Schema markup for AI: FAQ, Article, Organization - is it still worth it?

For the past two years, "just add schema and AI will cite you" has been the most common advice in SEO. By mid-2026, two major data points forced a rethink: Google officially retired FAQ rich results, and a controlled Ahrefs study showed that adding schema does not lift AI citations by itself. So is schema still worth doing, and how do you do it right?

What is schema markup and how does AI read it?

Schema markup, or structured data, is code added to your HTML that describes a page's content using the standard schema.org vocabulary - a project launched in 2011 by Google, Microsoft, Yahoo and Yandex. Instead of letting machines guess, schema states it plainly: this page is an Article, written by this person, published on this date, by this Organization, answering these questions (FAQPage). The recommended format today is JSON-LD: a <script type="application/ld+json"> block inside the page that has no effect on what readers see.

Here is the part most people miss: AI "reads" websites through two paths. One goes through a search index - Googlebot and Bingbot crawl your pages, process the schema, and AI Overviews or Copilot pick sources from that index. The other is AI fetching your page directly at answer time (ChatGPT, Perplexity) - and on this path, testing shows they only read the visible text and skip JSON-LD entirely.

The two paths AI takes to your content. Schema only matters on the search-index path. Source: compiled from the searchVIU experiment and Google Search Central documentation (2026).

What does the 2026 data say about schema and AI citations?

In May 2026, Ahrefs published the largest controlled study on this question to date: it tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared them against a control group of 4,000 pages. The result: adding schema produced no meaningful change in how often pages were cited by AI. Specifically, the change on Google AI Mode was +2.4%, on ChatGPT +2.2% (both statistically indistinguishable from zero), and on Google AI Overviews it was actually -4.6%.

Data source: Ahrefs, "We Tracked 1,885 Pages Adding Schema", published May 11, 2026.

An independent experiment by searchVIU reinforces the finding: when five AI systems (ChatGPT, Claude, Perplexity, Gemini, Google AI Mode) fetched a page in real time, none of them read the JSON-LD - every one extracted only the visible HTML content. Around the same time, on May 7, 2026, Google removed FAQ rich results from search entirely, completing a restriction that began in August 2023.

But there is another side to the picture. In March 2025, Fabrice Canel, Principal Product Manager at Microsoft Bing, confirmed at SMX Munich that schema helps Microsoft's LLMs (Copilot) understand website content. Google has also made clear that FAQPage remains a valid schema type, is still parsed to understand pages, and that unused structured data causes no harm. In other words: schema is no longer a "make my snippet bigger" button, but it is still the base layer that helps machines understand exactly who you are.

So should you drop schema markup?

No - but you should reset your expectations: schema is near-zero-cost identity infrastructure, not a direct citation lever. Four values still hold: Google uses it to understand entities and rank pages - and organic ranking remains the factor most strongly correlated with being picked as an AI Overviews source; Bing confirms schema helps Copilot's LLMs, while ChatGPT relies on Bing's index for many queries; other rich results (Organization, Breadcrumb, Product, Review) still work; and consistent schema feeds the Knowledge Graph - the foundation for AI getting your brand name right.

What schema cannot do is rescue thin content. If an article has no original information, no data and no direct answers, no amount of markup will make AI cite it. That is the practical conclusion of the Ahrefs study: the deciding factor lives in the visible content, not in a script tag.

The three schema types a business should prioritize

A small or mid-sized business website does not need dozens of schema types. These three cover most of the value:

• Organization - declares the business: name, logo, address, social profiles (sameAs). This is your brand's ID card; place it on the homepage and reference it (@id) from every other page. It is the starting point if you want AI assistants to mention your brand by name.
• Article / BlogPosting - declares a post: headline, description, publish and modified dates, and most importantly an author of type Person pointing to a real profile page. This is an E-E-A-T signal both Google and AI systems weigh heavily.
• FAQPage - declares question-answer pairs that exactly match the FAQ visible on the page. The rich result is gone, but concise 40-70 word Q&A pairs remain the format AI answers quote verbatim most often.

All three are written in JSON-LD. One note from running this very website: inside about or mentions arrays, use type Thing rather than Product unless the page actually sells a specific product - it avoids product-snippet errors in Google Search Console.

How to implement it right: a 6-step checklist

Step 6 is the one that matters most. The Princeton GEO study (presented at KDD 2024) found that adding source citations, statistics and expert quotes to content improved visibility in AI answers by up to 40% - a far larger effect than any technical intervention. If you are new to this topic, start with what GEO is and why it matters from 2026, then see how AI answers are already reshaping search traffic in the Chegg vs Google AI Overviews case study.

Where should a business start?

From our audits of client websites, most small-business sites either have no schema at all or only WordPress theme defaults, with no Organization block and no real author. That gap is an opportunity: adding the three core schema types takes a few hours, done once and used for years. A sensible roadmap: add proper schema (week 1), rewrite key pages in a GEO-friendly format with a visible TL;DR and FAQ (weeks 2-4), declare an llms.txt file for AI crawlers, then measure monthly by asking ChatGPT and Gemini about your brand - here is how AI Overviews is rolling out in Vietnam if that is your market.

This is the exact process Chạm AI applies to its own website and to clients in our GEO and AI Search service: schema as the base layer, direct-answer content as the deciding layer, and a consistent brand as the layer that keeps AI calling you by the right name.

References: Ahrefs, "We Tracked 1,885 Pages Adding Schema. AI Citations Didn't Move" (May 2026) · Google Search Central Blog, "Changes to HowTo and FAQ rich results" (Aug 2023) and the FAQ rich results removal notice (May 2026) · Search Engine Land, "Microsoft Bing/Copilot use schema for its LLMs" (Mar 2025) · Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024.

Frequently asked questions