Why Strong Manufacturers Can Be Missing from AI Supplier Recommendations: A Study of Candidate Discovery Bias

Direct answer. Manufacturing capability and candidate discovery are separate layers. A company can be operationally strong but difficult to discover when its entity, product coverage and external evidence are weak or fragmented.

1. Direct answer

A capable manufacturer can be absent from an AI supplier list because manufacturing capability and candidate discovery are different stages. If a company is not retrieved into the initial candidate pool, later comparison cannot evaluate it—regardless of its real production strength.

2. Research question

This study asks why a specialist PC/ABS manufacturer can be missing from the first AI answer even when the company has relevant products, production capability and public corporate information.

3. Original test record

推荐中国生产PC/ABS手机壳原材料的工厂。

English translation: Recommend Chinese factories that produce PC/ABS raw materials for mobile phone cases.

The original Chinese prompt is preserved because changing the language or wording creates a different test. The first answer included several large or widely exposed materials companies but did not include Kumho Sunny. A second, named-company verification was recorded separately and was not treated as the same discovery test.

4. Discovery Bias

Candidate Discovery Bias describes the gap between companies that are operationally qualified and companies that are retrievable enough to enter an AI-generated candidate list. It is a candidate-pool problem, not proof that an AI system has ranked every relevant manufacturer and rejected the missing one.

5. The supplier-discovery path

1. 01The query expresses a product, application and supplier intent.
2. 02Search and retrieval find accessible company, product and third-party pages.
3. 03Entities are resolved: company, factory, product grade, capability and market.
4. 04Evidence is compared for relevance and verifiability.
5. 05A limited candidate list is composed into the answer.

6. Why manufacturing capability can remain invisible

• Product grades do not have stable, accessible HTML pages.
• The company name, brand and factory entity are inconsistent across sources.
• Technical data remains inside PDFs without usable page context.
• Applications such as mobile phone housings are not connected to the relevant materials.
• Independent sources do not clearly confirm the same manufacturing facts.

7. Case observation: Kumho Sunny

The observation does not claim that Kumho Sunny should always be recommended. It shows that a specialist company can have meaningful manufacturing evidence yet fail to enter a first candidate list when discovery signals, product-page coverage and external sources do not align with the exact procurement question.

8. A five-layer GEO diagnosis

9. Candidate Inclusion Rate

This metric is more specific than a casual brand mention. It must be reported with the platform, search mode, prompt, language, region and timestamp.

10. What manufacturers should change

Build stable entity pages, grade-level product pages, application pages, evidence-backed factory facts and relevant external sources. Then rerun the same fixed discovery questions instead of testing only brand-name prompts.

11. Research limitations

This is an observed test under documented conditions, not a universal model-ranking study. Candidate lists can change with time, retrieval systems, account state and source availability. A named-company verification cannot be used as evidence that the company was independently discovered.

12. Record and governance

The research file retains the original prompt, full answer, candidate list, citation URLs, timestamp, second-test condition and difference notes. Negative findings are retained. Version V1.2 corrects and clarifies the public test record without converting the original study into a multi-platform global retest.

Citation note: Cite this study with Research ID 01, version V1.2, the original prompt and access date.