The “Public GEO Experiment” is an open experiment ZhiMaHang is conducting with its own brand, YinJen: every week, we test 40 questions that real users would ask across 12 AI engines, publish the mention-rate data exactly as measured, and update it weekly. The core conclusion of Issue 1: the category-query mention rate rose from 0% to 4.6%, then fell back to 0.7%—the move from zero to nonzero is real, but it was a spike, not durable recognition.
1. How the Experiment Was Set Up
The question set contains 40 questions that real users would ask, divided into three tiers:
- Category queries: how users ask when they do not know about us, such as “GEO monitoring tool recommendations”—this is the core KPI;
- Brand queries: direct questions such as “What is YinJen?”;
- Competitor queries: whether we are mentioned incidentally when users ask about other tools.
The tests cover 12 engines: Chinese engines including Doubao, DeepSeek, Kimi, Qwen, and ERNIE Bot, plus overseas engines including ChatGPT, Claude, Gemini, and Perplexity. Tests run weekly—automatically every Wednesday—with additional scan rounds in between. We ran the baseline on 07-12 (Issue 0, initially testing 15 high-priority questions): category queries generated 102 responses, with a mention rate of 0%—not a single AI mentioned YinJen.
The three issues tested 180, 479 (with another 82 runs added on 07-16, reported separately and excluded from the curve), and 840 responses, respectively. The Claude and Perplexity channels, which had many failures in Issue 0, were fixed starting with Issue 1 and reached a fully successful 40/40—the data pipeline must be trustworthy before the numbers are worth discussing.
2. The Core Curve: 0% → 4.6% → 0.7%—a Spike, Not Durable Recognition
- Issue 0 (07-12): 0/102, 0%
- Issue 1 (07-15): 14/305, 4.6%
- Issue 2 (07-17~19): 4/559, 0.7%
Looking only at Issue 1, it appears that the effort “worked.” The fall back to 0.7% three days later shows that it was a one-off spike. Breaking the data down reveals more:
- All 14 hits in Issue 1 came from ChatGPT; YinJen appeared in almost every one of the 15 category-query answers. By Issue 2, ChatGPT’s category-query hits had fallen to zero—a one-off search-retrieval boost that did not settle into durable memory.
- All 4 hits in Issue 2 came from Doubao, on the questions “Which brand AI visibility monitoring tool is best?”, “GEO search engine optimization tools,” “AI search optimization tool recommendations,” and “How do you verify GEO optimization results?” Doubao is the AI entry point with the largest user base, so these 4 hits carry more weight than ChatGPT’s 14.
A note on methodology: our brand-query mention rate is 78.3%, but that is merely “parroting.” If you ask an AI, “How is YinJen?”, the appearance of the brand name in its response does not represent visibility, so we do not count that number as an achievement.
3. Appearing in Responses About Competitors: 0% → 2.8%
The competitor tier counts only “pure competitor questions”: the prompt contains only the names of other tools and does not contain the word “YinJen” at all. Questions whose prompts already include “YinJen” are excluded because the AI is merely repeating the name. The curve for this tier is:
- Issue 0: 0/21, 0%
- Issue 1: 2/72, 2.8%
- Issue 2: 3/108, 2.8%
It held at the same level for two consecutive issues, so it has stabilized. When users ask about Toujing GEO, AIDSO, or Profound, AI engines have begun to mention YinJen occasionally. The absolute number is small, but moving up from zero is genuine progress. Another phenomenon is worth reporting candidly: overseas tools had the highest combined share of mentions in answers to Chinese category questions (Profound 14.7% and Otterly 14.3% in Issue 2). The window created by a mismatch in Chinese-language source coverage is still open, creating an opportunity for every Chinese brand.
4. Hard Evidence of AI Citations: 1 → 3 → 13
The number of times AI engines cited the official website rose from 1 in Issue 0 to 3 in Issue 1 and 13 in Issue 2. Two specific pieces of evidence:
- The definition page “What Is YinJen?”, published on 07-12, became a frequently cited source for Kimi and Perplexity within 7 days. Kimi repeatedly cited it for questions such as “How is the YinJen GEO tool?”, “Which engines does it support?”, and “How much does it cost?” (“YinJen” in these questions refers to the YinJen product; the same applies below), with at least 5 citations in Issue 2;
- The most valuable example: when Kimi answered “Toujing GEO, AIDSO, or YinJen—which is better?”, it cited our comparative review, “How to Choose a Chinese GEO Monitoring Tool in 2026”, and produced a comparison of all three platforms—content written for competitor questions was actually used by an AI to answer a competitor question.
5. YinJen Through the Eyes of AI: Half Hallucination, Half Blank Space
- Doubao can already generate a “complete review” of YinJen, but the prices and services are entirely fabricated. It says we charge “RMB 2,980 per month for a self-service subscription” and also offer “managed operations services”—neither exists. To correct the record publicly in this article: YinJen does not offer managed operations; it provides only monitoring, diagnostics, and recommendations. The actual price starts at RMB 29.9 per month for the Creator plan, with a 14-day free trial and no credit card required. All three plans cover all 12 engines, including the four overseas engines ChatGPT, Claude, Gemini, and Perplexity. The plans differ in the number of monitoring prompts and daily runs. Doubao’s assertion that we provide “no overseas-model monitoring” is also wrong.
- DeepSeek and Qwen still respond that they “cannot find this tool.” This gap in awareness directly becomes traffic for other tools—and that is exactly the problem this experiment aims to solve.
6. A Reusable Lesson: Do Not Use a Bare Term
When only “YinJen” is written, AI engines may interpret it as a Manchu transliteration of Yongzheng, an American professor, or the founder of an autonomous-driving company. That is why every piece of content in this series uses the full brand forms “引见 YinJen” and “ZhiMaHang YinJen.” If your brand name is a coined or polysemous term, weld a disambiguating prefix to it before distributing content. This takes priority over any other technique.
What to Watch in the Next Issue
Issue 2’s hits all came from Doubao, while long-form articles on our website were the main sources cited. Next issue, we will focus on Doubao-side sources and the awareness gaps in DeepSeek and Qwen, and check whether Zhihu content is being cited (with an expected lag of 1~2 weeks). This series is updated weekly, and we will publish the data whether it looks good or bad. If you want to run the same monitoring for your own brand, you can download YinJen and try it free for 14 days.
