Schema markup for AI search helps engines like ChatGPT, Perplexity, and Google AI Overviews parse your content and confirm it as a real entity, but it won’t guarantee a citation. Google says no special schema is required for AI Overviews, and a 2026 Ahrefs study found JSON-LD produced no meaningful citation lift. Ship it for clarity, not as a hack.
What Schema Markup Actually Does for AI Search
Schema markup is a block of structured code, almost always JSON-LD, that sits in a page’s <head> and tells search and AI systems explicitly what a piece of content is: an article, a product, a set of instructions, an organization. It translates human-readable HTML into machine-readable facts.
Here’s what most 2026 AEO content skips: when AI systems fetch a page in real time, they mostly read the visible text, not the hidden schema. Ahrefs tested five major AI systems (ChatGPT, Claude, Perplexity, Gemini, Google AI Mode) during live retrieval and found none of them used JSON-LD, hidden Microdata, or hidden RDFa. They extracted only what a person would see on screen.
So what does schema actually do? Two things, and neither is “guarantees a citation.”
- It speeds up understanding. Structured data removes ambiguity for the crawlers building the index AI systems draw from later, including Googlebot, GPTBot, and ClaudeBot. Instead of inferring a page is a how-to guide from headings and formatting, the crawler is told directly.
- It builds entity identity. Organization and Article schema link a page to a verified entity, your business, your author, through
sameAsreferences to Wikipedia, Wikidata, or verified social profiles. That entity graph is what AI systems and Google’s Knowledge Graph use to judge whether a source is trustworthy enough to quote.
Neither of those is the same as “add FAQPage schema and get cited by ChatGPT.” That claim shows up constantly in 2026 AEO content. The best available data doesn’t support it, and we’ll walk through why below.
The 5 Schema Types Worth an Agency’s Time
Out of dozens of schema.org types, five do almost all of the useful work for AI search visibility: FAQPage, Article, Organization, HowTo, and BreadcrumbList. Each does something different, and none of them is a silver bullet.
| Schema type | What it tells AI/search engines | Current status |
|---|---|---|
| FAQPage | Marks discrete Q&A pairs so engines can extract a self-contained answer | No longer produces a Google rich result (deprecated May 2026); markup is still valid and still helps structure Q&A content for extraction |
| Article | Identifies headline, author, publish/update date, and publisher, the freshness and authorship signals AI systems weigh for trust | Fully supported; still drives rich results and is read by most crawlers |
| Organization | Establishes your business as a verified entity, linked via sameAs to Wikipedia, Wikidata, LinkedIn, Crunchbase |
Foundational for Knowledge Graph and AI entity recognition |
| HowTo | Marks numbered steps for process or tutorial content | Google’s rich result carousel was deprecated in September 2023; markup is still valid and the step structure still helps AI parse process content |
| BreadcrumbList | Communicates site hierarchy and how a page fits into the larger site | Still an active Google rich result; helps AI systems understand topical context |
Use this table as the rollout order too: Organization and Article first, since they’re foundational and touch every page; FAQPage and HowTo next, only where content genuinely fits the format; BreadcrumbList last, since it’s usually a site-wide template change rather than a page-by-page job.
What the 2026 Data Actually Shows
The claim that “schema-rich sites get priority in AI Overviews” circulates constantly in SEO content, usually credited to “a 2026 study” with no link attached. We looked for a primary source, including checking for an IDC 2026 report on the topic, and couldn’t find one. Treat it as unverified until someone can point to a traceable methodology.
What is verified: Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, against a control group of 4,000 pages that changed nothing. Google AI Mode citations moved +2.4%, ChatGPT +2.2%, both statistically indistinguishable from zero, and Google AI Overviews citations moved -4.6%, which Ahrefs calls statistically significant and in the opposite direction from what most AEO advice promises.
There’s a real correlation in the same dataset: pages AI systems already cite are roughly three times more likely to carry JSON-LD than the average page. But correlation isn’t causation here. Sites that invest in structured data also tend to invest in technical SEO, authoritative content, and link building. The schema marks a well-run site; it isn’t proven to cause the citation.
Why FAQPage Schema Is More Complicated Than It Looks
FAQPage schema is the type most often recommended as “highest-impact for AI extraction.” The logic makes sense on the surface: a clean Q&A block mirrors how people phrase questions to ChatGPT and Google AI Mode, so it’s an easy pattern for an AI system to match and quote.
But Google deprecated FAQ rich results entirely on May 7, 2026, the final step in a process that began in August 2023, when it first restricted FAQ (and HowTo) rich results to well-known government and health sites. Google gave no public explanation and didn’t tie the move to AI search. FAQPage schema.org markup is still valid and won’t hurt a site; the Q&A structure is still a reasonable content pattern. What’s gone is the visible rich result, the little accordion in the SERP.
For agencies, the practical read: keep using FAQPage on pages that genuinely answer distinct questions, because the structure still helps with AI parsing and costs almost nothing to add. Don’t sell it to clients as a guaranteed citation lever. On the current evidence, it isn’t one.
The Implementation Checklist
An agency schema rollout works best as a repeatable checklist, not a one-off audit. Run this per client site, in order:
- Audit what’s already there. Use Google’s Rich Results Test and the Schema.org Validator. Most client sites have partial, broken, or duplicate schema before you touch anything.
- Ship Organization schema site-wide, including
name,logo,url, andsameAslinks to every verified profile: LinkedIn, Crunchbase, Google Business Profile, Wikipedia if applicable. - Add Article schema to every blog post and resource page, with accurate
datePublished,dateModified, andauthorfields tied to a real, named person. - Add BreadcrumbList across the site so crawlers and AI systems understand where a page sits in the hierarchy.
- Use FAQPage only where content is genuinely Q&A, five to eight real questions phrased the way a person would type them into ChatGPT, not a fabricated list added to fish for schema credit.
- Use HowTo only on true step-by-step content, and set expectations it won’t produce a visible Google rich result anymore.
- Validate every deploy. Schema errors are common after CMS updates and template changes, so check quarterly, not once.
- Track outcomes, not intentions. Schema is a hygiene layer. Knowing whether it moved the needle for a client requires a dedicated AI citation tracking process, since there’s no rich-result report for AI Overviews the way there was for FAQ.
The Honest Take on llms.txt
llms.txt is a plain-Markdown file, hosted at yoursite.com/llms.txt, that gives large language models a condensed summary of a site’s most important pages. Jeremy Howard of Answer.AI proposed it in September 2024, modeled loosely on robots.txt and sitemap.xml but written in Markdown, since the intended reader is a language model, not a crawler bot.
Google’s position is direct. John Mueller has called it “purely speculative for now,” pointing out that the file has existed for two years and no AI system actually requires it. Asked directly whether Google’s use of llms.txt on some of its own developer properties counts as an endorsement, Mueller answered on Bluesky with one word: “no.” Google’s own AI-features documentation states plainly that no special files, including llms.txt, are needed to appear in AI Overviews or AI Mode.
The usage data backs him up. Ahrefs analyzed 137,210 domains in May 2026 and found that about 28% publish an llms.txt file, but 97% of those files got zero requests from any AI bot that month. Of the fetches that did happen, GPTBot and Claude-Code (Anthropic’s coding tool, not a consumer search engine) were the top requesters, and roughly 12% came from GEO/AEO tools and researchers checking the file, not production AI systems serving real users.
So there’s no validated proof llms.txt lifts citations, and Google has said as much on the record. That’s the straight answer.
The reasonable case for shipping it anyway is smaller than the hype, but it isn’t nothing:
- It costs almost nothing. A well-structured llms.txt takes an hour or two once you know a site’s key pages.
- Adoption keeps growing even though usage is thin today. Anthropic, Cloudflare, and Vercel already publish one, and if agent-based browsing expands, being ready early costs less than retrofitting later.
- It forces a useful exercise. Writing a clean summary of a site’s most important pages is good discipline, whether or not any bot reads the file today.
What we don’t recommend: selling llms.txt as a citation strategy, or pricing it like real structured data or content work. Position it as a low-cost hygiene item, not a deliverable with a measurable return.
Where This Fits in a Real AI Search Strategy
Schema markup is a foundation, not a strategy. It makes your content easier to parse and your entity easier to trust. It doesn’t replace structuring content so engines can quote it or building the kind of citation-worthy content generative engines pull from.
In our experience working with agencies, this is where most schema rollouts stall: the code ships, nobody checks whether it changed anything, and the client eventually asks what they’re paying for. That’s less a technical problem than a packaging and reporting one, and it’s part of how agencies sell AI-era SEO services to clients in a way that survives a renewal conversation.
Frequently Asked Questions
Does schema markup help you get cited by ChatGPT and AI Overviews?
Indirectly. Schema helps crawlers understand and index your content correctly, and it builds the entity signals, like Organization and author data, that AI systems check for trust. It doesn’t directly cause a citation. A 2026 Ahrefs study found no meaningful citation lift from adding JSON-LD alone.
What is the most important schema type for AI search?
Organization schema is the best starting point because it establishes your business as a verified entity across the web. Article schema is a close second for content-heavy sites, since it carries the authorship and freshness data AI systems use to evaluate trust.
Is FAQPage schema still worth using in 2026?
Yes, but not for the reason most content claims. Google removed the visible FAQ rich result from Search in May 2026, so there’s no SERP benefit left. The markup is still valid, and the Q&A structure still helps AI systems extract self-contained answers, so it’s worth keeping on pages that genuinely answer distinct questions.
Do I need llms.txt for my website?
No AI system requires it. Google’s John Mueller has called it “purely speculative for now,” and Ahrefs data shows 97% of published llms.txt files get zero requests from AI bots. It’s a low-cost, optional addition, not a requirement.
What schema types does Google recommend for AI Overviews?
None specifically. Google’s own AI-features documentation states there’s no special schema.org markup required to appear in AI Overviews or AI Mode. The requirement is the same as normal Search: the page needs to be indexed and eligible for a standard snippet.
Does HowTo schema still work in Google Search?
The markup is still valid, but Google deprecated the visible HowTo rich result carousel in September 2023, on both desktop and mobile. Keep using HowTo schema for genuinely instructional content because the step structure still helps AI parsing, but don’t expect it to produce a rich result in Google Search anymore.
How do I check if my schema markup is valid?
Use Google’s Rich Results Test and the Schema.org Validator. Run both after any CMS update or template change, since broken or duplicate schema is one of the most common technical issues we find auditing client sites.
How long does it take an agency to implement schema across a client site?
For a typical mid-size site, Organization and Article schema can go live in under a week once templates are set up correctly. Site-wide BreadcrumbList and page-by-page FAQPage or HowTo review usually take longer, depending on how many templates and content types the site has.
Ship the Structure, Then Prove It Worked
Schema markup is cheap, low-risk, and worth doing well. It isn’t the lever that gets a client cited by ChatGPT on its own, and an agency that sells it that way loses that conversation at renewal. The real lever is the whole system: entity clarity, content structured to answer real questions, and a way to prove what’s happening in AI search results.
If you’d rather have a partner run that system behind your brand, from schema and entity work through content and citation tracking, that’s what we build for agency partners.