JSON-LD for content sites: Article, FAQPage, HowTo, BreadcrumbList

Short answer: Four JSON-LD schemas move the needle on a content site in 2026 — Article (or TechArticle) on every article page, BreadcrumbList on every page deeper than /, FAQPage only when there are 3+ genuine question-answer pairs, and HowTo only when the page is a real procedural walkthrough. Skip ImageObject, Review, AggregateRating, VideoObject, WebSite, and WebPage — they add no citation lift or carry penalty risk.

Schema.org has more than 800 types. Most AEO advice from 2018-2022 told you to use 50 of them. Most of that advice is wrong for 2026. This article is the actual subset that LLMs read on a content site, in order of impact, with the exact JSON-LD shape that gets cited.

The audience is the engineer doing the implementation work. If you are a marketing lead, hand this to the developer who owns the templates and ask them to apply the four shapes below. If you are the developer, the shapes are copy-paste-ready against any static- site framework that lets you emit <script type="application/ld+json"> in the document head.

The four schemas that move the needle on a content site

Content sites — blogs, documentation, knowledge bases, editorial publications — get most of their AEO lift from four JSON-LD types:

Article (or TechArticle for technical content) on every article page.
BreadcrumbList on every page deeper than the homepage.
FAQPage on pages with three or more genuine question-answer pairs.
HowTo on pages that walk the reader through an ordered sequence of steps.

That is the entire list. Skipping ImageObject, Review, AggregateRating, VideoObject, WebSite, WebPage, and the rest is intentional. They either add no citation lift, or they add risk (AggregateRating on a service page is a known manipulation flag), or they are noise that distracts from the schemas that do work.

Schema 1: Article — the canonical content surface

Every article page on a content site emits Article JSON-LD. That is the highest-leverage single change you can make. LLMs use the structured fields — headline, datePublished, author, mainEntityOfPage — to disambiguate similar titles across the web and to attribute citations correctly.

The minimum viable shape:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "JSON-LD for content sites: Article, FAQPage, HowTo, BreadcrumbList",
  "description": "The four JSON-LD schemas that move the needle on content sites in 2026, with the exact shapes that get cited by ChatGPT, Claude, and Perplexity.",
  "datePublished": "2026-06-01",
  "dateModified": "2026-06-01",
  "author": {
    "@type": "Person",
    "name": "Tudor Constantin",
    "url": "https://boringtechnologies.com/about"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Boring Technologies",
    "url": "https://boringtechnologies.com"
  },
  "inLanguage": "en-US",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://boringtechnologies.com/articles/schema-deep-dive-content-sites"
  }
}

What to use, why each field is there:

headline — the article title. Limited to 110 characters per Google’s Article schema guidance; LLMs do not enforce a hard cap but anything past 110 chars is a usability flag.
description — one to two sentences, ideally matching the meta description and the lede on the page. LLMs quote this verbatim more often than any other field.
datePublished and dateModified — ISO 8601. The first surfaces the article’s age in answers (“published in 2026”). The second informs whether the LLM treats the content as current or stale during retrieval ranking.
author as a Person (not a string) — gives LLMs an entity to link, which is how author-attributed citations get generated. The url should point to a real about page that also emits Person schema.
publisher as an Organization — pairs with author. The publisher entity should match the homepage Organization schema.
mainEntityOfPage — the canonical URL of the article. This is what closes the loop between the structured data and the HTML page; without it, LLMs sometimes misattribute citations to a different URL on the same domain.

What to omit:

interactionStatistic, commentCount, wordCount — noise. Not used by LLMs for citation ranking.
image and thumbnailUrl — if you have a hero image, emit it; if not, do not invent one.
articleSection, keywords — optional, low impact. Emit if your taxonomy is genuinely meaningful, otherwise skip.

For technical content specifically, swap @type: "Article" for @type: "TechArticle". The remaining fields stay identical. LLMs treat TechArticle as a stronger signal that the page contains technical content suitable for technical-question answers.

Schema 2: BreadcrumbList — the cheap citation lift

Every page deeper than / emits a BreadcrumbList. This is the cheapest single schema to implement (it is a function of the URL path and never changes after deployment) and it correlates with citation lift on category-broad queries because LLMs use it to understand which articles belong to which topic on your site.

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://boringtechnologies.com/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Articles",
      "item": "https://boringtechnologies.com/articles"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "JSON-LD for content sites",
      "item": "https://boringtechnologies.com/articles/schema-deep-dive-content-sites"
    }
  ]
}

Notes that matter:

position is 1-indexed and starts at the homepage.
The last item has the current page’s full URL. Some old examples on the web omit the item field for the final list element; that is wrong for 2026, every position needs item.
name should match the visible breadcrumb in the page UI if one exists. If your site has no visible breadcrumb, emit the schema anyway — LLMs read it without rendering it.

Implementation tip: build a small helper that takes the page path plus a category-name lookup table and emits the list. Templating it once and reusing across the site is ~30 lines of code in any framework.

Schema 3: FAQPage — only when the questions are real

FAQPage is the most-abused schema on the web. The 2018-2022 SEO playbook was: take a paragraph, rewrite it as three fake Q&A pairs, emit FAQPage, claim AEO points. LLMs caught up. Fake FAQs now get filtered out of citation selection in most providers and there is rumored future penalty risk in Google AI Overviews specifically.

The rules for emitting FAQPage in 2026:

There must be three or more questions. If there are fewer than three, do not emit it.
The questions must be questions people actually ask, not keyword variants padded as questions.
The answers must be standalone — readable on their own, without context from the surrounding article.
The questions and answers in the schema must match the visible questions and answers on the page. Hidden FAQ schema with no visible counterpart is a flag.

The shape:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Do I need both Article and BreadcrumbList schema on every article page?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Article schema attributes the content; BreadcrumbList places it in your site's topic graph. Together they roughly double the citation rate compared to Article alone."
      }
    },
    {
      "@type": "Question",
      "name": "Does FAQPage schema still help in 2026 with the manipulation crackdown?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes when the FAQ is genuine — three or more real questions, answers that stand alone, visible on the page. The crackdown targets fake FAQs padded for ranking; real ones still earn citations."
      }
    },
    {
      "@type": "Question",
      "name": "How do I test whether my JSON-LD is being picked up?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Run validator.schema.org against the URL for a syntactic check. Then wait 7 to 14 days and poll Perplexity for a category-relevant query. If your page starts surfacing, the schema is working."
      }
    }
  ]
}

Implementation notes:

The name field is the question text. Keep it natural, not keyword-stuffed.
The Answer.text field allows light HTML in some specs but plain text is the safer choice for LLM parsing.
Three questions is the floor, not the ceiling. Six to eight well-chosen questions is the sweet spot for an article that genuinely contains an FAQ. More than 12 starts to look like padding.

A small enforcement pattern that helps in code: a buildFAQ(questions) helper that returns null when questions.length < 3. This makes it impossible to ship a degenerate FAQ schema by accident.

Schema 4: HowTo — only when the page is actually a how-to

HowTo schema describes a procedural article: an ordered list of steps the reader follows to accomplish a task. Tutorials, runbooks, recipes, setup guides — these are real HowTo candidates. A think-piece with three numbered sections is not.

The shape:

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to add Article and BreadcrumbList JSON-LD to a static-site article page",
  "description": "Step-by-step procedure for emitting Article and BreadcrumbList JSON-LD on a static-site article template.",
  "totalTime": "PT30M",
  "step": [
    {
      "@type": "HowToStep",
      "position": 1,
      "name": "Build the Article schema object",
      "text": "Construct an object with @context, @type Article, headline, description, datePublished, author as a Person, publisher as an Organization, inLanguage, and mainEntityOfPage."
    },
    {
      "@type": "HowToStep",
      "position": 2,
      "name": "Build the BreadcrumbList schema object",
      "text": "Construct an itemListElement array with one ListItem per breadcrumb step, position-indexed from 1, with name and item URL on every element."
    },
    {
      "@type": "HowToStep",
      "position": 3,
      "name": "Emit both schemas in the document head",
      "text": "Render each schema as a separate script tag of type application/ld+json inside the head element, before the closing head tag."
    },
    {
      "@type": "HowToStep",
      "position": 4,
      "name": "Validate with the schema.org validator",
      "text": "Submit the published URL to validator.schema.org and confirm zero errors against both schemas before treating the change as shipped."
    }
  ]
}

What matters:

totalTime uses ISO 8601 duration syntax (PT30M is 30 minutes, PT2H is 2 hours). Skip if the time is not meaningful for the procedure.
Each step has position (1-indexed), name (a short imperative phrase), and text (the actual instruction).
Steps should be discrete and sequential. If your “steps” are really subsections of an essay, the page is an article, not a how-to. Use Article schema instead.

Sites that genuinely benefit from HowTo: documentation sites, technical tutorials, runbook collections. Sites that do not: opinion blogs, brand thought leadership, case studies.

What we do not use

A list of schemas content-site engineers commonly ask about and the reason each is skipped on a typical content site:

WebSite — adds nothing the homepage Organization schema does not already provide. Optional.
WebPage — implicit. LLMs treat every URL as a webpage by default. Emitting WebPage schema on every page is pure noise.
ImageObject — emit only inside the parent schema (the Article’s image field) when the image is a real, meaningful hero. A standalone ImageObject schema does not earn citation lift.
Review and AggregateRating on the brand’s own pages — manipulation flag. If reviews exist, link to a third- party review platform and let it emit the schema there.
VideoObject — only when there is genuine video content. Most content sites in 2026 do not have it.
Organization on every page — emit on the homepage and the about page only. Other pages reference it via publisher inside the Article schema.

How to verify the work

A three-step diagnostic that takes under 10 minutes per URL:

View the rendered page source. Search for application/ld+json and count the script tags. On an article page expect three to five blocks (Article, BreadcrumbList, plus Person and Organization references depending on how your templates compose them).
Run the published URL through validator.schema.org. Zero errors. Warnings on optional fields are fine.
Wait 7 to 14 days. Then poll Perplexity for a category-relevant query. If your article surfaces in the answer, the schema is doing its job. If not, the most common cause is not the schema — it is that the article does not yet match the query intent strongly enough to be cited. Iterate on the article, not the schema.

Where to see this in production

The exact shapes above are the schema we run on every Boring Technologies article including this one. View source on this page, search for application/ld+json, and you will see the Article, BreadcrumbList, Person, and Organization blocks. The Cristina Constantin site — /cristina living lab — uses the same shapes adapted for the Romanian language and the legal-content domain (with an added LegalService Organization on the homepage).