The New Challenge for Publishers in the Age of Generative AI Search
As generative AI search services such as ChatGPT, Perplexity, and Google AI Overviews become part of everyday life, the challenges facing Korean media companies are changing as well. The question is no longer simply, “Does our article appear prominently in search results?” Instead, a new performance metric is emerging: “Does AI cite our publication when generating an answer?”
Through a roughly four-month GEO (Generative Engine Optimization) consulting project conducted by Bluedot Intelligence for Media Company A from March to July 2026, let’s take a closer look at what actually changes AI citation rates.
Where We Started: The Reality of Citation Rates in the 0% Range
At the beginning of the project, the publication’s domain citation rate was 0% for generic queries that did not include its brand name—for example, questions about specific issues or events. Even for queries that did include the brand name, the citation rate was only around 8.8%.
By contrast, brand visibility—the percentage of AI responses that mentioned the publication at all—was already close to 100%. In other words, AI “knew” the publication, but did not cite it.
This is a typical problem faced by many media organizations. They may have plenty of content, but AI crawlers may not be able to access that content properly, or unstructured data may prevent the publication from being recognized as a reliable source.
What Changed: A Series of Technical Improvements
The measures implemented over approximately four months fell into three broad categories.
1. Improving Crawling Accessibility
- Opened access to five major indexing bots, including OAI-Search
- Registered the site with Bing and added Google Newsbot
- Cleaned up 301 redirects for AMP pages
- Updated outdated sitemaps and fixed 404 errors
2. Transitioning to Server-Side Rendering (SSR)
The most significant turning point came in early June, when the site transitioned to server-side rendering (SSR). Following the transition, according to Google Search Console, average daily impressions increased by 2.54x, while clicks increased by 2.01x. The lag between implementation and impact was also remarkably short—less than a day.
3. Enhancing Structured Data (Schema)
Schema markup such as Organization, NewsArticle, and BreadcrumbList was progressively implemented across the corporate information and article pages, followed by a full-scale expansion in mid-June.
The Results: Different Responses Across Channels, but Clear Gains Overall
As of the final measurement in July, the results were as follows:
- Citation rate for branded queries: 8.8% → 62.3% (a steady increase from the beginning)
- Citation rate for non-branded queries: 0% range → 23.3% (a sharp jump following the SSR transition)
- Visibility for non-branded queries: 25.1% → 47.9%
One particularly interesting finding was that the speed of response varied significantly across AI search channels.
One AI search channel that had previously performed particularly poorly showed a structural turnaround after an in-depth diagnostic in June identified wasted crawl budget as a key issue. Once the problem was addressed, its citation rate for branded queries shifted from 0% to 20%.
What the Data Tells Us: Correlation with Traditional KPIs
Perhaps the most significant finding from this project was that GEO performance was not limited to the new metric of AI citation rate. It also showed a statistically significant correlation with traditional search KPIs, including Google Search Console (GSC) impressions and clicks.
An analysis of daily data (n=111 days, based on a 7-day moving average) found that the correlation coefficient between GSC impressions and GEO domain citation rate was approximately 0.80.
This suggests that GEO consulting is not simply a separate effort designed to “improve visibility in AI.” Rather, it can be understood as an initiative that improves the overall technical health of a site across search engines.
The Remaining Challenge: Differences by Content Type
Of course, not everything has been solved.
Of the seven vulnerable non-branded query categories, six successfully moved from 0% to the 10–99% range. However, certain types of queries—particularly those requiring interpretation of historical events—remain at 0% despite the technical improvements.
This suggests that technical improvements such as crawling and structured data need to be accompanied by improvements in the design and presentation of the content itself.
Conclusion
This case highlights three key takeaways.
- AI citation is different from visibility. There can be a significant gap between AI “knowing” a publication exists and actually citing it. Closing that gap is at the heart of GEO.
- Technical improvements can translate into tangible results. SSR migration, schema markup, and improved crawling accessibility are not isolated initiatives; when implemented together, they can create a compounding effect.
- Different channels respond at different speeds. To see the full picture of a GEO consulting project, short-term performance indicators such as GSC metrics and medium- to long-term outcomes such as generative AI search visibility need to be designed and evaluated together.
As AI search increasingly becomes a new gateway for content consumption, the next competitive advantage for media companies and brands may lie not in “how often they are seen,” but in “how often they are cited as a trustworthy source.”
This case study is based on a GEO consulting project conducted by BlueDot Intelligence. The name of the media organization has been withheld.
Bluedot
