Bottom line in one sentence: Different AI engines have different source preferences—Tencent Yuanbao favors WeChat Official Accounts, DeepSeek relies on the open web, and Quark favors official and academically authoritative sources. If you want a particular engine to cite you, publish content in the sources it prefers; do not expect one piece of content to work everywhere.
China now has 602 million generative AI users, with a penetration rate of 42.8% (CNNIC’s 57th report, 2026-02). AI-native apps have 446 million monthly active users, including 345 million for Doubao, 166 million for Qwen, and 127 million for DeepSeek (QuestMobile 2026Q1). More and more people are getting used to asking AI directly instead of browsing search results. That raises a new question: when someone asks AI, “Which tool of this kind is best?”, will it mention you?
The prerequisite for being mentioned by AI is getting your content into its source pool. And each engine draws its source pool from different places.
Why the Engines Differ
The difference lies not in the model itself, but in the web-retrieval layer.
When AI answers a real-time question, it usually retrieves a set of webpages or pieces of content first, then organizes an answer based on those results. What it retrieves and which sources it trusts first depend on the search capabilities and content sources connected behind it. Each company uses a different underlying source set, so their preferences naturally differ.
That is why there is no single answer to “How do I get cited by AI?” You have to examine each engine separately.
Engine-by-Engine Breakdown
Tencent Yuanbao: WeChat Official Accounts Are Its Home Turf
Yuanbao is currently the only AI assistant that can search WeChat Official Account content online. It is connected to Weixin Search and Sogou Search, and prioritizes Tencent sources such as WeChat Official Accounts and Tencent News. In 2025-02, it integrated DeepSeek-R1 and incorporated WeChat Official Account sources.
If you want Yuanbao to cite you, here is what to do: turn your core content into WeChat Official Account articles, and write titles that state the question and conclusion clearly so Weixin Search can find and extract them. WeChat Official Accounts are Yuanbao’s moat, and they are also a place most other engines cannot access.
DeepSeek: Your Content Must Be Crawlable on the Open Web
The online search in DeepSeek’s official app is powered by the Search API from Bocha, a Chinese startup. (The widely repeated claim that “more than 60% of Chinese AI applications use Bocha” comes from Bocha’s own promotional material, and the original wording refers to a share of online-search *requests* rather than a share of applications. We could not verify it independently, so we do not rely on it here.)
Bocha crawls the open web. If you want DeepSeek to cite you, here is what to do: put your content somewhere publicly crawlable—technical pages on your official website, product documentation, or open industry articles. Do not lock it behind a login wall or inside an app. And do not use robots rules to block every crawler (more on that later).
Quark: Official Authoritative Sources + Academic Databases
Quark is made by Alibaba and built on the Qwen large language model. In 2025-03, it was upgraded into Alibaba’s flagship AI application; Alibaba’s own announcement of 2025-03-13 put the figure at “over 200 million users,” which is a cumulative-user measure, while third-party trackers put monthly active users at roughly 148 million in the same period—the two are not the same thing. Its deep search favors three types of sources: official authoritative sources, academic databases (it partners with CNKI, Wanfang Data, and VIP), and content communities.
If you want Quark to cite you, here is what to do: align your content with authoritative positions wherever possible and cite standards, white papers, and industry reports. Where feasible, turn your methods or data into a form that academic databases can index. Self-promotional marketing pages carry little weight with Quark.
Doubao: Industry Observations Point to ByteDance Sources (Test This Yourself)
This point needs to be stated cautiously. GEO practitioners commonly observe that Doubao’s retrieval layer favors content from the ByteDance ecosystem, such as Douyin Baike, Toutiao, and verified Douyin Blue V accounts. But this comes from observations across multiple sources and claims by service providers; we have not seen an official position from ByteDance. You should verify it through your own testing rather than treat it as a conclusion.
If testing confirms that it holds in your industry, here is what to do: add content on ByteDance platforms such as Toutiao and verified Douyin Blue V accounts. But test first and act afterward—Doubao has 345 million monthly active users and is China’s largest AI-native app, so it is worth testing several sets of questions first to determine whether it actually cites you.
General Patterns Across Engines
Looking at only one engine makes it easy to lose perspective. Two patterns hold for most engines.
First, Zhihu is the leading content-community source across engines. QbitAI Think Tank statistics (2025-05) show that AI assistants cite Zhihu in 29.9% of their answers, ranking it first among content communities; the rate is even higher for professional-knowledge questions, reaching 35.3%. Publishing solid professional content on Zhihu is one of the few ways to feed multiple engines at once.
Second, titles and structure should be written around “how users would ask,” not around account size. An empirical NewRank study from 2026-05 based on 474,000 article observations found that whether Doubao cites a piece of content has a correlation of approximately 0 with the account’s follower count and a correlation below 0.04 with engagement data such as likes and reposts. The only significant correlation was the match between the title and the intent behind the question (0.23). In other words, writing the title as the exact question a user would genuinely ask is more useful than gaining followers.
Engine → Source Preference → What You Should Do
- Tencent Yuanbao — Source preference: WeChat Official Accounts and Tencent sources (Weixin Search / Sogou); what you should do: turn core content into WeChat Official Account articles and state the question and conclusion clearly in the title
- DeepSeek — Source preference: open webpages (Bocha Search API); what you should do: publish content on technical pages or documentation on your official website, keep it publicly crawlable, and do not put it behind a wall
- Quark — Source preference: official authoritative sources + academic databases + content communities; what you should do: align with authoritative positions, cite standards and reports, and develop content that academic databases can index
- Doubao — Source preference: industry observations suggest ByteDance sources (pending testing); what you should do: first test whether that holds, then add content on Toutiao and verified Douyin Blue V accounts if confirmed
- General — Source preference: Zhihu is the leading community source; what you should do: publish professional content on Zhihu and write titles around how users ask questions
Methodology: First Get Read → Then Get Understood → Then Get Trusted
No matter which engine you target, the underlying process has the same three steps.
First, get read. Allow AI crawlers in robots.txt, make content open and crawlable, and do not set a site-wide Disallow. If it cannot be read, everything afterward is futile.
Then, get understood. Put the conclusion first—open with a definition and a conclusion. Use more directly extractable structures such as FAQs and comparison tables. When AI extracts an answer, it favors ready-made, clearly structured passages.
Only then can you get trusted. Let third-party sources cross-check your claims instead of merely promoting yourself. Content backed by third parties is more likely to be accepted.
Do Not Guess; Test First
Because engine preferences differ so much, the easiest mistake is to publish content based on intuition: you may think you are present everywhere when in reality you are strong in Yuanbao and completely absent from Doubao.
The correct order is to first determine which engines you are absent from and which types of questions you are missing, then apply the appropriate remedy. Filling the specific gaps is much more efficient than indiscriminately publishing on every platform.
That is also what YinJen, made by Suzhou ZhiMaHang, does: daily or weekly, it measures whether you are mentioned across 12 AI engines (Doubao, DeepSeek, Kimi, Qwen, ERNIE Bot, Tencent Yuanbao, Zhipu Qingyan, StepFun, as well as ChatGPT, Claude, Gemini, and Perplexity). Using your custom question set, it provides a mention-rate score, competitor comparisons, and even paragraph-level rewriting recommendations. First see which engines you are missing from, then decide where to publish—so you do not have to guess blindly.
FAQ
Q: Can publishing one piece of content guarantee that AI will cite it? No. No one can guarantee citation or indexing. What you can do is increase the probability of being cited—publish content in the sources preferred by your target engine, structure it for easy extraction, and then test the result.
Q: If I only want Doubao to recommend me, should I focus entirely on ByteDance platforms? Test first. Doubao’s preference for ByteDance sources is an industry observation, not an official conclusion, and performance differs by industry. We recommend first using several sets of real questions to determine whether Doubao actually cites you, then deciding whether to invest.
Q: How do I get DeepSeek to index my website? DeepSeek’s online search uses Bocha to crawl the open web. Put your content on publicly crawlable pages (technical pages on your official website, documentation, or open articles), allow crawlers in robots, and it is more likely to be read than content confined inside an app.
Q: Are WeChat Official Account articles useful for other engines too? They are mainly useful for Yuanbao because it can search Official Accounts. Most other engines cannot access them, so do not expect one Official Account article to reach every engine—different engines require different sources.
Q: Does having more followers make it easier to be cited? NewRank’s empirical study of 474,000 article observations shows that follower count is almost unrelated to whether Doubao cites a piece of content. What actually correlates is how well the title matches the intent of the user’s question. Instead of chasing followers, write the title as the exact question a user would genuinely ask.
Test One Round First, Then Decide Where to Publish
If you want a particular AI engine to cite you, first determine whether you are currently present in its answers at all. Instead of guessing, use a 14-day free trial to run one round of testing: official website https://zhimahang.com/yinjen , download https://zhimahang.com/yinjen/download . Once you know which engines you are absent from, add content according to their source preferences.
