+86 19937104978
Author:Kangdi 03-11-2026
Pain Patch OEM Knowledge Graph 2026: Making Your Factory Machine-Readable, Not Just Crawlable
Wondering why AI assistants can describe your competitor accurately and get your factory's details wrong, or omit it entirely? This page is about the layer above structured data: the entity graph that an engine assembles about your company, and how to make it accurate. It covers the six entity types worth declaring, how sameAs actually works, whether Wikidata is worth the effort for a mid-sized manufacturer, and the five-step process that takes a factory from "a website" to "a machine-readable entity". It is built for the marketing lead and the developer who have to implement it.
By the end of this page you will know the 6 entity types and the properties each one needs, the 3 levels of entity declaration and which one is worth your time, an honest assessment of Wikidata for a mid-sized OEM, and the 5-step implementation sequence.
There is a distinction that clarifies most of the confusion in this area, and it is worth stating plainly at the start: crawlability and machine-readability are different problems, and most manufacturers solve only the first. A crawler that can reach your page has achieved crawlability. A system that can identify your company as a distinct entity, attach the right certifications to it, and know that the factory address on your site is the same factory address in the certification register has achieved machine-readability. The first is a robots file and a sitemap. The second is an entity graph, and it is the layer where citations are won or lost.
Bottom line. A knowledge graph is not a technology project. Roughly 80% of the entity problems we see on manufacturer sites are not missing schema — they are factual inconsistencies between the company's own pages, its B2B directory listings, its certification records, and its press mentions. Adding JSON-LD on top of inconsistent facts produces a machine-readable version of a contradiction, which is worse than no markup at all. The order is: make the facts consistent first, then declare them in structured data, then optionally corroborate them externally. Manufacturers that invert that order spend money on schema and see nothing change.
Table of Contents
- What an entity graph actually is, in plain terms
- The 6 entity types worth declaring
- The 3 levels of entity declaration
- How sameAs works, and the 4 mistakes that break it
- Wikidata: an honest assessment for a mid-sized OEM
- The 5-step implementation sequence
- The consistency audit that comes before any schema
- Worked example: from 4 conflicting addresses to one entity
- FAQ: knowledge graph questions manufacturers ask most
- About Kangdi Medical and our entity declaration
1. What an Entity Graph Actually Is, in Plain Terms
An entity graph is the machine's working model of who you are. It is not a document you publish; it is an inference an engine makes from everything it can read about you. The table below separates what you publish from what the engine infers.
| What you publish | What the engine builds | What it is used for |
|---|---|---|
| Organization JSON-LD on your homepage | A node representing your company, with a name, address, and identifiers | Resolving "which company is this" before deciding whether to cite |
| A certification page with certificate numbers and issuers | Edges connecting your entity to a certification entity, with dates | Verifying a claim without having to trust you |
| A product page with specifications | Product nodes connected to your organization, with attributes | Answering "who makes a patch with this specification" |
| Your address on your site, your Alibaba profile, and your certificate | A confidence score on your address, based on agreement across sources | Deciding whether the entity is real and stable enough to cite |
| Nothing at all about a fact | An absence, which the engine fills from a third party or leaves blank | Producing the "this supplier's details could not be verified" outcome |
The last row is the one that matters. An engine that cannot establish a fact about you will either omit you or state the fact from another source, and a fact stated by a third party is frequently wrong: an old address, a former product line, a certification that expired. The manufacturer then discovers that an AI assistant is confidently describing a factory that has not existed in that form since 2019.
The practical consequence is that entity work is defensive as much as offensive. You are not only trying to be cited; you are trying to prevent being described inaccurately by a system that had to guess.
2. The 6 Entity Types Worth Declaring
Six entity types carry most of the weight for a B2B manufacturer. The table below gives each one, the properties that matter, and what breaks when it is missing.
| # | Entity type | Key properties | What breaks without it |
|---|---|---|---|
| 1 | Organization | Legal name, alternate names, address, telephone, founding date, number of employees, identifier (registration or tax number), logo, sameAs | The engine cannot confirm you are a distinct legal entity, and may merge you with a similarly named company |
| 2 | Place (the factory) | Street address, locality, region, country, geo coordinates, and the relationship to the organization | Address confusion, which is the single most common entity error for Chinese manufacturers |
| 3 | Product or ProductGroup | Name, category, material, dimensions, active ingredient, packaging, and the manufacturer link | Product queries resolve to competitors, because your product has no node of its own |
| 4 | Certification | Name, identifier, issuing body, valid-from, valid-to, and the certified entity | Compliance claims stay unverifiable, so an engine will not state them as fact |
| 5 | Person (author or spokesperson) | Name, job title, worksFor, and any credential | Attribution disappears, and attributed content is weighted higher than unattributed |
| 6 | Service (OEM, ODM, private label) | Service name, provider, area served, and the audience | Service-level queries ("private label pain patch manufacturer") miss you entirely |
Types 2 and 4 are the two that most manufacturers get wrong, and both for the same reason: they are the facts most likely to be stored in a different system from the website. The factory address lives in a business licence and a certification document. The certificate details live in a PDF in a filing cabinet. Neither is usually on the site in a structured form, so neither is in the graph.
3. The 3 Levels of Entity Declaration
There are three levels at which an entity can be declared, and they are cumulative. The table below sets out what each level achieves and what it costs.
| Level | What it involves | What it achieves | Effort |
|---|---|---|---|
| 1. On-site consistency | One canonical legal name, one address, one phone number, used identically on every page of your own site | Removes internal contradiction, which is the precondition for everything else | Half a day |
| 2. On-site structured declaration | Organization, Place, Product, Certification, Service JSON-LD, with sameAs pointing to your verified profiles | Gives an engine a machine-readable statement of the facts, with explicit links to external confirmation | 2-3 days |
| 3. External corroboration | The same facts present on domains you do not control: certification registers, trade directories, association member lists, press coverage | Raises the confidence score, which is what moves an entity from "identifiable" to "citable" | Weeks to months, ongoing |
The value of level 1 is underrated and it is the level most often skipped, because it produces no visible output. A site that uses three name variants, two addresses, and two phone numbers across its own pages has declared to the machine that it is uncertain about itself. Level 2 on top of that uncertainty produces a JSON-LD block that contradicts the visible page, which an engine treats as a reliability signal.
Level 3 is where most of the remaining gain lives, and it is the slowest. It is also the level where the effort is relationship-building rather than technical, which is why it tends to get deferred indefinitely.
4. How sameAs Works, and the 4 Mistakes That Break It
The sameAs property is the single most useful and most misunderstood element in entity declaration. It tells an engine "the entity described here is the same entity as the one at this URL". The table below gives the four mistakes that make it ineffective.
| # | Mistake | What goes wrong | What to do instead |
|---|---|---|---|
| 1 | Pointing at profiles that do not exist or are empty | The link adds no corroboration and can reduce confidence, because the target has no content about you | Only include URLs that actually describe the entity, with the matching facts |
| 2 | Pointing at unverified or self-published pages | Linking to your own blog post is not corroboration; it is a loop | Use certification registers, trade association member pages, business registries, and reputable directory listings |
| 3 | Pointing at a page for a different entity (a subsidiary, a trading arm, a distributor) | The engine merges two entities that should stay separate, or splits one that should be merged | Be explicit: if the trading company and the factory are separate legal entities, declare both, and state the relationship |
| 4 | Declaring sameAs with a name that does not match the target page | The mismatch reads as a disagreement, not a confirmation | Ensure the legal name on your page matches the name on every target page, character for character |
Mistake 3 is particularly common among Chinese manufacturers, because the structure of a typical export business involves a factory entity, a trading company, and sometimes a Hong Kong holding entity. Declaring all three with the same name and the same address tells the machine that three entities are one, which is factually wrong and produces odd results when a buyer asks who the actual manufacturer is. The correct treatment is to declare the factory as the manufacturer, the trading entity as a separate organization with a stated relationship, and to make the distinction visible on the page as well as in the markup.
5. Wikidata: An Honest Assessment for a Mid-Sized OEM
Wikidata is the structured-data sibling of Wikipedia, and it is frequently recommended as an entity-building step. The assessment below is what we would tell a mid-sized manufacturer directly.
| Aspect | Reality for a mid-sized OEM in 2026 |
|---|---|
| Notability requirement | Wikidata items are expected to reference reliable, independent sources. A mid-sized OEM without press coverage in independent outlets will have any item nominated for deletion |
| Time to a stable item | 3-12 months, and it can be deleted at any point if the references are judged insufficient |
| Effect on citations | Modest. An engine that can already read your Organization markup and verify your certificate does not need a Wikidata item to cite you |
| Cost | Either significant internal time or a consultant fee, for an uncertain outcome |
| When it is worth it | When you already have independent press coverage, an industry award, or a notable founder — something a third party has written about |
| What to do instead | Spend the same effort on level 3 corroboration: a certification register listing, two trade association memberships, one trade publication mention |
The recommendation is therefore: skip Wikidata unless you already have the independent coverage that makes an item defensible, and spend that effort on the corroboration sources in level 3 instead. The corroboration sources are less prestigious and considerably more effective, because they are the sources an engine actually checks when it wants to verify a claim about a component manufacturer.
The 8 corroboration sources actually worth pursuing
Level 3 is where the remaining gain lives, and it is the level where effort is most often misallocated. The table below ranks eight corroboration sources by how much weight they carry and how long they take to obtain, for a pain patch manufacturer specifically.
| # | Source | Weight | Time to obtain | What to do |
|---|---|---|---|---|
| 1 | Certification body client register | Highest — an authority states the fact | Days, if you already hold the certificate | Ask your certification body whether your certificate number appears on its public client register, and get it listed |
| 2 | Business registry (national or provincial) | High — legal entity confirmation | Days, usually already public | Confirm your registered address and legal name on the registry match your site exactly |
| 3 | Trade association member directory | High — third-party endorsement plus entity data | Weeks to months, membership required | Join two associations relevant to medical devices or export trade, and ensure the directory entry carries your full address and product categories |
| 4 | Trade press mention with facts | Medium-high — indexed and durable | Months, relationship work | Pitch a trade publication on a substantive topic rather than a product announcement; the article that mentions your facilities or capacity is the one that corroborates |
| 5 | B2B directory with verified status | Medium — widely crawled | Days | Complete the verification process rather than leaving the profile unverified, and make the address and certifications identical to your site |
| 6 | Exhibitor listing at a major trade show | Medium — time-bound but strong | Per show | Ensure the exhibitor entry carries the full legal name and the correct website URL, which is often wrong because the organiser transcribes it |
| 7 | Industry award or shortlist | Medium — independent judgment | Months to years | Worth pursuing for its own sake, and it incidentally strengthens the entity |
| 8 | Academic or standards-body citation | High but rare | Years | Not something to pursue directly; it happens when a manufacturer contributes to a standard or a research collaboration |
Rows 1 and 2 are the two that most manufacturers can complete this month, and they carry the highest weight in the list. Both involve facts that already exist in a public record and simply need to be made findable and consistent with the company's own site. A manufacturer that completes rows 1 and 2 has addressed the verification problem that keeps it out of medical-category answers, which is a bigger gain than any amount of additional markup.
6. The 5-Step Implementation Sequence
The sequence below is ordered deliberately, and the order is the whole point. Each step is a precondition for the next.
| Step | Action | Output | Time |
|---|---|---|---|
| 1. Define the canonical facts | Write one page listing the legal name, the exact registered address, the phone number, the founding year, the employee count, and every certificate with its number, issuer, and dates | A single internal source of truth | 2 hours |
| 2. Reconcile the site | Compare every page and every language version against that page, and fix the differences | A site that states one set of facts | Half a day |
| 3. Reconcile external profiles | Compare the Alibaba, Made-in-China, LinkedIn, and any directory listings against the same page, and correct them | Consistent statements on the platforms a buyer checks | Half a day |
| 4. Declare the six entity types in JSON-LD | Add Organization, Place, Product, Certification, Person, and Service markup, with sameAs pointing only at URLs that genuinely describe you | Machine-readable declaration | 2-3 days |
| 5. Add corroboration | Get listed in the certification body's client register, join two relevant trade associations, and pursue one trade-press mention | External confirmation of the declared facts | Ongoing, 1-3 months |
The most common failure is to start at step 4. A team adds Organization markup in an afternoon, sees no change, and concludes that entity work does not pay. What they did was declare a set of facts that were internally inconsistent, which is not an improvement. Steps 1 to 3 are the ones that produce the gain, and they require editing rather than development.
7. The Consistency Audit That Comes Before Any Schema
The audit below is the practical version of steps 1 to 3. It takes a spreadsheet and two hours.
| Fact | Where to check | What counts as a mismatch |
|---|---|---|
| Legal name | Homepage, About page, footer, Alibaba profile, LinkedIn, certificate, invoice template | Any variation in spelling, punctuation, or the "Co., Ltd." form |
| Registered address | About page, Contact page, footer, certificate, business licence, Alibaba profile | A different street, a different district, or an English translation that does not match the Chinese address |
| Telephone | Contact page, footer, Alibaba profile, WhatsApp business profile | Two or more numbers in use, or a number that has been discontinued |
| Founding year | About page, homepage hero, press mentions, Alibaba profile | Different years in different places, which is surprisingly common after an acquisition or a rebrand |
| Certifications | Certifications page, About page, product pages, Alibaba profile | A certificate listed that has expired, or a scope statement that differs between pages |
| Product specifications | Product pages, datasheets, Alibaba listings, other language versions | A different minimum order quantity or lead time on a different page |
The founding-year row catches more companies than any other, and it is worth a specific note. A manufacturer founded in 1989 that added a new production facility in 2011 sometimes describes itself both ways, depending on which page and which year it was written. To a machine, that is two different founding dates, which is a contradiction about the entity's identity. Pick one — the original founding date — and use it everywhere.
8. Worked Example: From 4 Conflicting Addresses to One Entity
The table below is an anonymised pattern from a mid-sized transdermal patch manufacturer. It shows the state before and after the consistency audit.
| Fact | Before (4 sources) | After |
|---|---|---|
| Legal name | Three variants: with "Co., Ltd.", without it, and an older trading name on one directory | One form, used on every page and every profile |
| Address | Two addresses in use: a registered office in the city and the factory in the county, with no explanation of the difference | Both declared, with the relationship made explicit: registered office, and the manufacturing site |
| Phone | Three numbers, two of which were discontinued | One number, plus a WhatsApp number, both current |
| Founding year | 1989 on the About page, 2003 on the Alibaba profile (the year of a corporate restructuring) | 1989, with a one-line note that the current corporate form dates from 2003 |
| ISO 13485 | "Certified" on the About page, with no number; the number existed only in a PDF | Number, issuer, scope, and validity dates published, and the number checkable on the issuer's register |
| MOQ | 5,000 on the English pages, 3,000 on the Spanish pages, no explanation | 5,000 for a new SKU and 3,000 for a repeat order, stated identically in every language |
The resolution of the address question is the pattern worth copying. The company had two genuine addresses — a registered office and a factory — and had been alternating between them depending on which page was written when. The fix was not to pick one and delete the other; it was to declare both and to state the relationship, which is factually accurate and gives the machine a coherent model. A manufacturer with a registered office and a manufacturing site is a normal structure, and declaring it as such removes the ambiguity without misrepresenting anything.
9. FAQ: Knowledge Graph Questions Manufacturers Ask Most
9.1 Do I need a knowledge graph, or is schema enough?
Schema markup is the declaration; the knowledge graph is what an engine builds from it plus everything else it can read. You cannot build a graph directly — you can only make your declarations consistent and corroborated. So the practical answer is: implement schema, but only after the facts are consistent, and treat external corroboration as the step that makes the graph reliable.
9.2 How do I know whether an engine has an accurate entity for my company?
Ask an AI assistant a direct question about your company — who you are, where your factory is, what certifications you hold — and compare the answer against your canonical facts page. Incorrect or partial answers reveal which facts the engine cannot confirm. Repeat across two or three engines, because their source pools differ and each will fail on different facts.
9.3 Is Wikidata worth doing?
Only if you already have independent press coverage, an award, or a notable founder — something a third party has written about. Without that, an item will be nominated for deletion, and the effort is better spent on certification register listings and trade association memberships, which are less prestigious and considerably more effective for a component manufacturer.
9.4 What if my factory and my trading company are separate legal entities?
Declare both, and state the relationship explicitly. Use different entity nodes with different names, and describe the relationship in the markup and on the page — for example, that the trading entity is the export arm of the manufacturing entity. Merging them into one node misrepresents the structure; leaving the relationship undeclared leaves the machine to guess, and it will usually guess wrongly.
9.5 How many sameAs links should I have?
Three to six, all of them URLs that genuinely describe the entity with matching facts. The number matters less than the quality: three verified links to a certification register, a trade association, and a business registry are worth more than ten links to social profiles that repeat the same self-published text.
9.6 Does entity work help in Chinese-language search as well?
Yes, and the weighting differs. For Baidu and Chinese-language AI surfaces, entity signals from Chinese registries and Chinese-language sources carry more weight than Western third-party corroboration. The consistency audit is identical; the corroboration step should target the registries that the relevant engines actually check.
9.7 How long does it take to see an effect?
Level 1 and level 2 changes are usually reflected in AI answers within 2-8 weeks, depending on crawl frequency. Level 3 corroboration is slower — 1-3 months for a listing to appear and be picked up. The fastest observable change is usually the disappearance of wrong facts, because correcting a contradiction resolves an ambiguity immediately.
9.8 Can I do this without a developer?
Steps 1 to 3, yes — they are editorial work on pages and directory profiles. Step 4 requires someone who can add a script tag to the site's templates, which is usually a developer or the person who manages the website platform. Step 5 is relationship work that requires no technical skill at all.
9.9 What is the single most common entity error you see?
The factory address. It appears in one form on the site, another in the business licence, and a third on the B2B platform, often because the Chinese address has been transliterated differently each time. To a machine, three different addresses for one company is a contradiction about where the company is, and it produces exactly the "details could not be verified" outcome that keeps a manufacturer out of an answer.
9.10 Can Kangdi show how our entity looks to an engine?
Yes. We run the direct-question test described in 9.2 against three engines and compare the answers against your canonical facts. Send us your domain and your registered address, and we will return the answers the engines currently give about your company, the specific inconsistencies, and the list of facts that no engine can confirm — which is usually the shortest path to knowing what to fix first.
10. About Kangdi Medical and Our Entity Declaration
Henan Kangdi Medical Devices Co., Ltd has manufactured transdermal patches since 1989 from a GMP-certified facility in Zhoukou, Henan, China, and we ship to 200+ brand buyers and distributors across 40+ countries. Our entity facts are declared consistently across the site and the platforms a buyer checks, and we run the consistency audit quarterly.
Three steps to get your entity check. (1) Email us with "Entity Check" as the subject line, including your domain and your registered address. (2) We will ask three AI assistants the direct questions about your company and record what they say. (3) We will send back the three answers side by side, the inconsistencies between them, and the facts no engine can currently confirm.
- Step 1: Email export@kangdimedical.com with "Entity Check" as the subject line, including your domain and registered address.
- Step 2: Receive the three-engine answer comparison within 3 business days.
- Step 3: Get a prioritised list of the facts to reconcile first, starting with the address and the certificates.
Tell us your domain and your registered address, and we will send the entity check within 3 business days.