Schema Markup for GEO: Which Schemas Actually Matter

Schema markup is one of the most over-talked, under-implemented elements of modern SEO. Most agencies either skip it entirely or deploy 12 different schemas hoping something sticks. Start with the uncomfortable finding: Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 against 4,000 control pages, and adding schema boosted AI citations on no platform — Google AI Overviews moved −4.6%, AI Mode +2.4%, ChatGPT +2.2%. If you were sold schema as an AI-citation lever, you were sold a story.

Read the scope before you rip out your JSON-LD, though. Those 1,885 pages were already earning 100+ AI Overview citations before schema was added, which Ahrefs says plainly. So the finding is precise: schema does not lift pages that are already visible to AI. It is not evidence that an unmarked, entity-ambiguous site gains nothing from finally telling engines who it is. That distinction is the whole reason four schemas still belong on your site — as hygiene, not as a lever.

This post is the operator manual for schema-for-GEO in 2026. For broader context, see the GEO pillar. For passage-level work that pairs with schema, see the inverted pyramid post.

Which schemas actually drive AI citations?

Four schemas. Organization (with sameAs and complete address/contact info) at the site level — typically in the layout component. Article (with author, datePublished, dateModified, headline) on every blog post and resource page. FAQPage nested under Article on pages with Q&A sections. Service or LocalBusiness on commercial pages. That's it. Everything else is secondary or noise for GEO purposes.

• Tier 1 (deploy on every relevant page): Organization + sameAs, Article + author + dateModified, FAQPage nested under Article, Service or LocalBusiness on commercial pages.
• Tier 2 (deploy where it makes sense): BreadcrumbList for site-structure understanding, Review/AggregateRating paired with LocalBusiness, Person schema for authors.
• Tier 3 (mostly cosmetic for GEO, useful for SERP rich results): VideoObject, Recipe, Event, Product.
• Skip entirely: Speakable (effectively dead since Google Assistant news clips deprecated), Course (very narrow use), JobPosting (only for hiring pages), QAPage (different from FAQPage and rarely correctly implemented).

Put plainly, here is what each schema actually buys you — and what the evidence says it does not:

The honest verdict: deploy the top four once, validate them, and never think about schema again. It is a one-week hygiene job with a real but modest payoff, and it is the single most over-sold line item in GEO proposals. If an agency's AI-search pitch leads with schema, they are selling the cheapest part of the work as though it were the hardest.

Why does the Article + FAQPage compound pattern matter?

Because it tells AI engines two things at once: "this is an editorial piece with an author" and "this contains Q&A content that maps directly to user questions." Don't over-claim its citation impact, though — Ahrefs' controlled study of 1,885 pages found schema alone didn't move AI citations. The value of the compound pattern is clean entity and editorial signaling plus SERP rich-result eligibility, on top of genuinely well-structured content. Schema supports good content; it doesn't substitute for it.

Implementation matters: the FAQPage should be nested as a sub-entity inside the Article schema, not deployed as a separate JSON-LD block. The FAQ questions should match the H2s on the page exactly (Google's quality check compares them). The Article needs author, datePublished, dateModified, headline, and mainEntityOfPage at minimum.

What does a proper Article + FAQPage JSON-LD look like?

A minimal correct implementation has the Article schema as the root entity with author, dateModified, datePublished, and headline, plus a mainEntity property pointing to a FAQPage object whose mainEntity is the array of Question/Answer pairs. The FAQ questions inside the schema must exactly match the questions surfaced on the page. The structure is well-documented at schema.org/Article and schema.org/FAQPage.

• Root @type: Article (or BlogPosting, which is a subtype of Article).
• Required Article properties: headline, author (with @type Person), datePublished, dateModified, mainEntityOfPage.
• Nested FAQPage as a child entity (not a separate JSON-LD block).
• FAQPage.mainEntity: array of Question objects, each with name (the question) and acceptedAnswer (with @type Answer and text).
• Question count: 5-8 per page. Below 5 reads as token compliance. Above 8 dilutes the signal.
• Match FAQ questions to actual H2s or near-H2s on the page. Mismatches get demoted by Google's quality check.

What is the role of Organization + sameAs?

Organization schema with a complete sameAs property (LinkedIn, Crunchbase, Facebook, X, Wikipedia if applicable, industry-specific directories) builds the entity graph AI engines use to identify your brand. Without it, the engines may treat your domain as anonymous and discount its authority. With it, your brand becomes a known entity that gets pattern-matched across the open web.

The sameAs property should include 5-12 high-quality external URLs that all point to your verified brand presence. Avoid stuffing low-quality directory links — quality matters more than count. Deploy this in your site layout component so it appears in the JSON-LD on every page.

What about LocalBusiness and Service schemas?

LocalBusiness schema on location pages and Service schema on service pages are the two commercial-intent schemas that move both classic SEO rich results and AI citation rates. LocalBusiness needs complete NAP (name, address, phone), opening hours, geo coordinates, and serviceArea. Service needs name, description, provider (linking back to your Organization), and serviceType.

Pair Service with AggregateRating only if you have legitimate, verifiable reviews. Fake or scraped reviews can trigger Google's review-spam policies and lead to manual actions against the whole domain. The penalty is real — and recovering from a manual action is far more expensive than simply never emitting unverifiable rating schema in the first place.

What schemas should I skip?

Speakable, JobPosting (unless hiring), Course (unless an actual course), QAPage (rarely implemented correctly), and any schema you can't legitimately back with on-page content. This isn't a folk rule — Google's own AI-features guidance explicitly lists 'making sure your structured data matches the visible text on the page' among its recommendations. Markup that describes content a human visitor cannot see is the one schema mistake with real downside: it is the definition of a mismatch, and it puts your whole domain's structured-data credibility on the line to win a rich-result snippet you probably don't need.

Speakable specifically: it was designed for Google Assistant news clips, which Google has effectively deprecated. There is no current product surface where Speakable schema delivers user-visible results. Deploying it does nothing for GEO and adds maintenance overhead.

How do you validate schema correctly?

Two validators. Google's Rich Results Test (search.google.com/test/rich-results) for what Google itself parses. Schema.org's validator (validator.schema.org) for strict schema.org spec compliance. Run both on every page before deployment. The Google validator catches issues that affect SERP rich results. The schema.org validator catches strict spec violations that affect AI engine parsers (which often use schema.org spec more strictly than Google does).

• Run Google's Rich Results Test on every commercial page after schema deployment.
• Run schema.org's validator for strict spec compliance.
• Check for warnings as well as errors. Warnings degrade signal quality even when they don't break parsing.
• Re-validate after every content update — schema can silently break when properties shift.
• Monitor Search Console's "Enhancements" report for ongoing schema health.

What's the 30-day schema deployment plan?

Week 1: deploy Organization + sameAs in the site layout. Week 2: roll out Article + author + dateModified across the top 25 blog/resource pages. Week 3: add nested FAQPage to the top 10 Q&A pages. Week 4: deploy Service or LocalBusiness on the commercial pages. Validate everything in Rich Results Test as you go. Budget roughly 8-16 hours of developer time for a 50-page site, once — this is a build-and-forget job, not a retainer line item.

Set the expectation before you start: do not run this sprint expecting AI citations to climb, because the best available evidence says they won't move on that basis alone. Run it because rich-result eligibility, entity disambiguation, and clean editorial signals are worth having, and because the marginal cost is one developer-week. Then spend the rest of your budget on the passage rewrites in the inverted pyramid post, which is where the citation lever actually is.

If you want this done for you — including the validation, the page-by-page audit, and the monthly schema health monitoring — book a strategy call. For the full GEO context, see our GEO service. For dental practices specifically, dental SEO covers the schema patterns built into that vertical's program.