Structured data for AI search is not the same thing as structured data for Google. If you’re still running the Schema.org playbook you built in 2019 and hoping AI engines will figure it out, you’re leaving most of the value on the table.
Schema.org JSON-LD still matters. It’s the baseline. But AI search engines read your site in a fundamentally different way than Google does, and the stack that actually moves the needle in 2026 includes signals that didn’t even exist two years ago. llms.txt, entity markup, speakable schema, AI-specific robots rules, organization knowledge graphs.
This is the full 2026 playbook for making your site readable by AI engines. Not a rehash of what you already know.
The old stack: Schema.org JSON-LD (necessary, not sufficient)
Let’s get the obvious out of the way. You still need classic Schema.org on every important page. AI engines read it, Google reads it, and it’s the closest thing we have to a universal language for “here’s what this page is.”
The minimum you should have running on every page that matters:
-
Organization schema on your homepage (with
sameAslinking to all your social profiles — this is how AI engines resolve your entity) -
Article schema on every blog post with full author details,
datePublished, anddateModified -
Product schema on product pages
-
BreadcrumbList on every page
This is the baseline. If you don’t have this, nothing else matters. If you already have this, keep reading because the interesting stuff starts now.
The new stack: what AI engines actually read in 2026
1. llms.txt — the sitemap for LLMs
llms.txt is a plain-text file you drop at the root of your domain (yourdomain.com/llms.txt) that tells AI models what your site is about and where the important content lives. Think of it as robots.txt meets sitemap.xml, but written for language models.
It’s a proposed standard, not an official one yet, but Anthropic, OpenAI, and Perplexity have all signaled they’re reading it. Early adopters are already showing up in more AI citations than sites without it.
A minimal llms.txt looks like this:
# YourCompany
> One-sentence description of what your company does.
Longer paragraph providing more context about who you are, what you offer, and why AI models should care about your content.
## Docs
- [Getting Started](https://yourdomain.com/docs/getting-started): Quick setup guide
- [API Reference](https://yourdomain.com/docs/api): Full API documentation
## Blog
- [Latest post title](https://yourdomain.com/blog/latest-post): Short descriptionShip this today. It takes fifteen minutes and the compounding effect over the next year is going to be large.
2. robots.txt with explicit AI bot rules
Your robots.txt is probably blocking AI bots without you realizing it. Go check. Right now.
The bots you want to explicitly allow:
User-agent: GPTBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Perplexity-User
Allow: /
User-agent: Google-Extended
Allow: /
User-agent: CCBot
Allow: /
User-agent: anthropic-ai
Allow: /If you have a blanket Disallow: / for unknown agents, or your CDN/WAF is blocking bots by default, these explicit allows fix it. This is the single most common reason sites with good content don’t show up in AI answers.
3. Organization schema with sameAs (entity resolution)
AI engines resolve brands through entities, not URLs. When Perplexity sees a mention of “RunAgents,” it tries to match it to a specific entity in its knowledge base. The way you make that matching work is the sameAs property on your Organization schema.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "RunAgents",
"url": "https://runagents.dev",
"logo": "https://runagents.dev/logo.png",
"sameAs": [
"https://twitter.com/runagents",
"https://www.linkedin.com/company/runagents",
"https://github.com/runagents",
"https://www.crunchbase.com/organization/runagents",
"https://en.wikipedia.org/wiki/RunAgents"
],
"foundingDate": "2024",
"founders": [{
"@type": "Person",
"name": "Your Name"
}]
}Every URL in sameAs is a reinforcement signal that ties your brand to a canonical entity. The more reputable profiles you link, the more confident AI engines become that you are a real, distinct entity worth citing. See our entity SEO guide for the full playbook on entity reinforcement.
4. FAQPage schema (critical for AI extraction)
FAQPage schema is the single highest-leverage schema for AI search because it gives engines clean question-answer pairs they can lift directly into an answer.
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What is structured data for AI?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Structured data for AI refers to machine-readable markup that helps AI search engines understand and cite your content. Beyond Schema.org, it includes llms.txt, entity markup, and AI-specific robots rules."
}
}, {
"@type": "Question",
"name": "Is Schema.org enough for AI search?",
"acceptedAnswer": {
"@type": "Answer",
"text": "No. Schema.org is the baseline. Modern AI engines also read llms.txt files, entity graphs via sameAs, and extract content from FAQPage and HowTo schema more aggressively than traditional search."
}
}]
}Every page with common questions should have this. Blog posts, product pages, help center articles, landing pages. It’s the closest thing to “please cite me” you can put in code.
5. HowTo schema (for tutorial extraction)
If you publish tutorials, how-to content, or step-by-step guides, HowTo schema is your best friend. AI engines love step-by-step content because it maps cleanly to how people ask questions.
{
"@context": "https://schema.org",
"@type": "HowTo",
"name": "How to add llms.txt to your site",
"step": [{
"@type": "HowToStep",
"name": "Create the file",
"text": "Create a plain-text file named llms.txt in your project root."
}, {
"@type": "HowToStep",
"name": "Add your content",
"text": "List your site name, description, and links to your most important pages."
}, {
"@type": "HowToStep",
"name": "Deploy",
"text": "Deploy the file so it's accessible at yourdomain.com/llms.txt"
}]
}6. Article schema with full author details
Author attribution is a major trust signal for AI engines. A blog post with a vague byline or no author info won’t get cited as often as a post with a verified author, their bio, and links to their other work.
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Your article title",
"datePublished": "2026-04-09",
"dateModified": "2026-04-09",
"author": {
"@type": "Person",
"name": "Ronak Kadhi",
"url": "https://runagents.dev/about",
"sameAs": [
"https://twitter.com/ronakkadhi",
"https://www.linkedin.com/in/ronakkadhi"
]
},
"publisher": {
"@type": "Organization",
"name": "RunAgents",
"logo": {
"@type": "ImageObject",
"url": "https://runagents.dev/logo.png"
}
}
}The sameAs on the author is just as important as the sameAs on the Organization. It ties the byline to a real person AI engines can verify.
7. Speakable schema (underused and underrated)
Speakable schema tells AI engines which parts of your page are suitable for voice and audio extraction. It was originally designed for Google Assistant, but AI engines are reading it too because it’s basically a map to the “most important sentence” on a page.
{
"@context": "https://schema.org",
"@type": "Article",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".article-summary", ".key-takeaway"]
}
}Tag your TL;DR and key takeaway sections with these CSS selectors and you’re telling AI engines “if you’re going to quote one thing from this page, quote this.” Almost nobody uses it. That’s the opportunity.
What to remove (outdated or ignored schema types)
Not all schema is worth the effort. These types are either ignored by modern AI engines or have been deprecated in practice:
-
CreativeWork by itself — too generic, always use a more specific subtype
-
Thing — the root type, basically useless
-
WebPage without context — rarely adds value beyond what AI engines already infer
-
Old itemprop microdata — switch everything to JSON-LD, microdata is dying
If you have any of these cluttering your templates, clean them up. Less is more. AI engines prefer a few high-quality signals over a wall of generic markup.
Testing your structured data
Three tools you should run every deploy:
-
Google Rich Results Test — validates JSON-LD and shows how Google will render it
-
Schema.org Validator — catches spec-level issues that Google’s tool misses
-
Your own llms.txt — just fetch it with
curl https://yourdomain.com/llms.txtand make sure it loads clean
For a deeper check across all your AI signals at once, run your domain through our free marketing audit. It checks Schema.org coverage, llms.txt, robots.txt, entity resolution, and flags what’s breaking your chances of getting cited.
How AI engines parse this differently from Google
Google uses structured data as hints for rich results. AI engines use it as ground truth.
When Google sees your FAQ schema, it might show a rich result in the SERP. When Perplexity sees your FAQ schema, it may lift your answer directly into its response and cite you. One is a cosmetic upgrade. The other is direct traffic and authority.
That difference is why structured data for AI search deserves its own playbook. It’s not the same game. The inputs look similar but the outputs are orders of magnitude different.
The implementation order
If you have one day to fix this, here’s the priority order:
-
Check your robots.txt — make sure AI bots aren’t blocked
-
Ship llms.txt
-
Add Organization schema with complete sameAs
-
Add Article schema with full author details to all blog posts
-
Add FAQPage schema to any page with Q&A content
-
Add HowTo schema to tutorials
-
Add Speakable to article summaries
Everything else is polish. Ship this stack and you’ll immediately see more AI citations. Our structured data for AI tool tracks the full stack for you and alerts you when something breaks.
Structured data isn’t a silver bullet. But it’s the closest thing to one AI search has right now. Run your free marketing audit, see where you stand, and fix the gaps this week. The sites that get this right in 2026 are going to dominate AI citations for the next three years.