Schema Markup Isn’t Optional Anymore: A No-Nonsense Guide to Getting Cited by AI

A no-nonsense guide to schema markup in 2026: what Google switched off, what the data actually says about AI citations, the four schema types worth your time, and how to implement and validate them properly.

The short version

• Google switched off FAQ rich results for every site on 7 May 2026. Writing FAQ blocks to chase a SERP feature is now wasted effort.
• Schema isn’t an AI-citation shortcut. Ahrefs’ test of 1,885 pages found no major uplift on any platform, and a small drop on AI Overviews.
• Its real job now is quieter: telling Google and AI systems who you are, who wrote the page and when it was last updated.
• Four types do most of the work: Organization, Article, Product/Service and Person. Keep every field in sync with what’s visible on the page.

On 7 May 2026, Google switched off FAQ rich results for every site on the web. No blog post, no warning shot — just a documentation edit ending a three-year retreat that started in 2023, when FAQ snippets were restricted to government and health sites only. By August, the FAQ filter, the Rich Results Test support, and the Search Console reporting for it are gone too. If you built an SEO habit around FAQ schema, the reward it was built for no longer exists.

At the same time, and this is the part most “schema is dead” takes miss, reporting on Google’s March 2026 core update points the other way: Gemini-powered AI Mode reads structured data more than it used to — not to decide what to show as a snippet, but to verify claims, establish entity relationships, and assess whether a source is credible. And a much bigger piece of evidence landed in between: Ahrefs ran a controlled test on 1,885 pages and found that adding JSON-LD produced no meaningful citation lift on ChatGPT, Google AI Mode, or Google AI Overviews.

So which is it — more important than ever, or a waste of a developer’s afternoon? This is the technical follow-up to AEO vs GEO vs SEO, and it exists because “add schema, get cited” is the single most oversold claim in AEO right now. Here’s what the data actually shows, what schema still does for you in 2026, and exactly what to implement instead of FAQ blocks nobody asked for.

The two schema stories colliding in 2026

Two things are true about schema markup in 2026, and they pull in opposite directions.

The first is decline. FAQ rich results are gone as of 7 May, and they’d already been losing ground for years — the March 2026 core update reportedly cut FAQ rich-result impressions by close to half in the run-up to the removal, and tightened HowTo eligibility so that markup describing supplementary steps, rather than a page’s primary content, stopped qualifying at all. If you were writing FAQ sections purely to earn a rich snippet, that trade no longer exists.

The second is a quieter shift in the opposite direction. That same March update is reported to have changed how Gemini-powered AI Mode treats structured data internally — using it not to populate a visual SERP feature, but to cross-check claims on the page, disambiguate who’s making them, and weigh whether the source is trustworthy enough to draw on. Schema stopped being a way to win a visible spot in the results and started being read, quietly, as supporting evidence for a decision the AI has already made about you.

Neither of those is a reason to panic or to gold-plate every page with markup. They’re a reason to be precise about which schema still earns its keep — which is what the rest of this is for.

What the data actually shows

Most AEO advice on schema is asserted, not tested. Ahrefs actually ran the experiment: they added JSON-LD to 1,885 pages that already had 100+ AI Overview citations, matched them against roughly 4,000 control pages, and tracked citation changes across three platforms between August 2025 and March 2026, using four separate statistical methods including difference-in-differences.

The results, in their words: “Adding schema produced no major uplift in citations on any platform.”

• Google AI Overviews: −4.6%, and this one was statistically real (odds of it being random were about 1 in 2,500) — but the cause is unexplained, and the study can’t say schema itself caused the drop.
• Google AI Mode: +2.4%, indistinguishable from noise.
• ChatGPT: +2.2%, also indistinguishable from noise.

Ahrefs also found that 53% of AI-cited pages carry schema markup — a number that sounds like proof until you notice it’s correlation, not causation. Sites that implement schema properly tend to also nail their content structure, internal linking, and technical fundamentals. The markup isn’t doing the work; it’s a marker of a site that’s doing everything else right too.

Separate testing on how AI systems actually retrieve pages backs this up from a different angle: several 2026 tests found that most AI crawlers tokenize a page as plain text rather than semantically parsing JSON-LD, and that content which exists only inside structured data — not also in the visible copy — gets extracted at close to a zero rate. If a fact isn’t in your sentences, wrapping it in a schema tag doesn’t put it in front of the model.

Worth flagging the study’s own caveats, because they matter for what you do next: it only tested pages already receiving heavy AI citations, so it says nothing about whether schema helps a currently invisible page get noticed for the first time. It also pooled every schema type into one number — Organization, Article, Product, and FAQ together — so a specific type could still be pulling weight the average hides.

Why schema still isn’t optional — just not for the reason you’ve been sold

Reframe the question. Schema was never going to be the thing that gets a sentence lifted into a ChatGPT answer — your visible copy does that, or it doesn’t. What schema still does, in 2026, is three narrower jobs that have nothing to do with citation volume:

• Entity disambiguation. Organization and Person schema with sameAs links to LinkedIn, Crunchbase, and Wikidata/Wikipedia tell Google’s Knowledge Graph — and by extension the systems that lean on it — unambiguously who you are. This is the “entity clarity” factor from the GEO side of the equation, and it’s markup-dependent in a way your body copy can’t fully substitute for.
• Claim verification inside AI Mode. If reporting on the March 2026 update holds, Gemini’s AI Mode is now cross-checking visible claims against your structured data as a trust signal. That doesn’t get you cited more often — but a mismatch between what your page says and what your markup says can quietly knock you out of the credible-source pool it’s choosing from.
• Everything schema always did. Voice assistant answers, Knowledge Panels, and the rich results that survived the FAQ/HowTo cuts — Product pricing and availability, Review stars, Recipe, Event — all still run on the same markup they always did.

None of that is glamorous, and none of it will show up as a citation-rate chart you can present to a client. It’s the same category of work as the technical SEO fundamentals from the AEO/GEO piece: necessary infrastructure, not a lever you pull for an overnight result.

The four schema types worth your time in 2026

Skip the exhaustive schema.org catalogue. For most SME and e-commerce sites, four types cover almost all of the real leverage left in structured data.

A minimal Organization block, done properly:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "WEBKANDY",
  "url": "https://www.webkandy.co.uk",
  "logo": "https://www.webkandy.co.uk/logo.png",
  "sameAs": [
    "https://www.linkedin.com/company/webkandy",
    "https://www.instagram.com/webkandy"
  ]
}

And Article schema, on a blog post like this one:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Schema Markup Isn't Optional Anymore",
  "author": { "@type": "Person", "name": "Steve Bell" },
  "publisher": { "@type": "Organization", "name": "WEBKANDY" },
  "datePublished": "2026-09-13",
  "dateModified": "2026-09-13"
}

Two fields do most of the work in that second block: a real author and an accurate dateModified. Both are exactly the trust signals AI systems are reportedly weighing when they decide whether to draw on a source at all.

What to strip out

• FAQ sections written only to chase a rich result. That result doesn’t exist anymore. Keep FAQs where they answer a genuine question your body copy hasn’t already covered — drop the rest.
• HowTo schema on supplementary steps. If the steps you’re marking up aren’t the primary content of the page, Google’s March 2026 tightening treats that as a mismatch risk, not just a missed eligibility check.
• Any markup that says something your visible copy doesn’t. The extraction research above says AI systems mostly ignore markup-only content anyway — and post-March-2026, a mismatch is now something Google’s update is reported to actively flag rather than quietly ignore.

You don’t need to go back and delete old FAQPage markup that’s still accurate — Google’s own guidance is that unused structured data doesn’t cause problems for Search. The instruction is narrower: stop adding more of it for a reward that’s gone, and stop letting it drift out of sync with the page underneath it.

Implementation playbook: matching markup to what’s actually on the page

1. Audit before you add anything. Pull every schema type currently live on the site (Rank Math’s site audit or a crawler will list them) and delete or fix anything that no longer matches the visible page.
2. Ship Organization schema once, sitewide, with a complete sameAs list — every LinkedIn, Crunchbase, and Wikidata profile that’s genuinely yours.
3. Add Article schema to every post, with a real author and a dateModified you actually update when you meaningfully revise the content — not on every typo fix.
4. Add Product/Service schema to commercial pages only, and keep price and availability fields identical to what’s rendered on the page, because these are the rich results that survived and mismatches here risk manual action.
5. Keep FAQPage markup only where the Q&A is genuinely on the page, in close to the same words as the markup — not a rewritten, keyword-stuffed version of it.
6. Re-check after every content update. A stale dateModified or a fact that’s changed on the page but not in the markup is worse for trust than having no schema at all.

Schema checklist: Organization + sameAs, sitewide · Article with a real author and an honest dateModified · Product/Service on commercial pages only · FAQPage only where the Q&A is visible on the page · re-validate after every content update.

How to validate it properly

• Schema.org’s validator for raw syntax errors — catches malformed JSON-LD before anything else does.
• Google’s Rich Results Test for the types still eligible for a visual result (Product, Review, Article); note that FAQ and HowTo testing support is being withdrawn from the tool alongside the rich results themselves.
• A manual side-by-side read, line by line, of every field in your JSON-LD against the visible text on the page. This is the check most sites skip, and it’s the one that matters most now that mismatches carry a credibility cost rather than just a missed snippet.
• Search Console’s Enhancements report to catch validation errors at scale once the page is live, not just at build time.

The bottom line

Schema markup isn’t the AI-citation shortcut a lot of AEO content sold it as — Ahrefs’ own numbers say adding it to a page that’s already visible barely moves the needle. But it isn’t dead either. Google pulled the visible reward for it in the same update that reportedly turned it into a quieter trust and disambiguation signal inside AI Mode.

The brands that treat schema as reinforcement — accurate Organization and Article markup sitting on top of content that already says what the markup claims — get the credibility benefit even though it’ll never render as a star rating again. The ones still writing FAQ blocks to game a feature that stopped existing in May are optimising for a search result that no longer draws.

Schema markup FAQ

Is schema markup still worth adding in 2026?

Yes, but for different reasons than it used to be. FAQ rich results are gone, so schema’s value now sits in entity clarity, authorship and freshness signals, and the commercial rich results (price, availability, reviews) that are still eligible.

Does schema markup help you get cited by ChatGPT or AI Overviews?

Not directly. Ahrefs added JSON-LD to 1,885 already-cited pages and found no major uplift on any platform. Your visible copy earns the citation; schema helps machines understand who’s saying it.

Should I delete my old FAQPage schema?

Not if it’s still accurate. Google’s guidance is that unused structured data doesn’t cause problems. Just stop adding more of it for a reward that no longer exists, and keep what’s there in sync with the page.

Which schema types should an SME start with?

Organization (with a full sameAs list), Article on every post, Person for named authors, and Product or Service on commercial pages. That covers most of the value for most sites.

Get your technical foundations right before AI reads them

WEBKANDY helps e-commerce and music brands get their technical foundations — schema, site structure, entity clarity — right before the content strategy sits on top of it, through a one-off SEO/AEO/GEO audit and roadmap, a fractional retainer, or full delivery. Get in touch to find out what’s actually worth fixing on your site.

Book a schema audit call →