On the subject of finance and medical care such as YMYL, the AI engine chooses a different standard of answer than any other domain: it would have preferred to be less quoted, rather than a vague source. The same paragraph, if placed on a travel blog enough to be taken from Perplexity, could be crossed to the question of whether "a blood sugar drop is safe". The difference is not the quality of the text, but whether the source can be identified by the machine as a "person or institution qualified to answer this matter". This is the real battleground in the AI era.
Why is the financial and medical content, the AI engine, particularly hard to trust?
YMYL refers to topics that can hurt the reader's package or body if the answer is wrong: investments, taxes, loans, drug interactions, disease interpretation, chronic care. Google has long applied a more stringent quality door to this type of inquiry, and the generative AI engine has pushed the logic even further. Because AI gives the answer directly to the user, there's no page in the middle that compares the results, and the sense of responsibility falls on the engine that provides the answer. The model was then adapted to the back end: in the case of YMYL queries, first-hand capture rights, professional endorsements, and source of retroactive origin.
That creates a difficult reality for brands. No matter how polished or comprehensive the content is, if an engine cannot identify who wrote it and why that person is qualified, it will fall back on sources such as Taiwan's Ministry of Health and Welfare, Financial Supervisory Commission, major medical centers, or established financial media. Many Taiwan SaaS and service brands appear in AI answers on general topics, then disappear at the YMYL boundary. The missing piece is often a machine-readable trust structure, not more content.
Four letters from E-E-A-T, each representing what they represent when they are drawn from AI
E-E-A-T is Experion (Experience), Expertise (Professional), Authoritaties (Power), Trustworth (Credibility). Human censors will read together, but the AI engine is the approximation of these four items by signalling. Take it apart so you know which piece you're missing:
- Experience: Are there any first-hand marks? The actual number of cases, operational intercepts, clinic or investment scenes are more proof than generalizations that the author did it himself.
- Professional (Expertise): Does the author's qualifications match the subject matter? People who write about drug safety have a chemist or doctor's background, and people who write about taxes are accountants, which needs to be clearly marked on the page, not hidden about us.
- Authority: Who quoted this person or brand and who recognized it? External references, co-namings and professional institutional links are external evidence of authority.
- Trustworthy: Can't the page itself be reliable? This is one of the four areas where AI is the easiest to quantify and the least to miss.
Author authority is not a name: an identity signal that can be read by a machine.
A lot of brands think there's a line at the bottom of the article, "This is written by a professional team." For AI, this is equivalent to having no author. The engine wants a cross-referenced entity: a named person, with a dedicated author's page, with a clear history and field of expertise on the page, as well as a description of the author and the agency through the structured data. When the name appears on the medical history books, journal titles, speakers' lists or LinkedIn professional files, the model is able to connect scattered signals to a "trustworthy person".
Quote and Source: Make AI willing to use you as a source
Generating engines prefer the content of "they will also invoke power." An article on the side effects of drugs, if every key comment goes back to a copy, a clinical guide or a director's bulletin, is read by an author who knows the origin, behaviour pattern and reliable source. In turn, the pages that are missing from the whole story, even if they are correct, are classified as more risky sources. Here's an easy point to ignore: the source is to be next to the commentary, not to focus on the end of the text, because AI was drawn by paragraph, and it needed to see the basis when it was removed.
In the YMYL domain, the AI engine is not looking for the most beautiful answers, but for a source with the lowest probability of error and the clearest responsibility.
Three fatal gaps common in Taiwan YMYL brands.
When we did the GEO audit for financial and medical clients, we saw the same questions repeatedly, which were not so serious as to add up to the fact that the brand was missing in the AI answer:
- Content is anonymous: the site is almost unnamed, professional experience is unverified and the engine cannot create two E-A signals.
- Externality is inadequate: all references are to home-made pages, and there is a lack of power links to competent institutions, schools or first-hand research.
- The update and legal compliance signal is missing: there is no clear update date on the page, no medical or investment exoneration statement, no identifiable legal compliance information and low credibility.
These three gaps need not be rewritten. From the top 10 to 20 YMYL pages on traffic and business, each supplement will be written by a well-known author and author, the key comments will be put back to power, with updated dates and the necessary exemption and compliance sections. When the same page is adjusted, the citation rate rises first on the engine of the Perplexity and Google AI overviews of the source of the label.
From the beginning of the diagnosis, fill the trust gap.
YMYL's E-E-A-T is not a one-time addition, but a trust-based foundation that continues to be maintained: the author needs to be real, to be out of the house, to be endorsed by the agency. If you're not sure which piece of financial or medical content you're missing in the eyes of the AI engine, the fastest way is to ask ChatGPT, Perplexity and Google about who it quotes, whether or not you are. If you want a list of specific gaps to the page level, you can set aside 30 minutes for a GEO diagnostic session, and we'll run it through your actual query, indicating which trust signals should be added.
