Most teams treat the URL as a routing setting, but the AI engine reads the URL as a paragraph of text and uses it to determine what the page is about and whether it is worth citing. Traditional Chinese websites have one more unclarified decision in this matter: whether the slug should be in Chinese or English. This article separates the path and slug planning, and you can modify them directly.
How does the AI engine "read" your URL?
LLM's search system will break the URL into words to understand, and display it when generating the answer and marking the source. A URL like /blog/url-structure-llm-friendly is equivalent to directly telling the model "this page is talking about URL structure"; if it is replaced with /p?id=48213, the model will not get any clues and can only rely entirely on the content of the page to guess. When several sources have similar conditions, readable and relevant URLs are often more likely to be selected for citation, because it reduces the interpretation cost of both the model and the user.
The path hierarchy itself also conveys context. /geo-audit/pricing makes it obvious at a glance that this is the pricing page for GEO auditing, and the model also reads which topic it belongs to. There is another price for going too deep: the farther a page is from the homepage and the fewer internal links pointing to it, the more likely crawlers will miss it, and the AI engine will naturally not be able to retrieve content that you haven't discovered at all. Readability and crawlability are two different things, but they are both supported by the URL structure.
The dilemma of Traditional Chinese slug: Chinese characters will be encoded into a string of garbled characters
As long as you put Chinese characters in the URL, the browser and server will percent-encode it. When the /blog/url structure plan is actually stored, shared, and referenced, it will become a long string like /%E9%83%A8%E8%90%BD%E6%A0%BC/…. What you see in the address bar is Chinese, but what is sent out is garbled code - AI summaries, community cards, and background reports all receive this string of codes, and all the semantic benefits originally intended are lost. To make matters more troubling, some third-party tools and AI crawlers do not handle such encoded URLs consistently, and occasionally truncate or fail to parse.
- Chinese slug (such as /blog/url structure planning): It looks clear in the address bar, but when transmitted and quoted, it will become the percentage encoding of %E9%83%A8..., and the AI summary, sharing links and reports are difficult to identify.
- English or romanized slug (e.g. /blog/url-structure-llm-friendly): Stable across systems, readable, and cleanly quoted by AI, it is the practical first choice for B2B SaaS websites.
- Use traditional Chinese for the title text and simplified English for the slug: currently the most stable combination for AI retrieval. It is not necessary to translate the entire title of the slug, just two or three English words that capture the core concept.
It should be emphasized that there is no shortage of places where Traditional Chinese should be used: the page title, H1, and content are still in Traditional Chinese, and that is the main body of being ranked and read. The slug is the underlying infrastructure, just grasp the core concepts. URLs are also always in lowercase. Some servers treat /Blog and /blog as different pages. Mixing uppercase and lowercase letters will silently create duplicate URLs. If your Chinese slug has been indexed for a long time, don’t rush to cut it all. Be sure to use 301 to take over the old URL when revising.
AI-friendly path planning: principles that can be followed directly
- The shallower the better: try to keep the main content within two to three layers (/blog/<slug>). Each additional layer increases the risk of being missed and crawled.
- The slug should be semantic: use words that describe the content, not serial numbers or dates. /blog/url-structure-llm-friendly is better than /blog/2026/07/post-482.
- Consistent across the entire site: Use the same path style for the same type of content, allowing crawlers and models to predict your structure.
- Lowercase with hyphens: All lowercase, use "-" between words instead of the underscore "_" to avoid duplicate URLs caused by capitalization or symbols.
- Don’t hide the main content behind query parameters: URLs such as ?id= and ?p= are more difficult to stably index and reference.
- Once the URL is public, don’t change it randomly: if you really want to change it, you must use a 301 redirect to transfer the accumulated weight and citations of the old URL.
Use path structure to match your content clusters
Reflecting the relationship between pillars, clusters, and subtopics into the internal links of the website can help the AI engine determine which topic you have accumulated depth in: Articles in the same pillar are interconnected with each other and point back to the pillar page. The model can see that this area contains systematic content rather than scattered individual articles. The two-way link between the cluster page and the sub-topic article also makes it easier for the model to read the entire topic along the link instead of just grabbing an isolated page. If internal links are well done, whether the URL should be layered according to clusters is a secondary issue.
Multilingualism in traditional Chinese websites: language prefixes and hreflang
If you want to serve multiple language families like many B2B websites, the language prefix should be placed at the beginning of the path, such as /zh-TW/blog/… and /en/blog/… to make the language family identifiable at a glance and facilitate crawler partition processing. Each language version must be marked with each other with hreflang and specify an x-default as the default. Only then will the AI engine and search engine serve the right language version to the right person, instead of treating them as duplicate content and diluting each other.
- The language prefix is placed at the beginning of the path: /zh-TW/blog/..., /en/blog/..., the language family is identifiable at a glance, and it is also convenient for crawler partitioning.
- Each language version is marked with hreflang and specifies an x-default as the default (Traditional Chinese sites often use the English version or zh-Hant as the default).
- Each language family should try to use the same slug: the content is translated, not the path, so that the mapping relationship is clean and will not be regarded as repeated content.
The easiest minefields to step on
The most common mistakes: writing the release date into the path, the URL will look outdated after half a year; using the database serial number as a slug, the model cannot read the topic at all; changing the slug without setting a redirect when changing the title, which is equivalent to dumping the accumulated references at once; the same page has versions with and without slashes, with www and without www, allowing crawlers to count one page as several pages. These are not difficult to fix, but leaving them alone will continue to dilute your chances of being cited by the AI.
URL structure is the kind of underlying technology that gets you right the first time and reaps the benefits in the long run. You don’t have to rewrite the entire site to please AI. In most cases, as long as you semanticize the slug, make the path shallower, mark the multilingual hreflang, and use 301 to close the old URL, the gap will be widened. If you want to know where your site's specific gaps in crawlability and citability are, Tenten GEO's 30-day GEO audit takes a page-by-page inventory of paths, structured data, and search performance. If you want to get the direction first, you can also go to /contact to make an appointment for a 30-minute GEO diagnosis. We will talk directly about your website structure.
