Structured data for AI. Google writes that it won’t get you into the answers. It tells them who you are.
There’s a trade in “schema for ChatGPT”, yet Google writes that generative search needs no special markup. Here is what structured data is actually for, with an annotated JSON-LD template in Word and PDF.
Structured data isn’t required for generative AI search, and there’s no special schema.org markup to add. Google Search Central says so in its guide updated on 10 July 2026, which still recommends it for SEO and for rich results.
For a small business its main job is to say who you are. Organization with legal name, VAT number, address and profiles; Person for the owner or the author; Article with the author and the dates. Google lists no required properties for Organization.
It doesn’t promise citations. The OpenAI, Anthropic and Perplexity documents read for this page say nothing about it; for Microsoft there is a report from a conference.
The annotated JSON-LD template is below, in Word and PDF, with no email address required.
Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.
The Organization and Person JSON-LD template, to download
A sheet with the Organization, Person and Article properties, what goes in each and where it must appear on the page, the JSON-LD block to paste into the home page and eight checks before you publish. Word to fill in, PDF to print. You’ll need the company registry extract, the list of official profiles and access to the site code or the CMS.

| File | Contents | Link |
|---|---|---|
| Wordto fill in, in Italian | The three property tables, the annotated JSON-LD block, the eight checks. | |
| PDFto print, in Italian | The same sheet, to fill in by hand with the real values. |
| Property | What goes in | Where it appears on the page |
|---|---|---|
| name | The name the business trades under, matching the site name. | Header, footer |
| legalName | The registered company name, as in the registry extract. | Footer, privacy notice |
| logo | An image of at least 112x112 pixels that works on a white background. | Header |
| vatID | The VAT number with the country prefix: IT and the number for Italy. | Footer |
| address | Street, postcode, town, province and country; more sites, more addresses. | Contact |
| telephone, email | The phone number with the international prefix and an inbox someone reads every day. | Contact |
| sameAs | The official profiles: LinkedIn, social accounts, Google profile, review sites. | Icons in the footer |
| founder, worksFor | The link between business and owner, with the same @id. | About |
| author | On articles: one author per field, with a url pointing to their page. | Byline |
{ "@context": "https://schema.org",
"@graph": [
{ "@type": "Organization", "@id": "https://www.example.com/#company",
"name": "Example", "legalName": "Example Ltd", "vatID": "GB123456789",
"sameAs": [ "https://www.linkedin.com/company/example/" ],
"founder": { "@id": "https://www.example.com/#owner" } },
{ "@type": "Person", "@id": "https://www.example.com/#owner",
"name": "First Last", "url": "https://www.example.com/about",
"worksFor": { "@id": "https://www.example.com/#company" } } ] }
On 7 August ChatGPT replied that it couldn’t find a Mattia Esposito linked to itria.io as founder. The structured data was already on the site. I put it like this on 21 September: “I’d written the llms.txt, the schema, all the nice touches. As far as AI was concerned, I didn’t exist.” The markup alone didn’t get me found, as I explain in the full post.
On 26 September I measured the websites of the 121 food producers listed in ICE’s directory for foreign buyers. All 121 let in Google, Bing and the assistants’ search crawlers. The gap was on the page: 53 sites, 43.8%, used no structured data to say who the company is. The figures and the method are in the study of food producers’ websites.
Anyone can read the structured data of itria.io in the source of every page. The home page declares the business as a ProfessionalService, a subtype of Organization, with VAT number, address and linked profiles; the About page is a ProfilePage holding the Person; every article points to its author with the same @id. I don’t credit this for the site’s appearances in AI answers, since Google says it isn’t needed. Its job is to make the facts about who stands behind the site add up.
Does structured data matter for AI search? What Google says in 2026
No, you don’t need it to appear in the AI features of Google Search. In its May 2026 guide to generative AI features, updated on 10 July, Google lists overfocusing on structured data among the things to stop chasing, and writes: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.”
The sentence ends with advice: keep using it as part of your SEO, because it makes a page eligible for rich results in Search. The Italian version of the page, updated on 16 September 2026, says the same.
Google’s broader view of GEO, which it still treats as SEO, is covered in the guide to generative engine optimization; the labels doing the rounds, GEO, SEO and AEO, are compared on the page about GEO vs SEO vs AEO.
In its introduction to structured data Google explains what it uses it for: understanding the content of a page and gathering information about the world, such as people, books and companies. It recommends JSON-LD, also accepts Microdata and RDFa, and prefers fewer properties that are complete and accurate to many that are incomplete or wrong.
Is there a schema markup for ChatGPT?
There’s no schema for ChatGPT, and no public OpenAI page read for this guide says ChatGPT search uses schema.org to pick its sources. For ChatGPT search OpenAI documents the OAI-SearchBot crawler, as the guide on how to appear in ChatGPT explains.
For shopping OpenAI uses a product feed with fields of its own, which merchants submit separately: the published specification doesn’t mention schema.org. In the documents read, Anthropic and Perplexity say nothing about structured data. Markup sold as “for AI” is a new label on the same old vocabulary.
| Who | What they say about structured data and AI | Source |
|---|---|---|
Not required for generative search, no special markup; useful for SEO and rich results. | Guide to AI features, 10 July 2026 | |
Organization helps Google understand administrative details and tell the business apart in results. | Organization, 8 September 2026 | |
| Schema.org | The vocabulary shared by search engines; version 30.1 of 16 September 2026. | schema.org, releases |
| Microsoft | NLWeb uses Schema.org and RSS to give sites a natural-language interface for people and agents. | Microsoft Source, 19 May 2025 |
| Bing | “Schema helps our LLMs”: a third-party report of a conference talk. | Search Engine Roundtable, 20 March 2025 |
| OpenAI | No page read mentions schema.org for search; for shopping, a product feed of its own. | Product feed specification |
| Anthropic, Perplexity | No statement found. | Public documentation |
Organization schema for a small business: the properties that count
Organization goes on the home page and tells Google who the business is: name, legal name, address, contact details, tax numbers and profiles. According to Google it helps it understand administrative details and tell the organisation apart from others with a similar name; some properties can influence the logo and the knowledge panel shown in results, without Google promising to show them.
Google lists no required properties: “There are no required properties; instead, we recommend adding as many properties that are relevant to your organization.” For a small company the relevant ones are almost always name (matching the site name), legalName, url, logo, address, telephone with the international prefix, email, vatID and sameAs.
The logo must be at least 112x112 pixels, reachable by the crawler and readable on a white background. An online shop can use the OnlineStore subtype, with contactPoint and a returns policy; a professional firm has subtypes such as ProfessionalService, the one itria.io uses. The more precise subtype describes the business better and leaves the rest unchanged.
VAT number, legal name and address: vatID, legalName and address
The VAT number goes in vatID, with the country prefix. Schema.org defines vatID as the value-added tax ID of the organisation or person “with national prefix”, and its official example is Italian: IT123456789. Google calls it a trust signal, because readers can look the business up in public VAT registries.
A tax code that differs from the VAT number goes in taxID, which Google says must match the country of the address. Address is a PostalAddress with street, postcode, town, region and country; a business with several sites can list more than one. legalName is the registered name, while name stays the one the business trades under.
Tax numbers vary from country to country, and so does the public service a foreign buyer uses to check them. The table brings together the schema.org and Google rules with the official checking services.
| Country | What goes in vatID and taxID | Where it can be checked |
|---|---|---|
| Italy | vatID: IT followed by the partita IVA. taxID: the codice fiscale, if different. | The European Commission’s VIES |
| Other EU countries | vatID with the country prefix; taxID from the country of the address. | VIES, which queries the national VAT databases |
| United Kingdom | vatID with the UK VAT registration number. | GOV.UK, Check a UK VAT number: validity, name and address |
| United States | There’s no federal VAT; businesses are identified for tax by the EIN, issued by the IRS. | No service named by Google |
sameAs: linking your site, LinkedIn, Google profile and reviews
sameAs lists the external pages that confirm the identity of the business or the person: social profiles, Google profile, review sites. Schema.org defines it as the URL of a reference web page that “unambiguously” indicates the item’s identity; Google accepts several addresses and also uses it to tell article authors apart.
Official, up-to-date profiles under the same name as the site are what count. An abandoned profile with an old name creates the very doubt the markup is meant to remove. On itria.io, sameAs points to the company’s LinkedIn page, its Google profile, Facebook and Instagram.
Person schema and article authors
Person describes the owner or the author of the articles, and links to the business with worksFor. Only a few properties are useful: name, jobTitle, url pointing to the About page, sameAs pointing to the LinkedIn profile, and knowsAbout, which schema.org says indicates a known topic and suggests possible expertise.
If the author’s page sits on your own site, Google advises marking it up as a ProfilePage, the type designed for sites where people or organisations share a first-hand perspective: it has one required property, mainEntity, which holds the Person. Articles can point to that page as the author’s url.
Google also has Search profiles, which gather a person’s or a publisher’s content from their site and social accounts on a single page. You connect one with a badge or a link, not with schema.org: identity on Google runs through there too.
Article schema: author, datePublished and dateModified
For Article, NewsArticle and BlogPosting Google lists no required properties, but it sets clear rules for the author. Every author shown on the page belongs in the markup, one per author field; two names in the same field are a mistake. Each author has an @type, Person or Organization, and a url or sameAs. In Google’s words: “Google can understand both sameAs and url when disambiguating authors.”
datePublished and dateModified use ISO 8601, with the time zone. The modified date changes when the content changes, and should match the date shown on the page. Publisher points back to the Organization on the home page with the same @id, which closes the loop.
What Microsoft says: Bing, Copilot and NLWeb
The Microsoft pages read for this guide set no rule on how Copilot uses structured data to cite sources. On 20 March 2025 Search Engine Roundtable relayed a report of a talk by Fabrice Canel, Principal Product Manager at Bing, at SMX Munich: according to an attendee’s account, which Canel commented on without denying, schema helps Microsoft’s models understand content. This is a second-hand source, and should be read as one.
The official piece is NLWeb, the open project Microsoft announced on 19 May 2025, conceived by R.V. Guha, creator of RSS, RDF and Schema.org. It uses Schema.org, RSS and the other data sites already publish to give every site a natural-language interface that agents can query too.
NLWeb helps your own site talk; it won’t get you into Copilot, which still depends on the Bing index. How to get in is covered in the guide on how to appear in Copilot.
The 121 food websites. In the study of 26 September 2026, 53 sites out of 121, 43.8%, used no structured data to say who the company is; 9, or 7.4%, had product structured data on the pages read. All of it is in the study of food producers.
The missing email. Cloudflare’s obfuscation removed the address from the HTML and put it back with a script: to an assistant, the contact page had no contact details. The case, and the same fault on 14 of 121 producers’ sites, is in the piece on the email AI assistants can’t see.
The site’s results. Between 11 October and 7 November 2026 Bing’s AI Performance report counted 268 citations of itria.io pages in Copilot, and Search Console 3,550 impressions in AI Overviews and AI Mode. The markup was already there in August, when the site had 24 impressions in 28 days.
Costly mistakes: hidden markup, fake reviews, manual actions
One rule matters most: structured data may only describe what the reader can see on the page. Google’s general guidelines are blunt about it: never mark up content the readers of the page can’t see. The same goes for irrelevant or misleading content, such as fake reviews.
Breaking these rules can bring a structured data manual action, which removes eligibility for rich results. Google adds that such an action doesn’t affect how the page ranks in web search. And even valid markup is no guarantee that rich results will appear.
The most common mistakes are practical ones: a name that differs from the site name, a logo that doesn’t load, a VAT number with no prefix, a sameAs pointing to a closed profile, an author shown on the page but missing from the markup. The downloadable sheet has a check for each.
How to test structured data: Rich Results Test and URL Inspection
While you write the code, use Google’s Rich Results Test; after publishing, the URL Inspection tool in Search Console. Google recommends checking markup during development and then following the rich result status reports, because templating or serving issues can break it once the page is live.
The schema.org validator also checks the syntax of types Google doesn’t use. How to open Search Console and read URL Inspection is in the Google Search Console guide; for Bing and IndexNow there’s the guide to Bing Webmaster Tools. The llms.txt file, often sold alongside schema, has a page of its own: llms.txt, what it is and whether you need it.
Markup says who stands behind a site. Getting named when a buyer asks an assistant who does that kind of work is a further step, covered in the guide on how to get your brand cited by ChatGPT. How I build it for a real business is on the page about presence in search engines and AI assistants. What to write on the page this markup describes is in the guide to the About us page.
Questions and answers
Does structured data help you appear in ChatGPT?
No OpenAI document read for this page says ChatGPT search uses schema.org to pick its sources. What counts is letting the OAI-SearchBot crawler in, being in the search engines’ indexes and publishing pages with facts that can be checked.
For Google the answer is in writing: structured data isn’t required for generative AI search.
Is there a schema markup for AI?
No. Google Search Central writes that there’s no special schema.org markup to add for generative search. The vocabulary is the one used for rich results: Organization, Person, Article, Product and the other schema.org types.
What structured data should a small business website have?
Organization on the home page, with name, legal name, logo, VAT number, address, contact details and profiles; Person for the owner or for the author of the articles; Article on guides, with author and dates. Google lists no required properties for Organization and recommends every relevant one, as long as it is complete and accurate.
Where does the VAT number go in structured data?
In the vatID property of Organization, with the country prefix: for Italy, IT followed by the number. Google calls it a trust signal, because readers can look the business up in public VAT registries. A separate tax code goes in taxID.
How do you test structured data?
With Google's Rich Results Test while you write the code, and with the URL Inspection tool in Search Console after publishing, to see whether Google has found the markup. The schema.org validator checks the syntax of types Google doesn't use as well.
Notes on sources
- Google Search Central, Optimizing your website for generative AI features on Google Search, updated on 10 July 2026 (Italian version of 16 September 2026): the quoted sentence on structured data. Content licensed under CC BY 4.0. Read on 10 November 2026.
- Google Search Central, introduction to structured data (10 December 2025), Organization, Article and ProfilePage (8 September 2026): properties, logo, vatID, authors. Read on 10 November 2026.
- Google Search Central, general structured data guidelines (10 July 2026) and Search profiles (16 September 2026); Rich Results Test.
- Schema.org, Organization, Person and release list (30.1, 16 September 2026). Read on 10 November 2026.
- Microsoft Source, Introducing NLWeb, 19 May 2025, and the project code. Search Engine Roundtable, report of Fabrice Canel’s talk, 20 March 2025: a secondary source, with no official transcript.
- OpenAI, product feed specification: no mention of schema.org on the page read on 10 November 2026.
- Your Europe, Check a VAT number (VIES); GOV.UK, Check a UK VAT number; IRS, Get an employer identification number. Read on 10 November 2026.
- The figures on the 121 sites come from Itria’s study of 26 September 2026, data under CC BY 4.0. The itria.io figures come from Search Console (generative AI report) and Bing Webmaster Tools, for the window from 11 October to 7 November 2026.
Markup says who you are. Getting chosen when a buyer asks is the work that comes next. Positioning starts from the market.
Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.