Schema Markup for AI Search: Structured Data That Gets Cited

Schema for AI search plays a different role from schema for rankings: instead of decorating your listing with stars and prices, it tells retrieval systems exactly what your company is, what you sell and why your facts can be trusted enough to cite. It’s a core layer of our AI search strategy for South Africa, building on the basics in our plain-language schema explainer.

The reason this matters now: AI systems compose replies from sources they can parse with confidence. Two SA pages can carry identical facts, yet the one wrapped in clean structured data gets understood — and quoted — while the other gets guessed at.

Schema for AI Search: Which Types Actually Get Cited?

Five structured data types do most of the citation work: Organization, FAQPage, Article, LocalBusiness and Product or Service. Each solves a different trust problem for the machine assembling an answer, which is why priority beats coverage — marking up everything poorly loses to marking up these five properly.

A note on FAQPage: Google retired most FAQ rich-result displays back in 2023, and plenty of SA agencies concluded the markup was dead. The display died; the parsing didn’t. Question-and-answer pairs in structured form remain one of the cleanest extraction surfaces you can hand a retrieval system — which is why every one of Growth Pulse Media’s 382 posts still ships with one.

How Do AI Systems Actually Use Structured Data?

AI systems use structured data for three jobs: disambiguating entities, corroborating facts, and lifting content with confidence. Google’s own structured data documentation describes it as how systems understand not just the page, but the people, companies and things the page describes.

Entity disambiguation is the underrated one for SA companies. “Apex Solutions” could be forty businesses; an Organization block with sameAs links to your LinkedIn, your registration listing and your official profiles collapses that ambiguity to one. Machines cite companies they can identify without risk of naming the wrong one.

Corroboration works the same way it does for earning AI referrals generally: when your markup, your visible copy and independent sources all state the same facts — same name, same services, same pricing — the machine’s confidence crosses the citation threshold. Contradictions anywhere drop you below it.

Provenance rounds out the trio. Article markup carrying a real author name, credentials and a dateModified stamp tells the machine a claim has an accountable origin and a freshness signal. Anonymous, undated pages force the system to treat every statement as unverified — attributable content is simply cheaper for it to trust, and cheap-to-trust is what gets quoted.

What Should SA Businesses Mark Up First?

Start with one site-wide Organization block, then work down: service or product pages with Rand pricing, your highest-traffic informational pages with FAQPage, then Article markup with genuine author details. That order front-loads entity clarity — everything else compounds on top of the machine knowing who you are.

The Organization block deserves an hour of care. Include your exact trading name, your Johannesburg or other SA location, and sameAs links to your LinkedIn company page, your Google Business Profile and any authoritative directory listings. Define it once, site-wide — duplicating slightly different Organization blocks across pages sends contradictory identity signals, which is worse than sending none.

For businesses serving specific regions, the LocalBusiness layer earns its keep on city-level questions. Include the physical address with province, the areas you serve, and opening hours — the concrete geographic anchors that let a machine answer “who does this in Sandton” with your name instead of a hedge. Service businesses covering all of Gauteng should say so explicitly; machines don’t infer coverage, they read it.

How Do You Implement It on a WordPress Site?

On WordPress, the practical split is: let Rank Math generate the automatic layer, then add the high-value blocks manually as JSON-LD. Rank Math handles Article, breadcrumbs and a basic Organization out of the box; the citation-earning extras — FAQPage per post, enriched Organization with full sameAs, Service blocks with Rand pricing — go in as hand-placed script blocks.

One WordPress-specific caution before the pattern: page builders and caching layers occasionally strip or reorder script blocks. After first setup, view the live page source in a private window and confirm your blocks survived the render — what the editor shows and what the crawler receives are not always the same document.

JSON-LD is the format to use for all of it — Google explicitly recommends it over the older inline formats because it sits in one self-contained block, separate from your visible HTML, and scales without breaking. This is exactly how schema for AI search gets maintained across a large site: one pattern, pasted per post, validated once.

The workflow we run on our own site: write the FAQ section in the visible content first, mirror it word-for-word into a FAQPage block, paste that below the article in the code editor, and re-sync whenever the visible answers change. The sync step is the one teams forget — and drift between markup and page is the fastest route to being ignored.

Validation closes the loop. Run new templates through Google’s Rich Results Test, fix what it flags, and re-test after any theme or plugin change. Broken structured data — a stray comma, curly quotes pasted from a document — fails silently: the page looks fine while the machine reads nothing.

Which Mistakes Get Structured Data Ignored?

Four mistakes account for most wasted markup: describing content that isn’t visible, contradicting yourself across pages, marking up fabricated reviews, and shipping broken syntax. All four end the same way — the machine discards your signals and falls back to guessing.

The review one carries extra risk in SA’s small markets. Ratings markup must reflect genuine, verifiable reviews; inflated or invented ratings invite manual actions that suppress the whole page. And syntax errors are more common than teams expect — smart quotes pasted from Word documents are the classic silent killer, invalidating an entire block over one character.

A quarterly cadence keeps all four risks controlled. Re-validate templates after theme or plugin updates, re-check that prices and answers in markup still match the visible page, and spot-check three key pages in the Rich Results Test. Fifteen minutes a quarter versus months of silently discarded signals — the maths argues for the calendar entry.

There’s also a strategic mistake worth naming: treating markup as a substitute for content quality. Getting schema for AI search right amplifies strong pages; it cannot rescue weak ones. The content-structure work and the markup work are two halves of the same job, in that order.

What Does It Look Like When It Works?

When it works, the first visible change is that machines stop misdescribing you — then citations follow. The composite below shows the typical shape for an SA professional-services firm implementing the priority stack over three months. Figures illustrate the pattern, not a single client’s audited results.

Fold this tracking into the same monthly prompt log you keep for referral monitoring — one extra column noting whether each reply described the firm correctly. The two habits share a spreadsheet and a half-hour ritual, and together they catch both kinds of drift: the market moving and your own signals decaying.

The first row is the one to watch early. Accuracy improvements show within weeks of the Organization and sameAs work landing, well before citation counts move — it’s the leading indicator that the entity work took, and the cheapest confidence you’ll buy in this whole discipline.

The GPM Differentiator: 382 Posts of Practised Markup

Growth Pulse Media doesn’t recommend structured data patterns — we run them. Every post on our 382-post site ships with hand-synced FAQPage JSON-LD, our Organization entity resolves cleanly across the major AI platforms, and our own Search Console shows the machine-read behaviour these signals feed. Our AEO services for South African businesses implement the same stack, with the same validation discipline, on client sites.

The operator detail that matters: we’ve watched which blocks moved grounding signals on real SA queries and which were decorative. That’s the difference between a markup checklist and a markup strategy — and it’s why our implementations start with entity clarity, not with whatever a plugin generates by default.

Who This Is NOT For

Questions SA Businesses Ask About Structured Data and AI

Is schema for AI search different from normal schema markup?

Same vocabulary, different priorities. Classic implementations chase rich-result displays like stars and prices; AI-focused implementations prioritise entity clarity, provenance and extraction — Organization with sameAs, author details, question-shaped markup. The syntax is identical; what you emphasise and validate changes.

Does FAQ markup still matter now that Google removed FAQ rich results?

Yes. The visual rich result was retired for most sites in 2023, but the markup remains valid structured data that retrieval systems parse. Question-and-answer pairs in machine-readable form are among the cleanest extraction surfaces for AI replies — the display died, the citation value didn’t.

Which schema type matters most for a small SA business?

Organization, implemented once and well. Entity clarity is the gate everything else passes through: a machine that can’t confidently identify your company won’t cite it regardless of how good your other markup is. Add sameAs links to your LinkedIn, Google Business Profile and key directories.

Can a plugin do all of this automatically?

Partly. Rank Math and similar plugins generate the Article, breadcrumb and basic Organization layer well. The citation-earning blocks — enriched Organization with full sameAs, per-post FAQPage synced to visible answers, Service markup with Rand pricing — need hand-placed JSON-LD and a validation habit.

How do I check whether my structured data is working?

Two layers: technical and behavioural. Technically, run pages through Google’s Rich Results Test and fix flagged errors. Behaviourally, ask the major AI platforms about your company monthly and log whether descriptions are accurate — accuracy improvements are the earliest signal the entity work has landed.

Will bad structured data hurt my rankings?

Broken syntax simply gets ignored, costing you the benefit rather than rankings. Deceptive markup is different: describing invisible content or fabricating reviews violates spam policies and can trigger manual actions that suppress the page. Honest-but-imperfect is safe; dishonest is not.

Still filing structured data under “technical stuff for later”? Later is when your competitors’ entity blocks have already taught the machines whose facts to trust in your category.

Dirk van Greuning

Founder of Growth Pulse Media and a specialist in South African search dominance. Dirk translates his experience in scaling South African businesses into high-velocity digital strategies for B2B and retail leaders. He writes about SEO, lead generation, and paid media from an operator’s perspective — prioritising pipeline value over impressions.

Connect with Dirk on LinkedIn