Schema Markup for the AI Search Era: What Still Matters

Does schema markup still matter for AI search?

Yes, but less than most GEO advice claims. Schema markup helps machines understand what a page is, who wrote it, and when it changed, and Google still reads it for the rich results it supports. It is not an entry ticket to AI answers: Google says there's no special schema.org structured data you need to add to appear in AI Overviews or AI Mode. And the two types GEO guides pushed hardest, FAQ and HowTo, no longer earn rich results in Google at all. Implement the basics, validate them, then spend your real effort on content.

Under classical SEO, schema powered rich results: star ratings, recipe cards, FAQ accordions. That layer keeps shrinking. In August 2023 Google limited FAQ rich results to "well-known, authoritative government and health websites", and on September 13, 2023 it stopped showing HowTo rich results. Per Google's documentation changelog, FAQ rich results stopped appearing on May 7, 2026, and the FAQ documentation was removed on June 15, 2026.

Under GEO, the case for schema is narrower than the hype. Structured data gives engines unambiguous labels for facts that are easy to get wrong: the author, the dates, the organization behind the page. Google says it uses structured data to understand the content of the page. What schema doesn't do is mark a passage as "the answer." Google's AI optimization guide calls overfocusing on it a myth: structured data isn't required for generative AI search, though it still helps with rich result eligibility.

That's why schema is the floor of GEO, not the ceiling. The Princeton GEO paper (Aggarwal et al., November 2023) measured effect sizes for citing sources, adding statistics, and adding quotations, but didn't test schema. We haven't seen a controlled study that isolates schema's effect on LLM citations.

Treat schema as foundational hygiene. The highest-leverage work is still adding statistics, sourcing claims, and restructuring for extraction.

The schema types that matter most for GEO

Six schema types cover almost everything; FAQPage and HowTo have dropped from "must have" to "where the content fits." Skip the rest until these are right.

1. Article and BlogPosting

The baseline schema for any content page, and still fully supported by Google. Engines use these fields to attribute authorship, establish recency, and disambiguate which page is canonical for a topic.

{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "Schema Markup for the AI Search Era",
  "description": "Which schema types still matter for AI Overviews, ChatGPT Search, and Perplexity.",
  "image": "https://example.com/images/schema-ai-search.jpg",
  "author": {
    "@type": "Person",
    "name": "Jane Doe",
    "url": "https://example.com/team/jane-doe"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Veritas",
    "logo": {
      "@type": "ImageObject",
      "url": "https://example.com/logo.png"
    }
  },
  "datePublished": "2026-05-08",
  "dateModified": "2026-09-29",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://example.com/blog/schema-markup-ai-search-era"
  }
}

The fields that matter most: author (Google strongly recommends the author's type plus a url or sameAs link), datePublished and dateModified (which Google says gives it more accurate date information), and mainEntityOfPage (disambiguates the canonical URL when content syndicates).

2. FAQPage (no longer a Google rich result)

This used to be the first type every GEO guide recommended. The Google side of that case is gone: FAQ rich results stopped appearing on May 7, 2026. FAQPage is still a valid schema.org type, and Google says unused structured data doesn't cause problems for Search. So if a page has a real FAQ section, marking it up is a cheap, clean label for any parser. Just don't add FAQs for the sake of markup, and don't count on a citation lift.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is Generative Engine Optimization?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Generative Engine Optimization (GEO) is the practice of structuring content so that large language models choose to cite or surface it when answering a user's question. The term was introduced in a November 2023 paper by Aggarwal et al. at Princeton."
      }
    },
    {
      "@type": "Question",
      "name": "Is GEO replacing SEO?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No. GEO is an additional optimization layer for AI-driven answer surfaces; SEO still governs the open-web SERP and remains the largest source of organic traffic for most domains."
      }
    }
  ]
}

A few rules:

• Keep each Answer.text answer-first and self-contained, roughly 50–150 words. Long, narrative answers don't extract cleanly, marked up or not.
• Don't duplicate FAQ blocks across pages. Boilerplate Q&A adds nothing; vary the questions.
• Your FAQPage schema must match what's visible on the page. Google's guidelines: don't mark up content that isn't visible to readers.

3. HowTo (no longer a Google rich result)

HowTo schema with each HowToStep marked up gives parsers an explicit, ordered step list. Google stopped showing HowTo rich results in September 2023 and dropped it from the Rich Results Test, so its value now is purely machine-readable structure. Use it only where the page genuinely is a procedure.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to set up GEO citation tracking",
  "description": "A four-step process to measure your brand's citation rate across major AI engines.",
  "totalTime": "PT30M",
  "step": [
    {
      "@type": "HowToStep",
      "name": "Define seed queries",
      "text": "Identify 30 to 50 representative queries that map to your target audience's actual research patterns."
    },
    {
      "@type": "HowToStep",
      "name": "Run queries across engines",
      "text": "Execute each query in Google AI Overviews, ChatGPT Search, and Perplexity. Record which domains appear in the citation chips."
    },
    {
      "@type": "HowToStep",
      "name": "Tag your brand mentions",
      "text": "For each result, note whether your brand was named in the answer text, whether your URL was cited, and which competitors were mentioned alongside."
    },
    {
      "@type": "HowToStep",
      "name": "Track weekly",
      "text": "Re-run the seed query set weekly. Citation patterns shift faster than SERP rankings, so weekly cadence is the minimum useful resolution."
    }
  ]
}

4. Organization (homepage only)

Google recommends Organization markup on your home page or one about page; it's how engines build entity recognition for your brand. The most under-implemented field is sameAs: links to verified profiles on Wikipedia, Wikidata, LinkedIn, Crunchbase, GitHub, and your social channels.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Veritas",
  "url": "https://www.getveritas.io",
  "logo": "https://www.getveritas.io/logo.png",
  "description": "Veritas generates marketing content from your knowledge graph with mandatory citations on every claim.",
  "sameAs": [
    "https://www.linkedin.com/company/veritas",
    "https://x.com/veritas",
    "https://github.com/veritas",
    "https://www.crunchbase.com/organization/veritas"
  ],
  "contactPoint": {
    "@type": "ContactPoint",
    "email": "hello@optivustechnologies.com",
    "contactType": "customer support"
  }
}

Each sameAs entry points to another page about the same organization. Engines can cross-reference these to build their understanding of who you are and what you do.

5. BreadcrumbList

Lower-impact but worth implementing, and still a Google-supported type. Helps engines understand site hierarchy and disambiguate pages that target related queries.

6. Person (for content authors)

Authors with linked credentials make attribution unambiguous. Give each author bio page jobTitle, worksFor, and profile links, wrapped in ProfilePage markup, which Google's Article docs recommend for internal author pages.

{
  "@context": "https://schema.org",
  "@type": "ProfilePage",
  "mainEntity": {
    "@type": "Person",
    "name": "Jane Doe",
    "jobTitle": "Head of Content",
    "worksFor": {
      "@type": "Organization",
      "name": "Veritas"
    },
    "url": "https://example.com/team/jane-doe",
    "sameAs": [
      "https://www.linkedin.com/in/janedoe",
      "https://x.com/janedoe"
    ]
  }
}

Schema you can probably skip

A short list of schema types that get a lot of recommendation airtime but don't move the needle much for GEO:

• Review and AggregateRating. Useful for e-commerce SERP rich results, less impact on LLM citation outside of explicit "is X any good" queries.
• Event. Only matters if you actually run events.
• VideoObject. Implement on YouTube content where it lives, not on your blog where it doesn't.
• Recipe. Irrelevant unless you're a food brand.

Don't waste implementation cycles here when Article and Organization aren't dialed in.

The emerging llms.txt standard

In September 2024, Jeremy Howard, co-founder of Answer.AI, proposed /llms.txt as a standardized way for websites to provide curated information to LLMs, and he published a revised v2 on August 10, 2026. It's a markdown file at your domain root (v2 also allows one at a subpath, such as /docs/llms.txt, covering the pages under it), the LLM equivalent of robots.txt but with a fundamentally different purpose: not "what should crawlers avoid" but "here's a curated map of our most important content, in a format LLMs can ingest cleanly."

The specification:

• An H1 with your project or site name. This is the only required section.
• A blockquote summary of what the entity is and does.
• Optional detail paragraphs or lists (anything except headings).
• H2 sections containing "file lists" of links, each optionally followed by a note.
• By convention, an "Optional" section for secondary links an agent can skip when context is short.

An example /llms.txt:

# Veritas

> Veritas generates marketing content from your knowledge graph with
> mandatory citations on every claim and structured output ready for
> AI-engine extraction.

## Documentation

- [Product overview](https://www.getveritas.io/features/content-generation)
- [SEO Intelligence](https://www.getveritas.io/features/keyword-research)
- [Visual Assets](https://www.getveritas.io/features/visual-assets)
- [Pricing](https://www.getveritas.io/pricing)

## Guides

- [Generative Engine Optimization complete guide](https://www.getveritas.io/blog/generative-engine-optimization-complete-guide)
- [How to get cited in ChatGPT, Perplexity, and AI Overviews](https://www.getveritas.io/blog/how-to-get-cited-chatgpt-perplexity-ai-overviews)

## Optional

- [Changelog](https://www.getveritas.io/changelog)
- [Customer stories](https://www.getveritas.io/customers)

A companion convention, /llms-full.txt, developed by Mintlify with Anthropic, holds the full content of those pages in one markdown file. The v2 proposal leans instead on a clean .md version of each page at the same URL.

Adoption status (September 2026): the v2 proposal reports that thousands of sites publish an llms.txt file, and OpenAI, Anthropic, and Gemini publish them for their developer docs. Mintlify rolled it out across every docs site it hosts in November 2024. It is still not an official standard in the way robots.txt is: there's no IETF RFC. And Google now says plainly that Google Search ignores llms.txt files; one "will neither harm nor help your site's visibility or rankings in Google Search."

The honest verdict: implement llms.txt if you have substantial technical documentation or product reference content that coding agents fetch on demand. The cost is one markdown file. The upside is cleaner ingestion by the tools that read it; it is not an AI Overviews lever. For purely marketing sites, the effect is smaller.

Common mistakes that quietly break your schema

Implementation issues that are easy to overlook and hard to debug after the fact:

1. Missing required fields. Each schema type has required and recommended fields, and a page missing required ones isn't eligible for the matching rich result, so the markup is quietly wasted. Always run Google's Rich Results Test and Schema.org's validator on every page that gets new schema.

2. A vague or merged author. Google accepts a Person or Organization as author, but wants every author shown on the page in the markup, each in its own entry, with a url or sameAs link. "By Jane Doe, Head of Content" with a linked bio gives engines far more to work with than "By Veritas Editorial."

3. Stale or fake dateModified. Update the timestamp when you update content, and only then. Google's helpful-content checklist asks whether you're changing the date of pages to make them seem fresh when the content has not substantially changed.

4. Markup that doesn't match visible content. Your JSON-LD must match the visible HTML. A structured data issue can result in a manual action that removes rich result eligibility (Google says it doesn't affect web ranking).

5. Duplicate Organization schema across pages. Google says you don't need it on every page. Repeating it doesn't amplify the signal; it just adds copies to keep in sync.

6. JSON-LD rendered only via client-side JS. Google can read JSON-LD injected by JavaScript because it renders pages. Many AI crawlers don't: Vercel's December 2024 analysis found none of the major AI crawlers, including OpenAI's, Anthropic's, and Perplexity's, rendered JavaScript. Server-render your JSON-LD in the <head> or before </body>.

The validation workflow

To write the markup in the first place, our free schema markup generator reads a live page and drafts JSON-LD for the types above. Validate its output like anything else. Two free validators, one paid alternative:

• Google's Rich Results Test. Confirms Google can read your schema and which currently supported rich results the page qualifies for. It doesn't cover retired types like HowTo.
• Schema.org's validator. Structural validation against the full vocabulary, including FAQPage and HowTo.
• Schema testing in your CI pipeline. For sites that publish frequently, automate checks in the publish flow. Platforms like Schema App manage markup at scale; for simpler setups, a script that extracts each page's JSON-LD and checks required fields works.

A practical rule: invalid schema is wasted schema, and misleading schema is worse than none. Unused markup is harmless, but markup that misrepresents the page can trigger a manual action. Validation is not optional.

A schema implementation checklist

If you're starting from zero, the order of operations:

1. Organization schema on the homepage (or one about page) with complete sameAs links to all verified profiles.
2. Article or BlogPosting schema on every content page, with Person schema for the author.
3. ProfilePage and Person schema on every author bio page.
4. BreadcrumbList schema in your site template.
5. FAQPage schema on pages with a genuine FAQ section, as a machine-readable label. Don't add FAQs just to carry markup.
6. HowTo schema on true step-by-step content: setup instructions, playbooks.
7. /llms.txt at your domain root if you have technical or product reference content worth highlighting.
8. Validation in CI if you publish more than weekly.

Then measure citation rates before and after rather than assuming a lift.

Closing

Schema markup under GEO is closer to plumbing than to magic. Done right, it removes friction in the engine's path to understanding your content. Done wrong or skipped, it leaves engines guessing, and engines that guess often choose someone else's content.

It's foundational work that compounds quietly: validated markup keeps authorship, dates, and entities unambiguous for every system that reads your site. Worth doing. Worth validating. Worth maintaining.

---

When Veritas publishes a post to WordPress, it adds JSON-LD built from the page: Article or BlogPosting, BreadcrumbList, and FAQPage where the post has questions and answers. Try Veritas free or explore SEO Intelligence.

Related reading: Generative Engine Optimization (GEO): A 2026 Guide · How to Get Cited in ChatGPT, Perplexity, and Google AI Overviews · AI Overviews Killed Your CTR. Here's the New Playbook..