JSON-LD Schema for AI Visibility

JSON-LD (JavaScript Object Notation for Linked Data) has been part of the SEO toolkit since Google recommended it in 2015. For a decade, its primary value was generating rich snippets — star ratings, recipe cards, FAQ dropdowns in search results. Useful, but not transformative.

That changed when AI models started generating answers instead of listing links. When ChatGPT, Gemini, or Perplexity answers a question, they don't just scan page text — they parse structured data to extract specific facts, prices, steps, and definitions. A page with a FAQPage schema gives the AI a pre-structured question-answer pair it can cite directly. A page without it forces the model to infer the answer from unstructured prose — which it does less reliably and less often.

The data is clear: the Authoritas study in 2025 found that pages with structured data are 2.3x more likely to be cited in AI Overviews. This makes JSON-LD the single highest-impact technical change you can make for AI visibility — and it sits squarely in Layer 2 (Comprehension) of the agentic web architecture.

The difference matters because it changes what you optimise for.

Search engines use JSON-LD for display purposes. They parse your Article schema to show a publication date and author in search results. They parse your FAQPage schema to show expandable Q&A sections. They parse your Product schema to show star ratings and prices. The structured data enhances the search result — but the ranking algorithm relies primarily on other signals.

AI models use JSON-LD for knowledge extraction. They don't display rich snippets — they generate answers. And when generating answers, structured data is significantly more reliable than unstructured prose. Here is specifically how AI models use each type:

• Fact extraction: When a model needs a specific number (price, date, quantity), it checks structured data first. A Product schema with "price": "3000" is more trustworthy than the phrase "our prices start from around three thousand pounds" buried in paragraph five.
• Direct citation: FAQPage Q&A pairs are used almost verbatim. When a user asks ChatGPT a question that matches one of your FAQ questions, the model can cite your answer directly — with attribution — because the structured format removes ambiguity about what the answer is.
• Entity resolution: Organization and Person schemas help AI models understand who is behind the content. This matters for trust signals — AI models weight content differently based on the authority of the author and publisher.
• Content structure understanding: BreadcrumbList and Article schemas tell the model how a page fits into the broader site structure. This helps when the model needs to determine if a page is an authoritative reference or a minor blog post.
• Step-by-step extraction: HowTo schema is particularly valuable because AI models can extract individual steps and present them in order, with attribution. "According to p0stman's guide, step 1 is..." — this only works well when the steps are in structured data, not when they're embedded in flowing prose.

Not all schema types are equal for AI visibility. Here is the ranking based on observed impact across multiple production sites:

FAQPage deserves its own section because it has the single highest impact on AI citations. The reason is structural: FAQPage schema provides pre-formatted question-answer pairs that AI models can cite with minimal processing.

When an AI model encounters a FAQPage schema, it gets:

1. An explicit list of questions the page answers
2. A structured answer for each question
3. A clear attribution path (the page URL)

This is dramatically easier to cite than scanning a 6,000-word page for the relevant paragraph. The model knows exactly which question is answered and exactly what the answer is.

Guidelines for writing FAQPage content that AI models will cite:

• Write the question as users ask it. "How much does a custom AI agent cost to build?" not "Pricing information for agent development." AI models match user queries to FAQ questions — match the natural language pattern.
• Keep answers between 50-150 words. Too short and the answer lacks detail. Too long and the AI model will truncate or summarise, losing your specific phrasing.
• Include specific numbers. "Between £5,000 and £20,000" is citable. "Varies depending on requirements" is not.
• Make each answer self-contained. Don't reference other answers ("as mentioned above"). The AI model may cite a single Q&A pair in isolation.
• Match the FAQ content to the page's visible FAQ section. Google requires that FAQ schema matches what's visible on the page. This is also good practice for AI models — they cross-reference structured data against page content.
• Include 6-10 questions per page. Enough to cover the topic comprehensively, but not so many that the schema becomes unwieldy. Prioritise the questions that drive the most search traffic.

Most pages should have multiple schema types. A guide page needs Article (authorship and publication context) + FAQPage (direct Q&A pairs) + BreadcrumbList (site structure). Rather than adding three separate <script type="application/ld+json"> tags, use the @graph pattern to combine them in one.

The @graph pattern is:

• Cleaner — one script tag instead of three
• Fewer DOM nodes — slightly better performance
• Recommended by Google — their documentation uses this pattern for multi-type schemas
• Better for AI parsing — all structured data is in one place, making it easier for AI crawlers to extract in a single pass

The simplest approach. Add the JSON-LD directly in the <head> section of your HTML file:

In Next.js, you can include JSON-LD in your page component using dangerouslySetInnerHTML. Next.js also supports the metadata API, but JSON-LD requires the script tag approach:

For single-page applications, use react-helmet or react-helmet-async to inject JSON-LD into the head:

Important SPA caveat: Most AI crawlers do not execute JavaScript. If your SPA renders JSON-LD client-side only, AI crawlers will not see it. For SPAs, you must either:

• Use server-side rendering (SSR) or static site generation (SSG)
• Pre-render pages for crawlers using a pre-rendering service
• Include the JSON-LD in the initial HTML served by your server

This is another reason to prefer Next.js or similar SSR frameworks — the JSON-LD is in the initial HTML response, visible to every crawler.

1. Invalid JSON

The most common mistake is invalid JSON — trailing commas, unescaped quotes, HTML entities in the JSON-LD block. AI crawlers parse JSON-LD as raw JSON. If it doesn't parse, it's ignored entirely. Always validate your JSON-LD output with a JSON validator before deployment.

2. FAQ Schema That Doesn't Match the Page Content

Google requires that FAQPage schema matches visible content on the page. If your schema lists 10 FAQ questions but the page only shows 6, Google may ignore the schema or penalise the page. AI models also cross-reference structured data against page content — inconsistencies reduce trust. Always ensure your FAQ schema matches a visible FAQ section on the page.

3. Missing dateModified

AI models heavily weight content freshness. A page published in 2024 with no dateModified field looks stale. Always include dateModified in your Article schema and update it whenever you meaningfully change the content. This is one of the simplest signals you can send to AI models: "this content is current."

4. Overly Generic Descriptions

A schema with "description": "Learn about our services" gives AI models nothing to work with. Be specific: "description": "Complete guide to building MCP servers for AI agent integration, with code examples in Next.js and pricing benchmarks for 2026". The description should be useful out of context.

5. Missing author and publisher

AI models use authorship signals to assess content authority. An Article without author or publisher schema is treated as anonymous content — lower trust, lower citation priority. Always include both, with real names and URLs that resolve to actual about/profile pages.

6. FAQ Answers That Are Too Long or Too Short

Answers under 20 words lack the detail AI models need to generate useful responses. Answers over 200 words get truncated or summarised, losing your specific phrasing. The sweet spot is 50-150 words — detailed enough to be useful, concise enough to be cited intact.

7. Not Using @graph

Multiple separate JSON-LD script tags work, but they're harder to maintain and occasionally cause parsing issues with some crawlers. The @graph pattern is cleaner and explicitly signals that the schemas belong to the same page context.

Three tools you should use, in order:

For AI-specific validation, manually check:

• Is the JSON-LD in the initial HTML response? View page source (not DevTools Elements tab) and search for "application/ld+json". If it's not there, AI crawlers can't see it.
• Are FAQ answers concise and self-contained? Read each answer in isolation. Does it make sense without the surrounding content?
• Are dates current? Check that dateModified reflects the actual last modification date.
• Do author/publisher URLs resolve? If your schema references https://p0stman.com/about, that URL must return a real page.

Three structured data formats exist. For AI visibility, JSON-LD is the clear winner. Here's why:

The decisive advantage of JSON-LD for AI is separation from HTML. AI crawlers can extract JSON-LD from a page without rendering the DOM. They find the <script type="application/ld+json"> tag, parse the JSON, and have structured data immediately. With Microdata or RDFa, the crawler must render the full HTML, traverse the DOM tree, and extract attributes from individual elements — a much slower and more error-prone process.

Google has explicitly recommended JSON-LD as the preferred structured data format since 2015. Every major AI model's crawling infrastructure (GPTBot, ClaudeBot, PerplexityBot) parses JSON-LD. If you're currently using Microdata, migrating to JSON-LD is the single most impactful change you can make for AI visibility.

Google's AI Overviews (the AI-generated answer box at the top of search results) is the most visible manifestation of AI using structured data. Understanding how it selects sources helps you optimise your JSON-LD:

1. Structured data presence is a strong signal. Pages with JSON-LD are 2.3x more likely to be cited. This is not correlation — structured data directly helps the AI extract and verify information.
2. FAQPage schema creates direct citation paths. If a user's query matches one of your FAQ questions, the AI Overview can cite your answer with high confidence. The structured Q&A format removes ambiguity.
3. freshness signals matter. dateModified tells the AI how current your content is. All else being equal, a page updated in March 2026 will be cited over a page last updated in 2024.
4. Authority signals (author, publisher) contribute. AI Overviews prefer sources from known entities. Organization schema helps Google's Knowledge Graph resolve your identity.
5. Content depth matters alongside schema. Structured data without substantive content won't perform. The page needs both — the structured data helps the AI find and cite the right parts of your comprehensive content.

Here are observed results from adding comprehensive JSON-LD schema to existing content pages:

These results are from a content site with 40+ pages that added Article + FAQPage + BreadcrumbList schema to all pages, plus answer capsules at the top of each page. The JSON-LD was the primary change — no new content was added during the measurement period.

The results compound over time. AI models learn which sources provide structured, reliable data. Sites that consistently have good JSON-LD build a reputation with AI models — they get crawled more frequently and cited more readily. This is the AI equivalent of domain authority.

JSON-LD and answer capsules are the two pillars of the Comprehension layer in agentic web architecture. They work together but through different mechanisms:

• JSON-LD (structured path): AI models parse the schema programmatically, extracting facts, Q&A pairs, steps, and metadata. This feeds the model's knowledge graph directly.
• Answer capsules (content path): AI models read the page content from top to bottom. The answer capsule — a direct answer in the first 30% of the page — gets cited 44% of the time because it's the most prominent, most direct content the model encounters.

A page with FAQPage schema and an answer capsule covers both paths. The schema gives AI models structured data they can extract and cite programmatically. The answer capsule gives them a direct, quotable answer they can cite from the content. Together, they maximise the probability that your page is cited in AI-generated responses.

The answer capsule on this page demonstrates the pattern: it directly answers "what is JSON-LD for AI visibility" in two sentences. The FAQPage schema in this page's JSON-LD provides 8 additional Q&A pairs. Between the two, an AI model has multiple citation paths into this page's content.