GEO Optimization Recalibration: DeepSeek and Tencent Yuanbao "Blacklisted" from Rankings? No, the Industry Is Returning to Rationality

Over the past week or two, discussions about "DeepSeek and Tencent Yuanbao no longer indexing rankings" have quietly gained momentum in the industry. A friend who has been in the content business for years told me that DeepSeek's search results seem to have become "sluggish", with cited data still stuck in September 2025; shortly after, peers sent over analytical articles, filled with concerns about sudden changes to platform rules.

I understand this anxiety. Over the past year and more, "rankings" have become almost the most abused content form in GEO (Generative Engine Optimization) due to their high replicability and strong dissemination power. When people say "rankings are no longer effective", many people's first reaction is not to reflect on the content, but to suspect the rules.

However, as someone who has long specialized in SEO and personally managed GEO projects, my judgment is quite the opposite: this time, it is not that the platforms have "banned" anything, but that the industry is being forced to return to a more rational state.

I. The Value of Rankings Never Lies in "Who to Recommend"

First and foremost, we must clarify a fundamental logic: in the eyes of AI, rankings are a "structure", not a "conclusion".

What AI excels at is not judging "who is the best", but classifying, comparing, summarizing, and establishing structured cognition. Rankings are precisely a highly structured way of presenting information—they have a clear scope of objects, comparison dimensions, and logical relationships. Therefore, in principle, AI cannot inherently reject rankings.

The real problem is not the rankings themselves, but the "fabricated rankings" that have been detected by AI. Over the past year, a large number of rankings have been characterized by "prioritizing rankings over logic, unexplained indicators, and vague data", essentially serving only a single commercial goal. Such content might have slipped through the cracks in the era of traditional SEO, but in the AI search system, as models evolve, it is inevitably being gradually downgraded. The reason is simple: AI cannot judge its authenticity and stability.

II. Bottom Line for Compliant GEO: Rankings Cannot Be Fabricated and Must Be Interpretable

Therefore, the first red line for compliant GEO is that rankings cannot be fabricated.

What constitutes fabrication? It means rankings without basis, unexplainable "comprehensive evaluations", no time/sample scope, and conclusions clearly serving marketing purposes. For example, "Based on comprehensive industry reputation, technical strength, and user feedback, the top ten brands are selected"—if you cannot even explain where the reputation comes from, how technology is measured, or the size of the sample, then in AI's view, this is a piece of subjective text that cannot be reused.

A compliant ranking must meet at least three criteria: having sources (industry reports, public data), having methods (able to explain the ranking logic), and having boundaries (applicable scenarios, time cycles). AI is not afraid of imperfection; it is afraid of you pretending to be perfect.

III. Risk Assessment from Titles: Which Are More Compliant, Which Carry Higher Risks?

To illustrate this intuitively, let's look at two groups of titles:

Relatively compliant ranking titles:

• 2025 Inventory of Mainstream Brands of Automated Packaging Machines: Categorized by Application Scenarios

• Inventory of B2B Official Website Information Types Frequently Cited in AI Search

• Distribution of SaaS Product Function Types Compiled Based on Public Information

Characteristics of high-risk rankings:

• Forcibly assigning labels like "No.1", "Authoritative", "Only"

• No sources or methodologies

• The ranking itself is the selling point

The core difference between the two is: the former is information collation and structural analysis, while the latter is a referee-style conclusion.

IV. In AI Search, "Searchable/Indexed/Cited" Are Three Distinct Concepts

This is the most misunderstood point in the current controversy. Many people simplify the problem to "AI no longer indexes rankings", but in reality, the issue lies in changes to the citation hierarchy.

• Searchable by AI: Only technically accessible, meaning "visible", not "usable".

• Indexed by AI: Refers to content entering AI's knowledge modeling layer, where concepts and entities are absorbed and may be used as background information, but the source is not necessarily displayed. You may find that the content is not cited by name, but the structure of AI's responses is clearly influenced by it.

• Cited by AI: This is the highest level of trust, meaning the content can be independently retold, with stable facts and a clear stance. Rankings, industry judgments, etc., are more commonly indexed but not directly cited.

V. Key Clarification: What AI Treats Cautiously Is Not Rankings, but "Content with Clear Advertising Attribution"

Currently, many clients mistakenly believe that AI does not cite "marketing soft articles". In fact, when AI sees content labeled "Advertisement/Promotion/Cooperation", it will judge whether it is influenced by commercial cooperation. Its approach is: information is absorbed, but not prioritized as an "objective factual source".

This is a risk control strategy, not a denial of value. Structurally clear and factually definite parts may still be cited under appropriate questions. It just has a higher threshold, rather than being rejected outright.

VI. Core of Compliant GEO: Not Avoiding Marketing, but Not Pretending

From my years of practice in SEO and GEO projects, my biggest takeaway is: what AI truly dislikes is not commercial or brand-promotional content, but disguising biased content as absolutely neutral facts.

As long as you clarify your stance, explain your sources, do not fabricate or exaggerate, your content will instead be more easily absorbed long-term.

Conclusion: