Wikipedia AI visibility: the highest-mentioned Western source.

The 21% in context

Three things to keep in mind before reading this number as a strategy directive. First, the panel is one category — luxury handbags. Watch and luggage panels were smaller and we have not yet confirmed the same Wikipedia rate carries. Second, the engine measured is DeepSeek; Qwen and Doubao have different source-mix preferences, and we have not yet decomposed the Wikipedia rate per engine for this category. Third, Wikipedia is one entry on the Western-source list, not the dominant source overall — Mainland-Chinese sources (SMZDM, Zhihu, Xiaohongshu) account for the bulk of citation surface across all engines.

Why Wikipedia surfaces so consistently

Wikipedia is the encyclopaedic anchor of the open web. Five structural properties make it dependable for engines:

• Encyclopaedic prose. Wikipedia articles read in third-person definitional voice — exactly the registered AI engines absorb most cleanly. There is no marketing copy, no "we believe", no editorial framing.
• Citation discipline. Every substantive claim is footnoted to a secondary source. Engines that absorb Wikipedia content inherit that citation chain, which improves the engine's ability to ground answers.
• Structured-data richness. Wikipedia articles ship dense Wikidata entity links (sameAs), infoboxes with typed fields, category memberships, and inter-language links. This is the kind of skeleton that JSON-LD tries to be on a single brand site, but Wikipedia has it across millions of entities.
• Stability. Wikipedia URLs change rarely; article histories are durable. Engines that cached an article in a training run can still cite it months later.
• Cross-language coverage. A Wikipedia entry usually has Chinese-language equivalents (中文 Wikipedia or Baidu Baike, depending on which the engine indexes). Western brands with English-only assets often surface in Chinese AI through this cross-language bridge.

None of this is unique to AI search — Google's Knowledge Panel uses Wikipedia as one of its primary entity sources, and has for over a decade. What changed is that the same encyclopaedic substrate that powered Knowledge Panels now also powers AI answer engines.

How to earn a Wikipedia entry without being deleted

The most common failure mode for brands trying to build Wikipedia presence is being speedy-deleted under the A7 (no credible claim of significance) criterion. Speedy deletion is more damaging than no entry at all — it leaves a record that hurts future attempts. The bar is notability, not visibility.

The four practical criteria for surviving review:

1. At least three independent secondary sources. Coverage in Search Engine Land, MarTech, SCMP, FT, Wired, TechCrunch — each providing substantial reporting (more than a brand mention in a roundup). Press releases do not count. Self-published content does not count. Sponsored content does not count.
2. Neutral tone in the proposed article. Encyclopaedic voice. No "innovative", "leading", "trusted by". State what the entity is and what it does, with citations.
3. Verifiable claims only. Every fact in the article — founding date, headquarters, products, key personnel — must be supported by a citation to one of the secondary sources. Unsupported claims are stripped during patrol.
4. Don't paid-edit. Wikipedia's conflict-of-interest policy requires disclosure of paid editing. Undisclosed paid editing leads to article deletion and can lead to a topic ban from the platform.

In practice the bottleneck for new-entity brands is the secondary-source collection. The other three criteria are tractable once the citations exist. A reasonable cadence is: target one quality piece per quarter from a recognised publication, building the source base for 9–18 months before a first submission attempt. Then the article writes itself.

What not to do

• Do not pay a "Wikipedia editing service". The vast majority of these are paid-editor sock-puppet networks; their work is detected and reverted, and the article is permanently flagged. Worse, the brand becomes associated with the network in editor-tracking databases, which makes future legitimate edits harder.
• Do not submit before the secondary sources exist. A submission with two press releases and a self-published blog post will be deleted A7 within hours. The deletion is logged.
• Do not direct-edit your own brand article. If an article exists, propose changes on the article's Talk page with the COI template. Wait for an uninvolved editor to make the change.
• Do not assume Wikipedia is the only encyclopaedic surface. Wikidata (the structured-data sibling) and Baidu Baike (the Mainland-CN encyclopaedia) carry independent weight for different engines. A complete encyclopaedic strategy covers all three.

Caveats and uncertainties

The 21% rate is from one engine (DeepSeek) on one category (luxury handbags) with n=1,620 responses. We have not yet confirmed equivalent rates on Qwen or Doubao. Reliability statistics for the panel: top-5 source-attribution κ = 1.00 (perfect agreement across re-runs), top-15 κ = 0.46 (moderate); the long-tail noise means individual long-tail Western sources are less stable than this top-tier number, but Wikipedia consistently appeared in top-5 results across re-runs. We recommend treating Wikipedia as a high-confidence Western citation surface for DeepSeek in the categories tested, and as a hypothesis worth verifying before extrapolating to other categories or other engines.

We are not measuring sales, conversion, or attributable revenue from Wikipedia-mediated AI mentions. Whether brands that surface in DeepSeek answers via Wikipedia citations close more business in Mainland China is a separate question we did not address.