Knowledge graph entity markup: Schema and validation for Organization/Product/Person

Required properties for three entity types, a three-step validation process, and common mistakes

Knowledge graph entity markup: Schema and validation for Organization/Product/Person

Bottom line first: To get AI engines to treat your company as a “recognizable entity,” you need to mark up all the required properties of three entity types on your site using Schema.org structured data: Organization, Product, and Person, and link your authoritative profiles on LinkedIn, Wikipedia, Crunchbase, and other sites through the sameAs field. One counterintuitive point: marking up only your own site won’t make models recognize you right away. Single-site markup is just your ticket in; you get added to the entity graph only when the same information appears consistently across multiple sources. After marking up, it usually takes 2 to 8 weeks to see changes in AI recognition, not same-day results.

1. Why AI often treats your company as a “vague phrase”

When a user asks ChatGPT “What do you think of crater, a Hangzhou company that builds independent sites for going global?” and the AI can’t answer or mixes it up with something else, the root cause is usually that it never recognized “crater” as an entity with clear properties. It has read the word, but doesn’t know the company’s full name, what it does, which site is its official one, or who founded it.

Entity recognition is different from ordinary keyword scraping. Keyword matching only requires the words to appear literally; entity recognition requires a set of mutually corroborating properties: full company name, official website, logo, founding date, associated people, associated products. The more complete this set of properties is and the more consistent it is across sites, the more confident the model is in treating it as a “thing” rather than a few characters. For how entities are built into AI answers, see Knowledge graphs and entity markup.

Recognition status	AI Behavior	Root cause
Not recognized at all	Answers off-topic or ignores	No entity markup on-site, no sources off-site
Treated as an ordinary phrase	Mentioned literally but properties are wrong	Incomplete properties, inconsistent across sites
Recognized as an entity	Correctly associates official site and business	Consistent properties across three sites + authoritative sources

2. Required property checklist for the three entity types

Each of the three entity types has a set of properties that must be filled in. Organization covers the company itself, Product covers specific products, and Person covers people who can represent the company. Don’t be greedy; get the required fields right first, then consider extended properties.

Organization needs at minimum name, legalName, url, logo, sameAs, and foundingDate. Product needs at minimum name, image, description, brand, and sku. Person needs at minimum name, jobTitle, affiliation, and sameAs. The sameAs field is the most critical of all: it connects your on-site entity to authoritative off-site profiles, so the model dares to “confirm” it’s the same company.

Entity type	Required properties	Suggested extensions	Deployment location
Organization	name, legalName, url, logo, sameAs	foundingDate, address, description	Site-wide homepage + About page
Product	name, image, description, brand	sku, category, material, specs	Each product detail page
Person	name, jobTitle, affiliation, sameAs	alumniOf, knowsAbout	Team/expert page

A real-world example: you’re a B2B company making industrial sensors. Add Organization markup to your homepage, listing three sameAs links: your website, your LinkedIn page, and your Crunchbase page. Add Product markup to each product detail page, filling in model number, measurement range, accuracy, and material. Add Person markup to the team page for your founder and lead engineer. Once these three steps are done, ask an AI “What’s the accuracy of X company’s temperature sensor?” and the odds it gets it right go up noticeably.

3. How to write Schema and where to deploy it

Embed it in JSON-LD format in the page head; it’s easier to maintain than microdata or RDFa. Put Organization on the homepage and About page, Product on each product detail page, and Person on the team page. The key is that your short name, full name, and English name must be exactly the same across all three markup types. If the short name doesn’t match the full name, the model treats them as two separate entities.

Don’t let a site-wide template auto-fill a pile of empty properties. An empty property is worse than no property, and a wrong property pollutes the entity. Every property you add must also appear in the visible content on the page, so what’s machine-readable matches what’s human-readable. Put JSON-LD in the page head, and only one primary markup block of the matching type per page. Don’t stuff Organization, Product, and Person all onto the same product page, or the model will misidentify the main entity. For a B2B checklist you can act on, see Structured data Schema: B2B website markup checklist.

Attribute	Writing essentials	Common mistake
name	Match the page title	Mixing short and full names
sameAs	List real, authoritative pages	Filling in your own social accounts for show
logo	Square or with a clear aspect ratio	Using a tiny transparent icon
url	https, consistent with or without trailing slash	Mixing http and https
brand	Point to Organization	Writing brand as a string inside Product

4. Validation: syntax, indexing, AI recognition, in three steps

Step one: run the syntax through Google Rich Results Test and confirm there are no parsing errors. Step two: check the structured data report in Google Search Console and confirm the markup is crawled and indexed with no errors. Step three is the real goal: take a fixed set of questions to ChatGPT and Gemini and see whether they can correctly state your company’s full name, business, and products.

Don’t skip step one and jump straight to step three. If the syntax fails, everything after it is wasted. And don’t expect results the next day once the syntax passes; engines take time to adopt structured data, usually 2 to 8 weeks. During that time, don’t keep changing the markup, because every change resets the observation window. To see how validated entities work with trust signals, see Trust pages and AI credibility.

Validation step	Tools	Pass criteria	Expected time
Syntax check	Rich Results Test	No errors, no warnings	Same day
Crawling and indexing	Search Console	Markup recognized and counted	1 to 4 weeks
AI recognition	Ask AI a fixed set of questions	Correctly states full name and business	2 to 8 weeks
Consistent across sites	Manually check all three sites	Names and attributes match	One-time

5. Pitfalls: the three most common entity markup mistakes

The first mistake is adding Organization only to the homepage and leaving product pages empty. The result: AI recognizes your company but can’t say what you sell. The second mistake is filling sameAs with only your own Weibo and WeChat accounts. Those don’t count as authoritative sources, and the model won’t accept them. The third mistake is adding Person markup to every employee, which dilutes the weight of your key people.

Also note that entity markup isn’t a one-time job. If the company moves, changes its logo, shifts its product line, or has a founder change, the Schema has to be updated in sync, or the model will answer users with old attributes. Product models and precision specs especially: after a product page changes, refresh the Product markup right away, within two weeks at most. Your monthly site maintenance checklist should include “check entity markup” as a line item, see Six Monthly Tasks for Site Maintenance.

Mistake	Consequence	The right way
Organization only	It knows the company, not the products	Add Product on product pages too
sameAs points to your own social accounts	The model doesn’t trust them	Use LinkedIn / Wikipedia / Crunchbase
Person markup on everyone	Dilutes your key people	Only founders and experts
Attributes don’t match visible content	Flagged as inconsistent	Keep machine and human readings aligned
Business changes, markup doesn’t	The model uses old attributes	Check it during monthly maintenance

FAQ

If I add Schema only on my own site, will AI recognize my company?

No. Markup on a single site is just the entry ticket. AI builds entities from consistent appearances across multiple sources: you mark up Organization on your site, but you also need profiles with the same name and business on LinkedIn, Wikipedia, and Crunchbase, linked together with the sameAs field. When multiple independent sources say the same thing, the model has the confidence to build it into an entity. If you only mark up your own site, it usually takes 2 to 8 weeks to see any change in recognition, and only if there are matching sources off-site.

What links should the sameAs field actually contain?

Your company’s official profiles on authoritative third-party platforms, not your own pages. Priority: Wikipedia entry, Crunchbase company page, LinkedIn company page, well-known industry directories. Don’t include your own Weibo, WeChat account, or subpages of your site, because the model doesn’t treat those as independent confirmation. Three to five is usually enough; too many dilutes the signal. Every link must be a real, accessible official profile whose content matches yours.

Which pages get Organization, Product, and Person markup?

Organization goes on the homepage and About page, covering the company itself. Product goes on every product detail page, one per product, linking model, specs, and brand back to the Organization. Person goes on team or expert pages, and only for founders, lead engineers, and industry experts who represent the company, not for regular employees. On the homepage, the three types reference each other via JSON-LD to form a connected graph.

After writing Schema, how do I verify it’s working?

Three steps. First, run the Google Rich Results Test to check the syntax and make sure there are no errors or warnings. Then open the structured data report in Search Console to confirm the markup has been crawled and indexed. Finally, ask ChatGPT and Gemini a fixed set of questions to see whether they can correctly state your full company name, what you do, and your products. The first two steps take anywhere from the same day to four weeks; the third usually takes two to eight weeks to show results. Don’t keep changing the markup in between.

What happens if my short name and full company name don’t match?

The model treats them as two separate entities. Say your website uses the short name “Company A” while LinkedIn uses the full name “Company A Technology Co., Ltd.” If the two don’t line up in your Schema, the model may decide they aren’t the same company, and your recognition weight gets diluted. The right approach is to stay consistent across your whole site: the short name, full name, and English name should be written exactly the same way on your homepage, About page, product pages, and external profiles. Put the full name in the legalName field and the common name in the name field.

How often should I maintain entity markup?

Check it once a month and add it to your website’s monthly maintenance checklist. Events that should trigger an update: a new logo, an office move, a change to your product line, a change in founders, or a site redesign. If these changes aren’t synced to your Schema, the model will answer users with outdated attributes. This matters most for product models and specs: update the Product markup as soon as a product page changes, or AI will cite stale parameters.

Final thoughts

Entity markup isn’t technically hard. The hard part is staying consistent over the long term. Writing the JSON-LD takes half a day, but keeping your short name, full name, English name, website, LinkedIn, and Wikipedia unchanged across all channels for three years takes discipline. What AI recognizes in an entity is exactly this kind of stability across time and sources. You can’t rush it, and there are no shortcuts.

Three steps to get it done: in week one, complete the required Organization fields on your homepage and About page, and list 3 to 5 authoritative profiles in sameAs. In month two, add Product markup to all your product detail pages. In month three, add Person markup for your founders on the team page. Then validate the syntax with the Rich Results Test and check AI recognition with a fixed set of questions. For building out external sources to go with this, see Off-site factual sources, and for the overall method, see GEO Practice.

One line worth remembering: Schema isn’t written for search engines to read, it’s an ID card written for models to recognize who you are. Marking up a single site is just signing up; being consistent across three sites is what gets you an entity registration.