In Taiwan, when the AI engine answers brand and shopping-related questions, the first to quote is not your official network, but the Mobile01 discussion string, the Wikipedia article, or the local media coverage. The reasons are straightforward: these sources have third-party endorsements, discussion networks, and the facts can be cross-checked, and models judge that they are more credible than self-soldled product presentations. So doing GEO in Taiwan, the real place to do the work is often outside the official web -- putting the quotes on the source that AI really turns.
AI, which sources do the engines prefer in Taiwan?
On the same question, in Chinese and English, AI gave a different source of reference. Asked "What B2B SaaS is worth considering in Taiwan", the model goes back to the many high-trust sources it has read during its training and inspection: Dcard, PTT, Mobile01, a group of people who have been talking about so much, the Wikipedia article, coverage by Central and local media, and public data from .gov.tw. There's a common denominator in these sources, which you don't control alone. It is because of not being manipulated by you that AI is willing to give them a higher weight, and that is why all the budget is thrown into the official web blog, often without a reference to the AI answer.
- Dcard, PTT: Monument and mine avoidance. When the user asks whether "a product is good or not", the model often uses this kind of reality experience as a reference.
- Mobile01: Compare, open and long-term use, especially hardware, tools and services.
- Wikipedia Chinese version: Definition check and background check of "who's this company" is where the model is built.
- Local news (Central, Economic Daily, Digital Times, etc.): Raising funds, co-operation, winning prizes, news events, etc., take time to verify with third parties.
- .gov.tw: Law, statistics and industry data, models will be quoted to ensure that the facts are correct.
There are three common conditions for what can be quoted.
Before laying anything, understand the logic of the model selection reference. It is not looking for the most beautiful paragraphs, but rather for the best, most credible sentences that can be matched elsewhere. The following three conditions, either of which is not satisfactory, have been cited as a significant decline in probability.
- Fact can be verified: Give the exact number, date, name and location so that the content can be matched by another source. "By 2018, team 40" beats "a highly experienced professional team."
- Language is neutral: reading is like describing, not selling. The model skips the adjective stack of sentences, because it doesn't make up the applicable facts.
- The structure is clean: one thing to say, put the conclusions at the beginning of the sentence. AI takes a period of reading, the cleaner it is, the easier it is to be quoted.
Template for forum post: Dcard, PTT, Mobile01
The forum is the main battleground for Taiwan’s slogans, but it has a doorway: fake accounts and business papers are reversed as soon as they become known, platforms are removed, people are captured, brands are awarded. The content of the forum that can be cited is a detailed sharing of experiences written by real users. What you can do is not put it down yourself, but sort out a real client's echo or internal reality into a good, quoted skeleton that gives people who want to share something to write and makes discussions that happen naturally easier to extract from models.
- Context opening: Who are you, what to solve. For example, "The company is going to get into the customer service system, comparing three to record them."
- Specific: List the rules, price ranges, use of comparable numbers, etc., which is the most popular part of the model.
- Test experience: Tell me what's good, where's good, where's good, where's good, what's good, what's good, what's good, what's good, what's good, what's good?
- Contraception and conclusion: to give you a reason to give out your options, in a clear judgment.
Press release template: for the media to upload and for AI to quote
The press release is worth investing because a report that has been retransmitted by the Central Agency or industry media will be indexed by the search engine at the same time, searched by AI, and will carry the media’s own confidence score. The problem is that most companies write news articles that read like ads, and journalists don't use them, and models don't quote them. Quoted articles go to the bottom of the pyramid: the most important and verifiable facts are in the front, adjectives and aspirations are in the end, and even not.
- Title with fact: Write a verifiable number or event into the title. "XX Finish A Round 80 million" is better than "XX Again."
- First paragraph self-contained: the first paragraph ends with people, events, times, places, and numbers, so AI can be quoted in this paragraph.
- A reference number: Put a representative number, e.g. number of clients, growth rate or market size, and indicate the basis of the calculation.
- A quote from the quote: name, title, point of view, not slogan, and model will make the quote a credible source.
- Corporate Introduction: Three to four sentences refer to the year of foundation, business and scale, which are often used by models to build brand facts.
wikis template: neutral, verifiable, reliable sources
The Wikipedia version is one of the finals of the model's "Who's This Company?" It's the highest and the easiest to step on. The entries are not promotional pages: insufficient attention, language sales, and only home-based news articles, which can be edited or deleted directly, but leave a negative record on the Internet. There is only one way to get in and stay, according to wiki's rules, to think of yourself as the subject of the story, not the author.
Every key fact in the article needs to be able to reach an independent and reliable third-party source. The content that can't be verified will be removed sooner or later, which is exactly the same rule as AI to judge credibility.— The focus of Wikipedia's "proveable" project.
Keep track of the template before you know if it works.
The content is just the first half. You have to know who, whether or not AI quotes you and on which source you are completely absent when asked about your type. And that's exactly what Tenten is doing with Brand Radar: keeping track of the mainstream AI engine's answers to your key questions, marking the sources and gaps cited, so that you know that the next item should be added to the Dcard, Wiki, or the press, instead of spreading it by feeling. If you want to see where your current citation gap is, you can schedule a 30-minute GEO diagnosis, and we'll ask you one round of AI, show you the spot.
