Does structured data help in ChatGPT, Gemini and Copilot? What Google, Microsoft and the studies say.

What structured data is

It is a short block of code in a page, usually in a format called JSON-LD. It labels what the page is about with words from a shared list at schema.org. A product page can label its price and whether it is in stock. A company page can label its name, address and logo. Visitors do not see it. Search engines read it.

What Google says

Google's guide to AI features says There's also no special schema.org structured data that you need to add. (Google).

Google still uses structured data across search. In May 2025 it called structured data useful for sharing information about your content in a machine-readable way that our systems consider. It added that everything in the markup should also be visible on the page (Google).

What Microsoft says

Microsoft is warmer. In October 2025 its Bing team wrote, Schema is a type of code that helps search engines and AI systems understand your content. The same post warned that nothing guarantees selection (Microsoft).

Copilot answers from Bing's search, so what helps Bing read a page can reach Copilot's answers too.

What OpenAI says

OpenAI has said little in public about structured data in web pages. For shopping in ChatGPT, it asks shops to send their product details in a product feed, a file they submit to OpenAI directly (OpenAI).

What the studies found

Pages that AI assistants cite are almost three times as likely to carry JSON-LD as pages they do not cite, Ahrefs found. That could mean the code helps. It could also mean that well-run sites add structured data and do many other things right at the same time.

So Ahrefs ran a cleaner test, published in May 2026. It followed 1,885 pages that added JSON-LD between August 2025 and March 2026, and compared each with similar pages that did not. Thirty days later, ChatGPT citations were up 2.2% and Google AI Mode citations up 2.4%, both too small to tell apart from chance. AI Overviews citations fell 4.6% more than for the other pages. Both groups were already falling, and Ahrefs could not pin that small gap on the code (Ahrefs).

The test has limits, which Ahrefs lists. Every page in it was already cited often by AI Overviews, so it says nothing about pages an assistant has not found yet. It also lumped all kinds of structured data together.

Two smaller tests looked at what an assistant reads when it opens a page. In one, by searchVIU and reported by Ahrefs, ChatGPT, Claude, Perplexity, Gemini and Google AI Mode each fetched pages live. All five took only the text a visitor can see and skipped the JSON-LD (Ahrefs).

In the other, Mark Williams-Cook put a made-up company's address only inside broken JSON-LD. ChatGPT and Perplexity still gave the address. Since the code was not valid, he concluded they read it as plain text on the page (Search Engine Roundtable, February 2026).

The two tests disagree on whether the code gets read at all. Neither found an assistant treating it as labeled data.

What to do

1. Keep structured data where search engines use it: products, reviews, events and your company's details. Google AI Mode answers from Google's search, and Copilot from Bing's.
2. Put every fact in the visible text as well: prices, opening hours, addresses, plan names. Google asks for that, and the tests found assistants read what a visitor can see.
3. If you sell online, send product feeds to Google's Merchant Center, and to OpenAI for shopping in ChatGPT.
4. Do not expect citations from the code alone. Clear answers, evidence and other sites that name you count for more. GEO vs SEO: what changes.
5. See what AI tools can read on your page. The free AI Readiness Check looks at structured data, titles and headings, and how much of your visible text reaches AI tools.