Structured data, entity markup and product feeds for AI surfaces

0:00 Structured data and feeds for AI

Let's settle one of the most repeated claims in AI search. Add schema, and AI will cite you. Is it true? Mostly, no. But structured data does matter, in a narrower and more useful way than the hype suggests. In this lecture you'll learn where machine-readable data genuinely helps, which rich results still exist in twenty twenty-six, how to build a clean entity graph in JSON-LD, and why product feeds may matter more than markup for anyone who sells online.

0:35 The honest position

Here's the honest position. Google's twenty twenty-six guide for its AI features says no special schema is needed, and no provider has shown that markup alone triggers citations. The defensible claim is this: accurate machine-readable data reduces ambiguity. Think of it like the label on a medicine box. The label doesn't make the medicine work, but without it, people make mistakes. Structured data tells systems who you are, what you sell, what it costs, and whether it's in stock, so they're less likely to get it wrong.

1:13 Three layers of machine-readable facts

Think in three layers. Layer one is on-page JSON-LD, the schema dot org markup in your HTML. Layer two is feeds, structured product catalogues you send to platforms like Google Merchant Center, and, for eligible merchants, OpenAI's product feeds for shopping in ChatGPT. Layer three is profiles and knowledge sources: your Business Profile, Bing Places, and official profiles. Different surfaces lean on different layers, so the goal is the same facts in all three.

1:45 Rich results in 2026

Now, which rich results still exist? Google has been narrowing them for years. HowTo rich results went in twenty twenty-three. Seven little-used types, including course info, claim review and estimated salary, were phased out from mid twenty twenty-five. And FAQ rich results stopped appearing on the seventh of May, twenty twenty-six. Still alive: product and merchant listings, review snippets within policy, articles, recipes, video, events, breadcrumbs, organisation details, local business and job postings. Leaving old FAQ markup in place does no harm. Promising a client an FAQ rich result in twenty twenty-six does.

2:25 Hands-on: a connected JSON-LD graph

Hands-on, part one: an entity graph. In the lesson text you'll find a JSON-LD block with an at-graph containing two entities. An Organization with a stable at-id, name, logo, same-as links and address. And a Product whose brand points to that same organisation at-id, with an Offer holding price, currency, availability and shipping destination. The stable identifiers are the clever part. Every page describes the same organisation in exactly the same way, so systems can join the dots. And the golden rule: generate this from the same database fields that render the visible price.

3:06 Feeds: the direct line to AI shopping

Hands-on, part two: feeds. For e-commerce, feeds often matter more than markup. Google Merchant Center feeds power Shopping and product information across Google surfaces. OpenAI documents product feeds so ChatGPT can surface products in its shopping experiences, and it co-developed the Agentic Commerce Protocol with Stripe for checkout inside ChatGPT. Availability varies by country, so check current merchant documentation before you plan. The rule across all of them is consistency: the same name, price, currency and stock status everywhere.

3:40 Example 1: duplicate Organization markup

Worked example one, simple. A bakery in Islamabad runs an SEO plugin and a theme that both output Organization markup, with two different phone numbers. Validators pass both blocks, but systems now see conflicting facts. The fix: remove the theme's block, keep one Organization entity with a stable at-id, and make sure its phone number matches the Business Profile. Five minutes, one less source of wrong answers.

4:09 Example 2: Jeddah abaya brand (illustrative)

Worked example two, with illustrative details. A Jeddah abaya brand finds assistants quoting last season's prices. The cause is a chain of small problems. The theme hard-codes the JSON-LD price. The Merchant Center feed updates weekly, but prices change daily during Ramadan sales. And a marketplace listing is months out of date. The fix: JSON-LD generated from live data, feeds updated several times a day during promotions, marketplace listings synced, and a weekly check of five products across page, markup, feed and assistant answers.

4:46 Validate and avoid

How do you validate all this? Use the Rich Results Test for Google eligibility on a rendered page. Use the Schema Markup Validator for general validity. Use your crawler's extraction to compare the JSON-LD price with the visible price on every template. Check Merchant Center diagnostics for mismatches. And ask assistants for your product's price and stock, logging any errors as accuracy issues. Common mistakes: selling schema as an AI ranking factor, markup that disagrees with the page, promising deprecated rich results, and updating the website but forgetting the feeds.

5:25 People are entities too

Let's add one more layer that often gets forgotten: people. Founders, authors and experts are entities too. An author page with Person markup, a consistent name spelling, a short bio, credentials, and links to official profiles helps systems attribute expertise correctly. Choose one transliteration of a name, for example Mohammed or Muhammad, and use it everywhere. Link the Person to the Organization with the same kind of stable at-id. It's a small amount of work, and it makes it far less likely that an assistant confuses your expert with someone else who shares the name.

6:06 Watch me do it: one product, four sources

Watch me do it. I'll run the consistency grid for one product, end to end. Step one: the page. I open the large brass lantern product page and note what a customer sees: eight thousand five hundred rupees, in stock, ships within Pakistan. Step two: the markup. I view source, search for application slash ld plus json, and copy the block into the Schema Markup Validator. It's valid. But the price field says seven thousand nine hundred. Step three: the feed. I open Merchant Center, search the product ID, and see eight thousand five hundred, in stock. Step four: an assistant. I ask, how much is the large hand-beaten brass lantern from this brand? It answers seven thousand nine hundred and cites a marketplace listing. Now I have a four-column grid: page, markup, feed, assistant. Two cells disagree. Step five: I trace the cause. The theme hard-codes the price in the JSON-LD template, and the marketplace listing hasn't synced since a price rise. I write two tickets: generate JSON-LD from the live price field, and sync the marketplace listing. Then I set a reminder to re-run this grid for five products every week during the next sale.

7:32 Pitch it honestly

And a note on Microsoft and other engines. Bing's team has said publicly that schema markup helps its systems understand content, and Bing Webmaster Tools reports on markup it finds. That's consistent with the ambiguity-reduction view: markup helps machines understand, which helps them represent you accurately. It is not a promise of citations. When you pitch structured data to a client, pitch accuracy and consistency across every surface, plus the rich results that still exist. That's a promise you can keep.

8:07 Recap and try this now

Recap. Schema isn't a citation switch, but accurate machine-readable facts reduce ambiguity everywhere. Check which rich results still exist before promising them. Build a connected entity graph with stable identifiers. And if you sell online, treat feeds as a first-class channel. Try this now. Pick five products or services. Write down the price, currency and availability shown on the page, in the JSON-LD, in your feed, and in one assistant's answer. Every mismatch is a ticket.