers.
Naver Shopping product pages were the main source of evidence in Naver AI Tab
| Source type | Share of Naver AI Tab citations |
|---|---|
| Retail and commerce | 73.6% |
| Of which: Naver Shopping product pages | 49.8% |
| Maps and places | 6.5% |
| Blogs | 5.0% |
| Media and news | 3.8% |
| Communities and social media | 3.2% |
| Brand and company sites | 1.2% |
In Naver AI Tab answers that mentioned a brand, 73.6% of citations were from retail and commerce, and half of all citations (49.8%) were from Naver Shopping product pages. Brand and company sites accounted for 1.2%. Over the 4 weeks, official sites were cited at least once for only 29.2% of brands, less than half the rate for ChatGPT (74.3%). In Naver AI Tab, product information on Naver Shopping was used as evidence for brands more often than official sites.
How to take action
Publish guides and comparison articles that answer recommendation questions on your official site
Brands whose official sites were cited as evidence were 1.2–2.0 times more likely to be recommended first, and guides and comparison articles were the most commonly cited official pages.
- Publish articles on your official site with titles based on the questions customers ask AI, such as “how to choose,” “comparison,” and “recommendations.”
- Include selection criteria, comparison tables, and recommendations for different use cases so AI can use them directly as evidence.
- Cover the entire category fairly rather than discussing only your own products.
Fill product and service detail pages with clear text
Product and service detail pages were the second most commonly cited official page type across all engines (11.1–30.1%).
- Present specifications, prices, uses, and target customers in text and tables, not just images.
- Add answers to frequently asked questions below the detail-page content.
For Korean ChatGPT, pay attention to your homepage and FAQs too
In Korean ChatGPT, 14.2% of cited official pages were homepages and 12.2% were customer support, FAQs, and documentation. These rates were about 8 times and 2 times higher, respectively, than in English ChatGPT.
- State in one sentence at the top of your homepage what your brand sells and to whom.
- Gather pre-purchase questions about shipping, refunds, pricing, and how to use the product on an FAQ page.
Choose channels outside your official site based on the country and engine
More than 80% of brand mentions appeared without that brand’s official site, and the sources used as evidence differed by country.
- For Korean Google AI Overviews and Gemini, focus on blog reviews and comparison articles, information on review and comparison platforms, and video reviews.
- For US English answers, focus on media articles, retail product pages, and mentions in communities such as Reddit.
- For Naver AI Tab, organize product names, attributes, and reviews on Naver Shopping, and build up reviews on Naver Blogs as well.
Measure official-site citations and first recommendations for each brand
Even for the same brand, its official site was cited on some days and not on others, and the first-recommendation rate was higher when it was cited.
- Track daily, for each engine, how often the official site is cited as evidence when your brand and competitor brands are mentioned.
- After publishing a new guide or comparison article, check for several weeks whether the citation rate and the rate of being recommended first rise together.
Limitations
- This is a 4-week observation from October 19–November 15, 2026. The official-site citation rate for Korean ChatGPT varied from 13.7–22.9% by week. The difference in first-recommendation rates went in the same direction in all four weeks.
- The relationship between official-site citations and being recommended first is an observed distribution. Comparing the same brands and answers mentioning multiple brands helped reduce differences in brand recognition and answer format, but the cause was not verified through an experiment. When an answer focused on one brand, its site may also have been cited.
- The brands are the own and competitor brands tracked by the measurement projects, and the recommendation order is limited to those brands. Brands that were not tracked were not included in the order.
- An official site means a registered official domain and its subdomains. Naver Smart Store, Brand Store, and official social media accounts were not counted, so the official-site citation rate for Naver AI Tab is low under this definition.
- Source types were categorized by domain, starting with the most frequently cited domains for each engine and language. Domains that could not be categorized were placed under other sites.
- Official page types were categorized by URL path. Naver AI Tab was excluded from the page-type comparison because its official-page citations were concentrated in one project.
- The figures are based on answers collected under Anymorph’s measurement conditions (with web search enabled). Answers may differ depending on user account history and settings.
- The questions were asked by Anymorph for its measurement projects, which primarily cover commerce. Differences in industry mix were reduced by comparing projects measured in both languages.
Conclusion
- AI search does not often use a brand’s own site as evidence when recommending that brand. Official sites were cited in 5.2–18.1% of Korean brand mentions and 10.1–12.5% of English mentions.
- Brands whose official sites were cited as evidence were recommended earlier. Their first-recommendation rate was 1.2–2.0 times higher, and the difference remained when comparing the same brands.
- Guides and comparison articles were the most commonly cited official pages. Outside official sites, blogs, review and comparison platforms, and videos were evidence sources in Korea; media, retail, and communities in the US; and Naver Shopping in Naver AI Tab.
If you want to be recommended near the top in AI search, start by creating official pages that AI can use as evidence when it mentions your brand. Publish guides and comparison articles that answer customers’ questions on your official site, make use of the external channels that matter in each country and engine, and measure whether your official site is actually cited for each brand.
FAQS
Frequently asked questions
Does AI search cite a brand’s official site when recommending it?
About one or two times out of ten. In Anymorph’s measurements from October 19–November 15, 2026, the share of brand mentions where that brand’s official site was cited in the same answer was 18.0% for Korean ChatGPT, 18.1% for Google AI Overviews, 13.9% for Gemini, and 5.2% for Naver AI Tab. The rate in US English answers was 10.1–12.5%.
If an official site is cited, does AI recommend the brand higher?
That was the pattern we observed. Brands whose official sites were cited as evidence were 1.2–2.0 times more likely to be recommended first. In Korean ChatGPT, the rates were 55.5% and 35.1%; the difference remained when comparing the same brands, at 44.1% and 29.0%. Since this was not verified through an experiment, measure whether the first-recommendation rate actually rises after changing a page.
Which official-site pages are cited in AI search?
Guides, comparison articles, and blog posts were the most common. Among cited official pages, these made up 69.2% in Korean Gemini, 48.4% in Google AI Overviews, 25.8% in ChatGPT, and 47.8–57.9% in English answers. Product and service detail pages came next. Korean ChatGPT also often cited homepages (14.2%) and customer support, FAQs, and documentation (12.2%).
What does Naver AI Tab use as evidence when recommending brands?
Naver Shopping product pages are the main source. Of the citations in Naver AI Tab answers mentioning brands, 73.6% were from retail and commerce, and 49.8% were from Naver Shopping product pages. Brand and company sites accounted for 1.2%. If you want to appear in Naver AI Tab, start by organizing product information and reviews on Naver Shopping.
How do sources used as evidence in AI search differ between Korea and the US?
Korean answers had a larger share of blogs, review and comparison platforms, and videos, while US English answers had a larger share of media, retail, and communities. For example, Google AI Overviews citations included 25.2% from blogs and 14.9% from videos in Korean, and 20.1% from communities and social media and 18.1% from retail and commerce in English.
How do you measure official-site citations?
For each answer mentioning your brand or a competitor, check whether each brand’s official domain is included in the citations. Since the same brand may or may not be cited on different days, measure daily for each engine and track both the official-site citation rate and the rate of being recommended first.
Appendix: Research data
Study overview
The study was conducted by Anymorph, and the measurement period was October 19–November 15, 2026. We asked 19,384 questions from 42 measurement projects every day on ChatGPT, Gemini, and Google AI Overviews, and also asked some of the Korean questions on Naver AI Tab. We analyzed 1,042,348 answers that mentioned at least one tracked brand and included citations, 2,811,717 brand mentions, and 10,165,503 citations.
Samples by engine
| Engine | Language | Measurement projects | Questions | Answers | Brand mentions | Citations analyzed |
|---|---|---|---|---|---|---|
| ChatGPT | Korean | 28 | 9,477 | 166,825 | 409,769 | 657,211 (citations in answer text) |
| ChatGPT | English | 22 | 9,196 | 182,646 | 526,174 | 762,365 (citations in answer text) |
| Gemini | Korean | 26 | 8,004 | 126,527 | 353,321 | 1,517,229 |
| Gemini | English | 20 | 8,635 | 171,567 | 531,402 | 1,822,625 |
| Google AI Overviews | Korean | 26 | 9,066 | 154,111 | 376,652 | 1,083,975 |
| Google AI Overviews | English | 20 | 8,939 | 188,976 | 482,600 | 1,192,927 |
| Naver AI Tab | Korean | 15 | 3,035 | 51,696 | 131,799 | 3,129,171 |
Sample for official page types
Citations with verified URLs among official-site citations from November 2–15. ChatGPT Korean: 24,645; English: 25,141; Gemini Korean: 54,443; English: 41,093; Google AI Overviews Korean: 29,766; English: 17,807.
Related reading



Find out what AI cites as evidence for your brand with Anymorph
Measure every day whether ChatGPT, Gemini, Google AI Overviews, and Naver AI Tab cite official sites when recommending your brand and competitors, and how highly they recommend them, then decide which pages to create first.