ChatGPT, in reply to a question from Taiwan users, cited a variety of sources that are not as large as expected in the natural search ranking of the site. With Brand Radar, we follow up on a series of B2B and SaaS-related questions, the same key word, Google’s first page is not likely to appear in AI’s answer, but rather a few well-structured and re-written conclusions on the front page. This means that the Taiwan brand wants to be quoted by AI, and the starting point is not to write another blog, but to find out who it is now.
Where did this report come from?
We've picked out a list of complex questions covering software acquisition, B2B services, finance and the four domains of local life, asking ChatGPT in a language that will actually be entered by Taiwan users, recording the reference links and the name of the name mentioned in each response, and then sorting, weighting, marking the source with Brand Radar. The point is not to pursue a beautiful, full-stop sample number, but to see one thing clearly: when the question is in Chinese and the answer is Chinese, the model reaches out to the type of page. The following percentage is an observation of this tracking sentence, which is not a web-based survey, and is intended to give the Taiwan brand a baseline against which to view.
The five most cited sources
Looking at all named sources, the multiple AI answers are highly focused on a few page types. The following grouping is the most common source category in our questionnaire.
- Encyclopedia and Integration: Wikipedia, knowledge integration stations in various domains, answers to definitional and background questions almost invariably appear.
- Posters on the brander network: fixed price pages, functional comparison pages, frequently available question pages, especially when asking "What's the difference between A and B" are clearly high.
- Media and industry coverage: analysis of technology and financial media, providing a third-party perspective for AI.
- Forums and community discussions: PTT, Dcard, and associated communities are cited in their practical use of experience and reputations.
- Structured directories and evaluation stations: tabulations, lists, scoring pages, and "what options are there?"
It's worth noting the second category. Many Taiwanese marketing teams think it's hard for officials to be quoted by AI and place their resources in their blogs, but the numbers are the opposite: as long as the officials' web pages are clear enough to write questions and answers, they are given the opportunity to avoid media coverage. The reason is simple, and the model needs a source that is clear about attribution, stable content, price-fixing pages and FAQ pages that are born to say dead answers.
Official Internet is actually more frequently quoted than you think.
We've tracked a pattern that re-emerges: the same brand, which is often quoted not as the longest article it ever wrote, but rather as the one that has no one to defend. The reason for this is that AI has preferred the paragraph "A question, a sentence, and a stand-alone one." One of the questions is the title, the answer is two or three sentences of FAQ, which can be used directly. This also explains why some small and medium-sized business networks, which are not in high traffic, are quoted by AI over and over again on a particular Leki issue -- they're just so clear.
There are three common features to what has been quoted.
Open the cited pages one by one, and the winnings are not in fancy text, but in three very basic structures. These three points are also the easiest part of Taiwan's brand.
- Preface the answer: The core conclusion is in the first sentence of the paragraph, without laying out, without going around, so that the model can capture the quoted sentence at once.
- Language is self-contained: each paragraph can be understood even if it is taken out alone, and does not depend on the previous paragraph's pronoun or previous connection.
- Determinants of specific information: gives a vague description of a model that produces numbers, versions, dates, precise conditions, rather than "very good" "very good."
AI does not quote you, usually not because you're not good at writing, but because your answer cannot be used in a clean manner alone.
The three most common gaps in Taiwan brands
In view of these quoted pages, we saw the same three gaps over and over again as we were doing the GEO audit for clients. First, there is a shortage of content, many of the B2B brands have an official web host page in English, with only a brief translation in the medium version, and the model naturally does not capture enough sources to answer Taiwan’s questions. Second, the content is all in blogs, and the most easily quoted pages are vaguely written, setting prices, comparing problems and common problems. Thirdly, there is nothing to track the situation of being quoted, and re-edited, new pages are not known if the answer to AI has changed, which is to do in the box. The gap itself is not difficult; the difficulty is that no one tells you where it is.
What we can do in the next three months.
If you want to make the content easier for AI to quote, don't rush the expansion. The first month has been devoted to changing the central pricing page, the comparative page, the frequently seen question page to a pre-requisite, self-informative writing, which is the highest rate of reporting. The next month, the main medium-sized content was completed, making the English-only key pages into a really complex content, not a machine translation. In the third month, we started regular tracking of our presence in the AI answer, and we used data to determine which adjustments worked. This process is exactly what Tenten's CEOs do for clients in 30 days -- measure who you're covered up with now and decide where to fix it. If you want to know the actual gaps in your busy AI search, you can expect a 30-minute GEO diagnosis, and we can run it over and over with your own key word field.
