If you have spent any time in SEO circles this year, someone has told you to add schema markup to your website for AI search. The advice sounds reasonable: structured data helps machines understand your content, and AI search engines are machines, so schema should help you get cited. New research says it does not work that way.
Pilot Digital, a Portland-based agency, ran a controlled experiment in August 2026 and published the results in October. They built a static HTML page with 14 hidden markers (“canaries”) placed in different locations and containers, then asked ChatGPT, Gemini, and Claude to fetch the page and report what they found. The findings were conclusive: all three models found zero of 15 JSON-LD blocks.
The reason is simple. JSON-LD schema lives inside a <script> tag. When AI models fetch a page, they strip all script tags during content extraction. The schema disappears before the model ever sees it.
What the Experiment Found
Pilot Digital placed JSON-LD blocks in five different spots on the page: header, body, footer, and other locations. They also hid the same JSON data inside a regular <div> tag instead of a <script> tag, to test whether the container matters.
Here is what happened:
- JSON-LD in script tags: 0 of 5 found by ChatGPT, 0 of 5 by Gemini, 0 of 5 by Claude. Adding two common scraping tools (Firecrawl and Chrome reader) brought the total to 0 of 25.
- The same JSON in a hidden div: All three models found it.
- Normal visible text (positive control): All three models found it.
- Meta description: Only Claude found it.
Location on the page did not matter. Header, footer, body, it made no difference. The containing element is the reason schema does not survive the fetch. Gemini itself confirmed this in its response: “This tool strips script tags, including Schema/JSON-LD, during content extraction.”
What Four AI Search Experts Say
97th Floor surveyed four experts on the same question. Their answers disagree less than they seem:
- Thomas Peham (OtterlyAI): Schema had “zero, literally zero impact” on citation visibility in ChatGPT and Gemini in their tests.
- Chris Long (Nectiv): Schema probably does not help much with LLMs because they strip markup, but it is cheap to implement and still matters for Google.
- Chima Mmeje (Moz): Schema is important, but llms.txt is not. Focus on what actually moves the needle.
- Martha van Berkel (Schema App): Schema is not a rich-result checklist. It is the knowledge graph AI uses to understand your brand. Build for entities, not rich results.
The thread connecting all four: schema still matters for Google’s regular search results and rich snippets, but it probably does not directly influence whether ChatGPT or Claude cites your page.
Does This Mean You Should Remove Schema?
No. Here is what the research actually tells us:
- Keep schema for Google. Google’s crawler reads JSON-LD and uses it for rich results, knowledge panels, and entity understanding. That has not changed. If you need help with this, our SEO services include structured data implementation.
- Do not expect schema alone to get you cited by AI. If your entire AI search strategy is “add schema and wait,” you will be disappointed. AI models cannot see your JSON-LD when they fetch your page.
- Put critical information in visible HTML text. If something is important for AI to know, it needs to be in readable text on the page, not hidden in structured data. Business name, services, hours, location, FAQs, all of it should be in plain HTML.
- Build for entities, not just rich results. Schema App’s Martha van Berkel makes a useful point: schema is how you connect the dots across your website. Use it to link your organization, products, people, and locations with sameAs and breadcrumb properties. This helps Google’s knowledge graph understand your brand.
What Actually Gets You Cited by AI
If schema is not the answer, what is? The Houston AI citation study by Andres Rusic, published October 30, offers the clearest evidence so far. His team asked four AI engines 232 local questions and traced every source cited.
The factors that correlated with getting cited:
- AI-readable HTML: Sites with at least 120 words of text readable without JavaScript were cited 4.6 times more often than sites that failed basic checks.
- Review volume: Businesses with 1,000+ Google reviews were twice as likely to be named by multiple engines.
- Directory presence: Being listed on the directories each engine reads (Yelp for Claude, BestProsInTown for ChatGPT) mattered more than on-site schema.
- Content worth citing: Clean HTML, clear answers to common questions, and a site structure AI crawlers can follow.
None of these require JSON-LD schema. They require readable content, real reviews, and presence on the platforms AI engines actually read.
The Bottom Line for Local Businesses
Schema markup is still good SEO practice. It helps Google understand your site and can earn rich results. But do not treat it as your AI search strategy. The research is clear: when ChatGPT, Claude, or Gemini fetches your page, they cannot see your JSON-LD.
If you want to improve your AI search visibility, focus on what the models can actually read: visible HTML text, a fast and crawlable website, strong review profiles, and listings on the directories each engine trusts. You can learn more about our approach on our about page, or book a meeting with us to discuss your specific situation.
Frequently Asked Questions
Does schema markup help with AI search visibility?
According to Pilot Digital’s October 2026 research, ChatGPT, Gemini, and Claude all strip script tags during page fetches, which means JSON-LD schema is invisible to them. Schema had zero impact on AI citation visibility in tests by OtterlyAI as well.
Should I still use schema markup on my website?
Yes. Schema markup still helps Google understand your site for regular search results and rich snippets. It is cheap to implement and good technical practice. Just do not expect it to directly improve your visibility in ChatGPT or Claude.
What should I do instead of schema for AI search?
Focus on visible HTML text that AI models can read when they fetch your page. Ensure at least 120 words of readable content without JavaScript, build review volume on Google and relevant directories, and maintain presence on the platforms each AI engine cites.
Why can’t AI models see JSON-LD schema?
JSON-LD schema lives inside a script tag in the HTML. When AI models like ChatGPT, Claude, and Gemini fetch a page, they strip all script tags during content extraction as a security measure. The schema data disappears before the model processes the page.
Is llms.txt worth adding for AI search optimization?
The evidence is mixed. Moz’s Chima Mmeje says llms.txt will not make the difference people claim, and Google reportedly does not use it. It is low-cost to add, but should not be a priority compared to readable content and directory presence.