Schema Markup for AI Answers: What Actually Gets You Cited

Last updated: October 15, 2026

In May 2026, a study of 1,885 pages that added JSON-LD between August 2025 and March 2026 found citations moved by only +2.4% and +2.2% on two AI answer platforms (both within noise) and fell 4.6% on a third, measured against matched pages that never added markup.

My verdict: schema markup for AI search is useful plumbing, not a citation lever. It helps search engines understand who you are and what a page is. It does not make a model pick your page over a better one.

That still leaves real work. Below: the evidence, the four schema types a B2B SaaS site should keep, what schema can't do, and how to implement it without wasting a sprint.

What the evidence actually says

Three kinds of evidence exist right now: official guidance, controlled studies, and practitioner tests. They mostly agree, with a couple of gaps worth knowing about.

Google: helpful for rich results, not required for AI features

Google's guide to optimizing for generative AI features in Search, published in May 2026, is blunt. Under a heading about overfocusing on structured data, it says structured data "isn't required for generative AI search, and there's no special schema.org markup you need to add."

The same guide still says it's "a good idea to continue using it" because it helps pages qualify for rich results. And it groups "special" AI markup with llms.txt as things Google Search ignores. (If you're weighing that file too, see the llms.txt breakdown.)

Glenn Gabe summed up that mythbusting section on X on May 15, 2026, listing "Don't overfocus on structured data" next to "No llms.txt needed." That's a fair one-line reading.

The controlled study: adding schema didn't move citations

The study from the intro is the best data we have, so it's worth knowing how it worked. Ahrefs took 1,885 pages that added JSON-LD and matched them against 4,000 control pages that didn't. Then it compared citations in the 30 days before and after the change.

Two details matter more than the headline. First, AI Mode citations for these pages rose 43% before you subtract the controls. Control pages grew almost as fast, so the "lift" was just the platform growing. That's exactly the trap a simple before-and-after test falls into.

Second, the same research team found that AI-cited pages were almost three times more likely to have JSON-LD than pages that weren't cited, and 53% of cited pages carried it. That correlation is the stat you'll see on conference slides. The controlled test suggests it comes from better-maintained sites doing many things well, schema being one of them.

The fetch test: chatbots read your visible page, not your JSON-LD

searchVIU ran a cleaner, smaller experiment in late October 2025 (published December 2, 2025). It built a test page for a fictional product and hid different prices in different places: visible HTML, JavaScript, JSON-LD, hidden Microdata and hidden RDFa.

ChatGPT, Claude, Gemini, Perplexity and Google AI Mode were each asked for the prices. None of them found the price that lived only in JSON-LD. Prices in visible text, including visible Microdata and RDFa, were picked up by ChatGPT and Gemini.

The authors are careful about the limit here, and so should you be. They only tested live fetching. Schema could still be used when a search engine builds its index, and that's the layer Google's and Microsoft's AI answers draw on.

Microsoft: the most positive statement, and the oldest

Bing is the one major engine that has said its AI uses schema. At SMX Munich in March 2025, Fabrice Canel said schema markup helps Microsoft's LLMs understand content, as Search Engine Land reported on March 20, 2025.

I'd treat that carefully for two reasons. It's older than everything else here, and I couldn't find a newer Microsoft statement that adds detail. And "helps understand" is not "earns citations." It most plausibly means schema helps Bing's index, which Copilot then retrieves from.

Where the evidence is weak

Be honest with your team about what nobody has tested well yet:

• Pages that aren't cited at all. Every page in the controlled study already had 100+ AI Overview citations in February 2025. Schema might help new or obscure pages get understood, and we just don't know.

• Individual schema types. The study pooled Article, FAQ, Product, HowTo and Organization together. One type could help while another doesn't.

• Longer windows. Thirty days may be too short to catch a slow effect.

• Entity-level effects. Organization markup feeding a knowledge panel is a different mechanism from page-level citations, and it's hard to measure.

So the fair summary is this: no evidence that adding schema wins citations on pages already in the running, decent evidence that live-fetching chatbots ignore it, and open questions everywhere else.

Which schema types are worth it for B2B SaaS

My rule is simple. Add markup that describes something real and visible on the page, and that either earns a Google feature or clarifies who you are. Skip anything you're adding only because a checklist said AI likes it.

Organization: yes, on the homepage

This is the one I'd implement first. Google's Organization documentation says it helps Google understand your administrative details and disambiguate your company in search results. It has no required properties, so you add what's relevant.

For a SaaS company, that usually means:

• name, legalName and url

• logo

• sameAs links to your LinkedIn, X, G2, Crunchbase and GitHub profiles

• contactPoint for sales or support

Entity clarity is the part of schema most likely to matter for AI answers indirectly. If your product name is also a common word, consistent Organization data across your site and profiles makes it easier to tell you apart.

SoftwareApplication: yes, if the facts are on the page

Product is mostly built for physical goods and merchant listings. For a SaaS product page, SoftwareApplication (or WebApplication) is the more specific fit. Google's software app documentation requires a name, an offers.price (set it to 0 for free), and a rating or review.

That last requirement is where SaaS teams get into trouble. If you don't show genuine ratings on the page, you can't mark them up. Pulling a G2 score into JSON-LD without displaying it breaks Google's rule against marking up content readers can't see.

Your pricing page is the other good candidate, provided the prices in the markup match what's visible. An AI answer about "how much does X cost" will lean on the text, so keep both in sync.

Article and BlogPosting: yes, keep it boring

Article markup on blog posts is cheap and low-risk. Include headline, author (as a Person with a url to a real bio page), datePublished, dateModified and image.

The author and date fields are the useful ones. They make your freshness and authorship explicit. Just don't bump dateModified without changing the content. That's a small lie, and small lies in structured data add up.

BreadcrumbList: yes, sitewide

Breadcrumb markup describes where a page sits in your site, for example Home → Blog → AEO. It's easy to generate from your CMS templates and helps search engines understand your structure. Google's breadcrumb documentation only requires an ordered itemListElement.

Make sure the markup matches visible breadcrumbs on the page. If your design has none, add them for users first.

FAQPage: optional now, and never for AI alone

This is the one with the most confusion. Google limited FAQ rich results to authoritative government and health sites in August 2023. Then, as Search Engine Journal reported on May 10, 2026, FAQ rich results stopped appearing in Google Search entirely on May 7, 2026. Search Console reporting and Rich Results Test support were scheduled to go in June, and API support in August 2026.

Google's position on retired markup hasn't changed: unused structured data doesn't cause problems. So you don't need to rip out existing FAQPage markup.

You'll still see posts claiming AI platforms "actively extract" FAQ schema. I haven't seen a test that shows it. The searchVIU result points the other way for live fetching. If your FAQ content helps buyers, keep the visible questions and answers. Those are what models can quote. The markup itself is now close to decorative.

HowTo: skip it

HowTo rich results have been deprecated since September 2023, when Google stopped showing them on desktop after dropping them on mobile. That's older history, but it's the current state. There's no Google feature left to earn, and no evidence it helps AI citations. Write clear numbered steps in the page instead.

A quick reference

What schema can't do

Schema can't make thin content citable. Google's AI guide says unique, non-commodity content will influence your presence in AI search "more than any of the other suggestions" it gives. Markup describes a page. It doesn't improve it.

It can't get a page into AI answers that isn't indexed. Google's requirements for AI Overviews and AI Mode are the normal Search ones: a page must be indexed and eligible for a snippet. Fix crawling and indexing before you touch JSON-LD.

It can't carry facts the page doesn't show. Hiding a price, a feature list or a comparison in JSON-LD won't get it into ChatGPT's answer, based on the fetch tests. Put the fact in the visible text.

It can't stand in for third-party mentions. Models form views about your product from reviews, forums, comparison articles and docs across the web. Markup on your own site doesn't change what others write about you.

And it can't be measured in isolation easily. If you add schema and rewrite a page in the same week, you'll never know which change mattered.

Lily Ray shared Fili Wiese's take on X on May 22, 2026: schema, rich snippets and voice were "all still just SEO," rooted in retrieval from the index. I agree. Treat schema as part of the technical SEO you'd do anyway. For how that work maps across both disciplines, see AEO vs. SEO.

How to implement schema the right way

Here's the process I'd follow on a typical SaaS marketing site.

1. Audit what you already ship

Before adding anything, see what's there. Many CMS themes and SEO plugins already output Organization, WebSite and Article markup, sometimes twice.

1. Pull a sample of templates: homepage, product page, pricing, a blog post, a docs page.

2. View the raw page source and find every application/ld+json block.

3. List the types on each template and note duplicates or conflicts.

2. Generate markup from templates, not by hand

Hand-written JSON-LD drifts out of date the first time someone edits a page. Build it from the same CMS fields that render the visible content. Then the price in the markup can't differ from the price on the page.

Use JSON-LD. Google recommends it over Microdata and RDFa in its general structured data guidelines, and it's the easiest to maintain.

3. Server-render it

Put the JSON-LD in the initial HTML rather than injecting it with JavaScript after load. Google can render JavaScript, but the controlled study only measured schema in raw HTML and notes that AI crawlers seem to treat injected schema differently. Server-side output removes that doubt.

4. Connect your entities with @id

Give your Organization a stable @id (for example, https://yoursite.com/#organization). Then reference it as the publisher of every Article and the provider of your SoftwareApplication. Google's guidelines show this pattern for linking related items. It turns a pile of separate blocks into one consistent description of your company.

5. Validate twice

Run two checks, because they catch different problems:

• Google's Rich Results Test for eligibility on the types Google still supports.

• The Schema Markup Validator for general schema.org syntax, including types Google doesn't show as rich results.

After launch, spot-check pages with URL Inspection in Search Console to see what Google actually picked up.

6. Check your escaping

This one is new. As Search Engine Roundtable reported on August 21, 2026, Google now applies only one pass of HTML unescaping when it extracts JSON-LD. Double-escaped entities such as & are no longer cleaned up for you.

If your company name contains an ampersand, or your plugin escapes text twice, look at the raw output. Google's advice is to use standard JSON escapes or Unicode escapes like \u0026.

7. Measure honestly

If you want to know whether schema changes anything for your AI visibility, run a small test. Pick a handful of pages to change and a matched set to leave alone, change nothing else, and compare both groups after 30 days or more.

Track AI citations alongside normal search metrics. The guide on how to measure AI search visibility covers the tracking setup.

Common mistakes

These are the ones I'd look for first in any audit:

• Marking up reviews or ratings that aren't on the page. Google's guidelines forbid marking up content hidden from readers, and in July 2026 Google added explicit language against fake or undisclosed incentivized reviews in review snippets, as Barry Schwartz noted on X.

• Prices in JSON-LD that no longer match the pricing page. Someone updates the page, nobody updates the markup.

• Several plugins each outputting their own Organization block with different names or logos.

• FAQPage markup on every page, wrapped around questions nobody asked, added in the hope AI will quote it.

• Treating a "valid" result in a validator as proof the markup is accurate. Validators check syntax, not truth.

• Adding schema and rewriting content in the same sprint, then crediting the schema for any change.

• Spending a week on exotic types while the page still has thin copy or indexing problems.

Final thoughts

Schema markup for AI search sits in an awkward spot. Vendors talk about it like a cheat code, and the best available test says it barely moves citations for pages already being picked up.

My view: implement Organization, SoftwareApplication, Article and BreadcrumbList properly, keep them accurate, and then stop. Put the rest of your time into content that's worth citing and into the visible facts models actually read. If you're new to this area, start with what AEO is and how GEO fits in.

Miraa is a B2B SEO agency whose services include technical SEO and AEO/GEO work, which is where structured data audits usually sit.