In September's four weekly monitoring rounds, 60 Doubao answers were flagged as mentioning our product, YinJen. We took apart the reference links recorded for 47 of those answers, one by one: 869 links in total, and not a single one points to our website, zhimahang.com.
This is part of the public GEO experiment we run on our own brand: the questions are fixed and repeated round after round, and the answer text and reference links are stored (reference links are deduplicated and stripped of tracking parameters, and at most the first 20 per answer are recorded; see the method section). This article answers one question only: when an AI (here, mainly Doubao) mentions us, does the list of sources it gives include our own website, and if not, what does it cite instead?
Five readings first; every section below gives the method and the source:
- Zero from our website, and not only in the answers that mention us. Across all 180 valid Doubao answers in the four rounds and the 2,649 sources recorded for them, zhimahang.com appears 0 times. Across all engines, it appears 5 times among 17,780 sources in the four rounds, from ChatGPT, Perplexity and Kimi, and all 5 point to the same article on our site; 2 of those 5 are in the single answer from ChatGPT and the single answer from Perplexity that mention YinJen. So "mentions us but does not cite our website" describes Doubao, not every AI.
- The single most-cited domain is iesdouyin.com, Douyin's video share pages. 181 links, or 20.8%; 39 of the 47 answers with sources cite at least one. The 181 links point to 149 different videos. Checked one by one against the Douyin video IDs we have registered as our own: 0 matches.
- By category, two groups are larger than Douyin share pages: news portals and self-publishing platforms, 232 (26.7%); sites run by GEO/SEO/marketing service providers and tool vendors themselves, 214 (24.6%). Developer and tech communities account for 146 (16.8%), and the remaining 96 (11.0%) go under "other".
- Content we wrote ourselves accounts for 1 of the 869 (checked against our own domain and the article and video IDs we have registered for our off-site content). It is an article we published on the Tencent Cloud developer community in early September, 10th of the 19 sources recorded for one answer in round 12. Across the four rounds, that same article appears in 7 Doubao answers; the other 6 answers do not mention YinJen.
- Some of what Doubao says about YinJen does not match public facts. Four answers to the region-specific questions mention YinJen, and all 4 place it in a group described as "not legally based in Suzhou", "not headquartered in Jiangsu" or "has no local entity in Jiangsu", while YinJen is developed by Suzhou ZhiMaHang Technology Co., Ltd.; 2 answers put YinJen in a "self-serve" tier and give annual price ranges that are both higher than the public list prices of our two self-serve plans; 1 answer lists only 5 engines (7 short), while the product covers 12 AI engines.
A disclosure up front: YinJen is our own product, and this article counts how AI engines talk about us. The data come from a single merchant's question set and are not an industry average.
1. Method: read this section before the numbers
| Item | Value |
|---|
| Data source | Self-monitoring records of YinJen (zhimahang.com), built by Suzhou ZhiMaHang Technology Co., Ltd. |
| Question set | 45 Chinese-language GEO questions in five layers: core category 21, long-tail Q&A 7, technical and integration 7, region-specific 4, competitor comparison 6. None of the 45 questions contains "YinJen", "引见" or our company name in its text. However, our monitoring ran brand questions that named YinJen in July (retired on July 29), and 2 answers that clearly mention YinJen list it among what Doubao itself calls "the few you mentioned earlier": the influence of account history cannot be ruled out; see section 8 |
| Engines | The product covers 12 AI engines; 10 to 11 of them actually produced valid answers in these four rounds (11 in rounds 9 and 10, 10 in rounds 11 and 12), 1,881 valid answers in total. This article focuses on Doubao, which completed all 45 questions in each round: 180 valid answers |
| Rounds | Rounds 9 to 12 of our weekly monitoring, split by round ID rather than by date window. Doubao's answer times (Beijing time): round 9, 09-01 14:15 to 09-02 22:09; round 10, 09-09 09:59 to 22:10; round 11, 09-17 11:45 to 09-18 11:41; round 12, 09-24 17:47 to 09-29 17:42 |
| How a "mention" is decided | The monitoring tool first flags whether our brand name appears in the answer text; we then read all 60 Doubao answers one by one. 57 clearly write "引见 YinJen"; 3 contain only the two characters "引见", run together with another tool's name, and we cannot tell whether they refer to us. None of those 3 has any sources, so they do not affect any source figure in this article |
| What a "source" is | The part of an answer's reference links that our monitoring tool records: it records links in the order it reads them, removes fragments and common tracking parameters (such as those starting with utm_), drops the engine's own pages and duplicates, and keeps at most the first 20 per answer; anything after the 20th is not recorded, and neither is how many links the engine originally gave. It is a reference list for the whole answer, not a per-sentence citation: it cannot show that any particular link supports the sentence that mentions YinJen |
| Counting unit | 1 source = 1 URL in a list. The same URL in different answers counts once per answer; no answer contains the same URL twice, because duplicates are removed when links are recorded |
| Domain | Host name with a leading www. removed; subdomains are not merged, so jiemian.com and m.jiemian.com count separately |
| Data pulled | Database snapshot taken 2026-10-08 19:08 to 19:24 Beijing time; the ByteDance comparison in section 4 and, in section 8, the number of other engines' answers hitting the 20-source cap were pulled at 20:01, when the 60 / 47 / 869 figures for Doubao's answers mentioning us were unchanged; which subsection each "earlier" phrase in section 8 sits under was checked one by one in a reread at 20:00, and the text and sources of those 60 answers in the reread were identical to the main pull. Monitoring records can be overwritten in place by re-runs: 4 of the Doubao answers mentioning us in round 12, and the one ChatGPT answer mentioning us in round 11, are re-run results; recomputing later may give slightly different numbers |
| Not included | Round 13 (started in October): it had not closed when the data were pulled |
2. Doubao overview: mentioned in every round, never sourced to our website
| Round | Valid Doubao answers | Mention YinJen | Of which have sources | Sources | Douyin share pages | Our website |
|---|
| Round 9 | 45 | 12 | 11 | 220 | 47 | 0 |
| Round 10 | 45 | 19 | 16 | 282 | 60 | 0 |
| Round 11 | 45 | 16 | 10 | 172 | 29 | 0 |
| Round 12 | 45 | 13 | 10 | 195 | 45 | 0 |
| Total | 180 | 60 | 47 | 869 | 181 | 0 |
- The 47 answers with sources have an average of 18.5 recorded sources each (869 ÷ 47); the fewest is 12 and the most is 20. Twenty is our monitoring tool's recording cap: 26 answers have exactly 20, and whether those 26 had any sources beyond the 20th, and how many, was not recorded, so 18.5 cannot be read as how many sources Doubao gives per answer. The 869 sources contain 691 distinct URLs from 260 domains.
- 13 of the 60 answers mentioning us have no sources at all. Among the 120 Doubao answers that do not mention YinJen, 25 have no sources, a similar share.
- The 60 answers mentioning us come from 31 questions. By layer: core category 37, technical and integration 9, competitor comparison 8, region-specific 4, long-tail Q&A 2. The layers have different numbers of questions (21 versus 4), so compare their structure, not their counts.
- Two other engines each have 1 answer mentioning YinJen, both in round 11: ChatGPT (6 sources) and Perplexity (3 sources). Both source lists include the same article on our site, How to Choose a Domestic GEO Monitoring Tool in 2026: Comparing Engine Coverage and Pricing. That is a sample of 2 and cannot be compared with Doubao.
- The ups and downs between rounds are differences between four observations, not a trend.
3. Where the 869 sources go
We classified sources by hand, by domain and with the help of page titles. Anything we were unsure about went to "other", so read the table as a structure, not a precise classification.
| Category | Sources | Share of 869 | Answers it appears in (of 47) | Domains |
|---|
| News portals and self-publishing platforms | 232 | 26.7% | 45 | 60 |
| Sites run by GEO/SEO/marketing service providers and tool vendors (homepages, blogs, help pages) | 214 | 24.6% | 45 | 111 |
| Douyin video share pages | 181 | 20.8% | 39 | 1 |
| Developer and tech communities | 146 | 16.8% | 34 | 18 |
| Other (academic papers, research reports, government sites, classifieds, personal blogs, etc.) | 96 | 11.0% | 38 | 70 |
| Our website, zhimahang.com | 0 | 0 | 0 | — |
Vendor-run sites are not listed individually. Excluding that category, the 10 most-cited domains are:
| # | Domain | Sources | Answers it appears in |
|---|
| 1 | iesdouyin.com | 181 | 39 |
| 2 | blog.csdn.net | 34 | 15 |
| 3 | m.toutiao.com | 31 | 21 |
| 4 | jiemian.com | 28 | 20 |
| 5 | m.sohu.com | 25 | 15 |
| 6 | cloud.tencent.com | 22 | 14 |
| 7 | developer.volcengine.com | 17 | 11 |
| 8 | m.jiemian.com | 15 | 11 |
| 9 | xhby.net | 14 | 11 |
| 10 | developer.aliyun.com | 11 | 10 |
ithome.com also has 11; ties are broken alphabetically by domain, which puts it after developer.aliyun.com.
How to read this:
- In the news and self-publishing category, about half of the pages are, judging by title, "help you choose" articles. Of the 232, 117 have a title containing one of these Chinese words: recommend, roundup, which one, service provider, review, evaluation, comparison, selection, chart, reliable, top list, hands-on test, how to choose, which ones. That is a rough keyword filter; it only shows that about half of these sources are selection-style articles, not who wrote them.
- The same platform is often spread across several addresses. The Tencent Cloud developer community appears under 5 different host names, 43 sources across 21 answers in total; 4 host names under the CSDN domain total 40 sources across 16 answers. Looking at single domains alone underestimates platforms like these.
- Vendor-run sites show up in almost every answer. 45 of the 47 answers cite at least one homepage, blog or help page run by a GEO/SEO/marketing vendor, covering 111 domains in China and abroad. We do not evaluate or name any of them.
- A count is only a count. It says nothing about the quality or credibility of any site.
4. Douyin share pages: 181 sources, all of them other people's videos
- The 181 sources point to 149 different video share pages; no single video is cited more than 4 times.
- We checked them one by one against the Douyin video IDs registered in our records as ours: 0 matches. All 181 are other people's work, and we do not name any creator.
- Compared with the answers that do not mention us: of the 120 Doubao answers in the four rounds that do not mention YinJen, 95 have sources, 1,780 in total, of which 337 (18.9%) are Douyin share pages; on the mention side the share is 20.8%. The two are similar, so there is no sign that the answers mentioning us lean especially toward video (this is a direct comparison of two shares, with no significance test). Why Doubao cites videos so often is not something these data can answer.
- Putting together the platforms owned by ByteDance (Douyin share pages, Toutiao, Volcano Engine, Juejin and Feishu, matched by host-name suffix; Juejin was acquired by ByteDance in 2020, as reported publicly at the time) gives 240 sources, 27.6% of the 869, and 45 of the 47 answers cite at least one of them. In How to Get Doubao, DeepSeek, and Quark to Cite You: Breaking Down Chinese AI Engines’ Source Preferences we wrote that "Doubao leans toward ByteDance properties" was an industry observation worth testing. Our article Where Should You Publish Content to Get Cited by AI? did test it once and found that 36% of Doubao's citations came from its own company's ecosystem; that test covered the window 07-28 to 08-11 and classified citations by whether they belong to a content platform owned by the same company. This section matches the 5 platforms above by host-name suffix, which gives 25.0% (661 of 2,649) across all 180 Doubao answers in the same four rounds. The windows and methods differ, so the two figures cannot be compared directly, and they do not show a change. Here is a further comparison using this section's method: in the 95 Doubao answers that do not mention YinJen and have sources, these platforms account for 23.7% (421 of 1,780); over the same four rounds, all sources from the other engines combined give just 3.8% (571 of 15,131), and the highest single engine is 13.6% (all of these are computed on records capped at 20 sources per answer, and the share of answers hitting that cap differs widely between engines; see section 8). On this question set, ByteDance platforms make up a clearly larger share of Doubao's sources than of the other engines'; but this is four rounds of observation on one category and one merchant's question set, it cannot be extended to other categories, and it does not explain why.
5. Why Doubao's sources never include our website: what we can and cannot say
What we can say:
- It is not only that the answers mentioning us happened to skip our website: none of the 2,649 sources recorded for Doubao in these four rounds is our website (84 of the 180 valid answers hit the 20-source recording cap, and whether those answers had any sources beyond the 20th, or our website among them, was not recorded).
- Across all engines in the four rounds, 1,881 valid answers and 17,780 sources, zhimahang.com appears 5 times (Perplexity 2, ChatGPT 2, Kimi 1), all pointing to the same comparison article; only 2 of those 5 answers also mention YinJen, the two described in section 2.
- Of the content we have published on third-party platforms, 1 item appears among the 869: an article we posted on the Tencent Cloud developer community in early September, 10th of the 19 sources recorded for one answer in round 12. We found it by matching, one by one and exactly, the article (and video) IDs we have registered for our off-site content; it sits under the cloud.tencent.cn address, so a check against cloud.tencent.com alone would miss it. The same article appears in 7 Doubao answers across the four rounds, and the other 6 do not mention YinJen. Being cited and being named are two different things.
What we cannot say:
- Why our website is absent from Doubao's sources. There are many possible reasons: whether the site is within its retrieval scope, how the questions are phrased, its preferences among source types, and so on. These data cannot test any of them, and we will not guess.
- Whether "our website is absent from the sources" and the "descriptions that don't match" in the next section are causally linked. The sources are a reference list for the whole answer, and we cannot tie any particular description to any particular link.
6. Does what it says about YinJen match what we say?
With no source from our website, how does Doubao describe YinJen? We read the 57 answers that clearly mention YinJen one by one. Most simply list YinJen alongside several similar tools; those that describe it on its own do not agree with each other. Examples (fragments of Doubao's wording, translated, with other companies' names removed):
- "lightweight GEO monitoring"
- "focused on auditing AI mindshare"
- "focused on source tracing plus entity credibility (E-E-A-T) analysis"
- "leaning toward knowledge assets and source governance; a tool for building a GEO knowledge base"
- "leaning toward being a GEO service provider with a built-in monitoring module ... rarely sells standalone tool accounts"
For how we describe YinJen ourselves, see What Is YinJen? A GEO Tool for Monitoring Brand Visibility in AI.
Of the statements that can be checked against public facts, the table below lists three that do not match (there is also half a sentence, explained after the table):
| What does not match | Answers | What Doubao says | Public fact |
|---|
| Where the company is | 4 | Groups YinJen with several similar vendors and says this group is "not legally based in Suzhou", "not headquartered in Jiangsu" or "has no local entity in Jiangsu" | YinJen is developed by Suzhou ZhiMaHang Technology Co., Ltd. These 4 are exactly all of the region-specific answers that mention YinJen |
| Self-serve price | 2 | Both answers put YinJen in a "self-serve" tier: 1 gives an annual range shared by several tools including YinJen, and in the other the price line sits under a subheading that also carries another company's name | The lower end of both ranges is higher than the public list prices of the two self-serve plans on our pricing page (monthly price × 12) |
| Which engines it covers | 1 | The question asked whether a tool can monitor 4 Chinese engines at once; under "coverage" the answer lists only 5 Chinese engines | The product covers 12 AI engines, 7 more than listed |
This table counts only statements that can be checked against public facts and do not match them, but the three rows are of different kinds: the location is wrong; neither of the 2 price answers is a quote for YinJen alone, so all we can say is that the self-serve price does not match; and the 5 engines listed are all among the 12 we cover, so that answer says too little rather than something wrong. Most of the positioning descriptions quoted above differ from ours without being right or wrong, and they are not in the table; the one exception is the half sentence "rarely sells standalone tool accounts", which can be checked: the two self-serve plans on our pricing page are tool accounts anyone can buy on their own, so that half sentence does not match either. It sits inside a positioning description, and we have not given it a row of its own.
7. What this is useful for in GEO
What you can use it for:
- Look separately at what the answers that mention you cite. A mention count only tells you that you appear in AI answers (for how to measure the mention rate itself, see How Can Companies Measure Their AI Mention Rate in Doubao, DeepSeek, and Qwen? (Includes a Free Manual Method)); pulling out the sources of those answers and counting them by domain shows which references the AI listed when it mentioned you. That list is a reference list for the whole answer; it does not mean the AI learned about you from those places (see "What a source is" in the method section, and section 5).
- Check what the AI says about you against the facts on your own website, item by item. Location, price and scope of capabilities all have public facts to check against; the mismatches in this article's table happen to be exactly these three.
- To check whether your own content made it into the sources, match at the article level: compare by the ID the platform assigns to each article (or video), and count all of the platform's host names (the Tencent Cloud developer community has 5 in this article). Looking at domains alone both misses and overcounts; comparing the registered URL character by character also misses: our Tencent Cloud article is registered under a cloud.tencent.com address, but all 7 Doubao answers that cite it in the four rounds use a cloud.tencent.cn address with an extra query string (see section 5).
What you can't use it for:
- This article makes no promise about indexing, rankings or any other outcome. The figures are observations from four past rounds and do not predict whether any site or any piece of content will be cited or mentioned in future.
- Extrapolating to other categories, brands or engines. The 45 questions cover only the GEO category and look at only our own brand; the two answers from ChatGPT and Perplexity that mention YinJen both include an article on our site among their sources.
- Reading it as "post on Douyin or a developer community and you will be cited". The videos in the sources are all other people's; these data answer only "when we are mentioned, what is cited", not "how to get cited".
- Judging which source type is better by count.
8. Limits of these data
- Doubao's answers may be affected by account history. 8 of the 60 answers mentioning us contain phrases like "the ... you were looking at earlier" or "the few you mentioned earlier". In 2 answers that clearly mention YinJen, YinJen is listed under what Doubao itself calls "the few you mentioned earlier" (neither of those two questions mentions YinJen); in 1 of the uncertain answers, the two characters "引见" also sit in a string of tool names introduced as "the ones you saw earlier". Our monitoring ran brand questions that named YinJen in July; they were retired on July 29, and their run records can no longer be retrieved; on August 29 we also created a question that names YinJen, and the database holds no run records for it at all, so whether it was ever run cannot be verified either. We cannot check whether these "earlier" references point to those conversations, and they may also just be Doubao's way of talking; so we cannot say that the mentions in this article have nothing to do with our name having been asked about. Asking from a different account could give different results.
- At most 20 sources are recorded per answer. This is our monitoring tool's recording cap and does not mean Doubao gives only 20: 84 of the 180 valid Doubao answers hit exactly 20 (26 of the 47 answers that mention YinJen and have sources), and whether those answers had any sources beyond the 20th was not recorded, so "0 from our website" in this article holds only for what was recorded. Among Doubao's answers, those that mention YinJen and those that do not hit the cap at similar rates (26 of 47, 58 of 95); the other engines differ widely, ranging from engines where no answer hit the cap to one where 164 of 180 did. So the cross-engine shares in section 4 are computed on source lists truncated to different degrees.
- "Mention" relies on automatic flagging plus a manual reread. The 3 uncertain answers are listed separately; none has sources, so they do not affect the source figures, but they do affect the total of 60.
- The classification was done by hand. There are gray areas between vendor sites, news and self-publishing; for example, an article a vendor publishes on a news platform is counted by platform, not by author.
- Only four rounds, and only one engine at the center. ChatGPT and Perplexity have just 1 answer each that mentions us, which says nothing reliable about their habits.
- Snapshots change. Re-runs overwrite monitoring records in place; every figure in this article is a snapshot as of the time given above.
Sources and scope: All figures come from rounds 9 to 12 of YinJen's weekly self-monitoring (Beijing time 2026-09-01 to 09-30), pulled 2026-10-08 19:08 to 19:24 Beijing time (ByteDance comparison and other engines' cap counts pulled at 20:01, section 8 subsection placement reread at 20:00); 45 Chinese-language GEO questions; the product covers 12 AI engines, 10 to 11 were actually run across the four rounds; 180 valid Doubao answers in the four rounds, 60 of them flagged as mentioning YinJen (3 uncertain), 47 with sources, 869 sources in total (at most 20 recorded per answer); measurements on a single merchant's question set, not an industry average.
This article was written by Suzhou ZhiMaHang Technology Co., Ltd. (the YinJen team). The answer-level details used in this article have not been published; our public citation datasets are under the GitHub organization ZhiMaHang (verified domain ownership: zhimahang.com). Website: zhimahang.com
Data in this article comes from production-environment measurements by a GEO monitoring tool built in-house by the author's company. The text was drafted with AI assistance and reviewed by the author before publication.
Further reading
