Schema Markup for AI Search: What Google Actually Reads

If you're an SMB owner being pitched an "AI schema package" in 2026, here's the short answer before the invoice lands: Google has stated, in its own documentation, that structured data is not required for generative AI features and there is no special schema.org markup to add for them. Read that sentence twice. Then read it on Google's own page: the AI features and your website doc says exactly that.

Schema still matters. Just not the schema most agencies are selling. This piece walks through what Google actually reads for AI Overviews and AI Mode, which schema types still earn their keep on an SMB site, and why the visible text on your page now does more work than the JSON-LD wrapping it. If you want the broader SEO context first, our SEO services page covers the full stack.

TL;DR

• Google's own docs confirm no special markup is required for AI Overviews or AI Mode.

• FAQ rich results have been restricted to government and health sites since August 2023.

• Four schema types still matter for SMBs: Organization, Service, LocalBusiness, BreadcrumbList.

• AI systems retrieve from the same Search index, so visible answer text outranks markup.

• Audit your current schema in 10 minutes before paying anyone to "upgrade" it.

1. What Google says, verbatim

Google's AI features documentation, published as part of Search Central, contains a passage that should end most of the schema-for-AI debate on the spot. The page tells site owners that structured data isn't required for generative AI features, and that there is no special schema.org markup to add. It then notes that schema is still useful for classic rich results.

Google's separate generative AI optimization guide closes the loop. It confirms that AI Overviews and AI Mode draw from the same Search index that powers classic ranking. Standard indexing, standard SEO, standard content quality. The same plumbing, with a generative layer on top.

So when an agency tells you their proprietary schema bundle will get you cited in AI Overviews, you can ask one question: which Google document are you reading that I'm not? There isn't one. The retrieval path runs through the regular index, and the markup that helped you win a featured snippet in 2022 is the same markup that helps you get cited in an AI answer in 2026.

For a deeper read on getting cited specifically, see our piece on how to get cited in Google AI Overviews.

2. The FAQ schema story: restricted since August 2023

This is the one that won't die. FAQPage schema stopped being a general rich-result play almost three years ago, and agencies are still selling it.

In August 2023, Google restricted FAQ rich results to authoritative government and health sites. Search Engine Land's retrospective on FAQ schema walks through the timeline. If your site isn't a state health department or a federal agency, your FAQPage markup is no longer producing the accordion snippet in regular search. It hasn't been for nearly three years.

Is FAQ schema actively harmful? No. It's just inert for rich results on a commercial site. The reason it gets sold as an "AI search optimization" tactic is that the conversation moved on, the markup stayed the same, and the upsell language got rebranded. "FAQ schema for AI Overviews" sounds new. It isn't.

What replaces it: clear, scannable Q&A copy in the visible body of your page. The AI systems that summarize your content are reading the rendered HTML, not interrogating your JSON-LD for ground truth. We'll come back to this in section 4.

3. Schema that still earns its keep

For a typical SMB site, the schema types worth shipping in 2026 are narrow and boring. Boring is the right answer here.

Schema types ranked by what they actually do in 2026, per Google's documentation.

A few notes on the four that matter:

Organization sets your brand identity in Google's knowledge graph. Logo, name, social profiles, sameAs links. One block, sitewide, in the root layout. Ship it once and forget it.

LocalBusiness earns its keep if you have a physical location or service area. Address, hours, geo coordinates, phone. This is the schema that actually drives map-pack visibility and "near me" intent, both of which feed AI Mode's local responses.

Service describes what you sell in machine-readable form. For an agency, a clinic, a contractor, this clarifies the offering when Google's systems are trying to match query intent to your page. It's quiet work, but it compounds.

BreadcrumbList is the unglamorous one. It tells Google your site hierarchy, which improves how URLs render in results and helps crawlers understand context. Ten minutes to implement, runs forever.

Notice what's not on the list: HowTo (restricted to mobile in 2023, then largely deprecated), FAQPage (covered above), Article on every blog post (useful, but not load-bearing for AI retrieval). If your agency's "AI schema package" leads with these, that's the tell.

4. Visible FAQs vs FAQPage markup

Here's the part most operators miss. The AI systems summarizing your content are doing extractive and generative work on the rendered text of your page. The JSON-LD is metadata, not the source of truth.

Which means: a visible FAQ section, written in plain question-and-answer prose under an H2, will do more for your AI retrieval than the same content wrapped in FAQPage schema and hidden in markup. The answer text is what gets pulled. The markup is, at this point, decorative.

This flips the usual SEO instinct. For a decade, the move was "hide the structure, wrap it in schema, let Google parse it." In 2026, the move is the opposite: write the answer plainly, make it scannable, put it where a human can read it on the page. The AI layer reads what the human reads.

One practical implication: if you've been told to add FAQ schema without also adding visible FAQ content, you're doing the work backwards. Write the questions and answers into the page first. Add markup later, if at all.

This is the same logic behind why we wrote What is llms.txt and does it actually do anything. Same principle, different file.

5. Implementing JSON-LD without a plugin mess

On a typical small-business WordPress site, schema gets installed by three different plugins fighting each other, each emitting overlapping JSON-LD blocks. You end up with two Organization entries, three BreadcrumbList declarations, and a FAQPage block that's been inert since 2023.

The minimal SMB setup, in practice:

Pick one schema source. If you're on WordPress, that's usually your SEO plugin (Yoast, Rank Math, or SEO Press all emit valid Organization, LocalBusiness, and BreadcrumbList markup). Disable schema output in every other plugin. In our client work, deduplicating schema sources alone resolves about 80% of structured-data warnings in Google Search Console.

For headless or custom stacks, ship JSON-LD as a static block in your root layout. One Organization. One BreadcrumbList component that renders per page. LocalBusiness on the contact page if applicable. Service schema on each service page.

Validate with Google's Rich Results Test and Schema.org's validator. If both pass, you're done. Stop adding more.

The instinct to ship more schema is the same instinct that fills agency invoices. Resist it. The marginal AI-retrieval impact of your eighth schema type is approximately zero, per Google's own documentation.

6. The 10-minute audit of your current markup

Before you pay anyone for schema work, run this:

1. Open Google's Rich Results Test and paste your homepage URL. Note which schema types are detected.

2. Repeat for your top service page and your contact page.

3. Check Search Console under Enhancements for any structured data errors or warnings older than 30 days.

4. Count how many Organization blocks render on your homepage. If more than one, you have a plugin conflict.

5. Confirm LocalBusiness is present on your contact page if you have a location.

If all five pass, your schema is fine. If your FAQ markup is being flagged as inert, that's expected, not broken. If an agency is proposing $1,500 of structured-data work and your audit comes back clean, the schema isn't the constraint. The content is. The information architecture is. The internal linking is.

Schema is plumbing. Plumbing that works gets noticed only when it doesn't.

FAQ

Does schema help AI Overviews?

Not directly. Google's AI features documentation states that structured data is not required for generative AI features and no special schema.org markup exists for them. AI Overviews retrieve from the same Search index as classic results, so general SEO quality matters more than AI-specific schema work.

Is FAQ schema deprecated in 2026?

FAQPage schema isn't deprecated as a spec, but Google restricted FAQ rich results to government and health sites in August 2023. For a typical commercial SMB site, FAQ schema no longer produces the accordion snippet in regular search and has no documented AI-feature benefit either.

What structured data should I use for LLMs?

Google has not published any LLM-specific schema requirement. The same Organization, LocalBusiness, Service, and BreadcrumbList markup that supports classic search visibility also feeds AI retrieval, because both pull from the same index. Visible answer text on the page matters more than wrapper markup.

Will adding more schema types improve my AI search visibility?

Probably not. Per Google's AI optimization guide, AI Overviews and AI Mode use the standard Search index. Adding obscure schema types past the core four typically generates Search Console warnings without changing retrieval behavior. Audit before adding.

Should I remove FAQ schema from my site?

No urgent need. Inert FAQ markup isn't harmful, just decorative. If you're cleaning up plugin conflicts or reducing markup bloat, removing it is fine. If it's stable and validates, leave it and focus your time on visible Q&A content that AI systems can actually read.

Where this leaves you

If you take one action this week: open Rich Results Test, paste your top three pages, and see what's actually there. Ten minutes. No invoice required.

If the audit surfaces something messy and you want a second read, drop us a line. No pitch deck, just the actual answer.