Is schema.org markup necessary to be cited by AI?
Published on
No. Schema.org markup is not a condition for being cited by AI. Google writes that it is not necessary to create markup to appear in its AI responses. Ahrefs, by comparing pages that added structured data to control pages, measured no net increase in citations. And in our measurements, between 12 and 31% of pages cited by Claude and Gemini carry no markup. Structured data describes a page; it does not make it chosen.
For what type of question? It all depends on what you expect from the markup. It remains useful for classic Google Search to display certain rich results. To be cited as a source in an AI response, on an informational or market question, it is the text of the page that matters.
What Google says about markup and its AI responses
Google Search Central's documentation on AI features, updated on December 10, 2025, is explicit. There is no additional requirement or special optimization to appear in AI Overviews or AI Mode, and it is not necessary to create new machine-readable files, text files for AI, or markup to appear there. The only condition to be a source link is that of classic search: to be indexed and to be able to display with a snippet.
This official position does not state what Google's systems favor in practice. It does state, however, that no markup opens a closed door: a non-indexed page will not be cited because it carries JSON-LD, and an indexed page that does not carry any is not dismissed for all that.
The guide Google dedicates to optimization for its generative AI features, updated on July 10, 2026, confirms it in black and white: structured data is not required for generative AI search, and no particular schema.org markup needs to be added. It also specifies that it is useless to create text files for AI, since Google Search itself does not use them.
Ahrefs compared pages with and without adding schema
The most robust study on the subject is that of Ahrefs, published on May 11, 2026, because it uses a control group, which is rare in this field. Ahrefs tracked pages that added JSON-LD markup between August 2025 and March 2026, and compared them to similar pages that did not add any, thirty days before and thirty days after. Ahrefs' conclusion: adding schema produced no notable increase in citations, on any platform.
In detail, citations increased by 2.4% in AI Mode and by 2.2% in ChatGPT, two non-significant gaps, and they decreased by 4.6% in AI Overviews, a significant gap. Ahrefs points out two limitations: the studied pages were already highly cited, and the addition of markup could not be totally isolated from other changes made to the pages.
What we measured on the cited pages
We recorded the markup of each HTML page cited by AI in our corpora: HVAC in French, artificial intelligence in English, about ten sectors in French then in English, professional software troubleshooting. For Claude, the share of cited pages without any markup ranges from 12 to 25% depending on the corpus; for Gemini, from 12 to 31%. At least one page out of eight, and sometimes nearly one out of three, is therefore cited without the slightest structured data.
FAQ markup, often recommended for AI, is a minority among the cited pages: between 10 and 27% depending on the corpus and the AI. The majority of cited pages carry markup of another nature, or none. These counts do not prove that markup is useless, since many sites implement it by default; they prove that it is not a condition for citation.
Where the advice to add schema for AI comes from
The advice comes mainly from observations: cited pages often carry structured data. But large sites, which are also the most cited, use tools that implement them automatically. A correlation of this type does not say that markup causes the page to be cited; this is precisely what Ahrefs' study, with its control group, verified, without finding any effect.
Seer Interactive, which analyzed Google's AI responses in the United States from May 7 to 13, 2026, goes further: according to its study, FAQ and HowTo markups are of no interest for AI Overviews, and an Article markup accompanied by a breadcrumb trail seems to suffice. These results apply to Google only, and over a single week.
What this changes for a business
Simple and accurate markup, which describes the article, the business, and the breadcrumbs, costs nothing and remains useful for traditional search. Multiplying schema types in the hope of being cited by ChatGPT, Claude, or Gemini is, on the other hand, an effort with no measured effect.
The saved time is better spent on what has a measured effect: a precise question in the title and the page address, an answer in the very first lines, and developed paragraphs that use the words AIs search with. The pages in this section feature Article markup and breadcrumbs, and no FAQ markup.
What we do not know
Our counts describe the markup of the cited pages; they do not measure what happens when a page adds or removes markup. They focus on the models' API, and not on AI Overviews or AI Mode.
Sources
- Google Search Central — “AI features and your website”, .
- Ahrefs, Louise Linehan — schema and AI citations, .
- Seer Interactive, Matt Buxbaum — what it takes to appear in AI Overviews in 2026, .
- Google Search Central — “Optimizing your website for generative AI features on Google Search”, .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".