// HOW-TO · AI SEARCH

How to implement structured data for AI visibility (2026)

Implement structured data for AI visibility: build an entity-first schema stack, mark up pages and media in JSON-LD, match it to the page, and validate.

Last verified · 2026-10-01 · by Moe Ameen

Structured data will not turn AI citations on by itself — Google is explicit that it is helpful, not required, and that there is no special schema for AI Overviews or AI Mode. What it does is remove ambiguity: it gives answer engines a labeled, machine-readable copy of what your page already says and a verifiable claim about who published it, so a retriever parses your facts and your identity without guessing. That is a real lever for AI visibility across Google's surfaces, Bing and Copilot, ChatGPT, and Perplexity — an indirect one, worth doing well.

This walkthrough is the implementation order that compounds. It is entity-first on purpose: the Organization and Person markup pays off across every page, so you build that backbone before you touch a single article. Then you map each page type to a small, honest schema stack, add the video and image markup most sites skip, align everything to the visible page, and validate. If a crawler cannot reach your content in the first place, fix that in [make content visible to AI search](/how-to/make-content-visible-to-ai-search) before any of this matters. For the mechanism behind why each engine treats schema differently, read the guide on [structured data for AI visibility](/guides/structured-data-for-ai-visibility); for the citation-specific tactics, see [using schema markup to get cited by AI](/how-to/use-schema-markup-to-get-cited-by-ai).

The steps

Start with the entity, not the page. Define one Organization block — exact legal name, logo, one-line description, and a sameAs array linking the profiles that corroborate you (LinkedIn, Wikipedia or Wikidata, Crunchbase, your verified social accounts) — and reference it site-wide. This is the single highest-value piece of structured data for AI visibility because it pins down who you are, and it benefits every page at once. Build the backbone before you mark up any individual article.
Add Person markup for every named author. Give each author a Person block with a real, linked bio and their own sameAs array, and tie it to the Organization. An engine that sees the same author described consistently across your byline, their LinkedIn, and their other writing treats them as a verifiable identity it can name — which is exactly the provenance ChatGPT, Perplexity, and Google's AI surfaces weigh before leaning on a claim. A named, corroborated author is worth more than any content-level tweak.
Map each page type to a small, honest schema stack. Match the markup to what the page actually is: Article or BlogPosting for editorial content, Product, Service, or LocalBusiness for commercial and local pages, QAPage or FAQPage for a genuine question set. Most pages deserve two or three types nested together — an Article that references its author Person and publisher Organization — not every type you can find. Never mark up a page as something it is not to chase a feature.
Mark up your video and images — the lever most sites skip. As engines read media directly, VideoObject and ImageObject markup is moving from optional to load-bearing. VideoObject declares a video's title, description, thumbnail, upload date, and a transcript or caption track the engine can read as text; ImageObject ties an image to a real entity so it is eligible for visual and multimodal answers. If you publish video or imagery, this is where the easy, under-contested AI visibility is in 2026.
Write it as JSON-LD and nest entities by @id. Use JSON-LD — a single <script type="application/ld+json"> block — not inline microdata or RDFa. It is the format Google recommends, the cleanest for a machine to lift whole, and it keeps the markup separate from the HTML so it is easy to audit. Give each entity an @id and reference them together rather than repeating the same Organization in every block, so your author, publisher, and content form one connected graph.
Make every value match what a human sees. This rule overrides every tactic: never let the markup claim something that is not on the visible page. If your JSON-LD lists a price, rating, author, or date, that exact value must appear in the rendered content too. Engines cross-check markup against the page and against your other surfaces; a mismatch reads as manipulation and gets your structured data discounted or flagged — the opposite of the trust you were trying to build.
Validate before you ship, then deploy across the CMS. Run every template through the Schema Markup Validator on schema.org and Google's Rich Results Test, fix all errors and warnings, and confirm the parsed types match your intent — malformed JSON-LD parses to nothing, so a clean validation is the floor, not the finish. Then roll the markup out at the template level in your CMS so new pages inherit it automatically, and re-check after any CMS or plugin update that might alter the output.
Monitor AI visibility, not rich-result reports. Rich-result reports tell you less as those features retire — Google dropped FAQ rich results for every site on May 7, 2026, finishing a phase-out that began in 2023. What matters now is whether the engines retrieve you correctly. Re-test your target questions in ChatGPT, Perplexity, Gemini, and Google AI Overviews over the following weeks, and when a model gets a fact wrong, trace it to the stale value or mismatched field and repair both page and markup.

Common gotchas

Marking up pages before the entity. The Organization and Person sameAs backbone is the highest-value schema for AI visibility and it helps every page — build it first, not after a pile of per-page Article blocks.
Treating schema as a citation switch. Google says it is helpful, not required, and there is no special schema for AI Overviews. Controlled tests found little standalone lift on already-cited pages; it removes ambiguity, it does not manufacture authority.
Claiming what the page does not show. A price, rating, author, or date in your JSON-LD that is not visible on the page gets the markup discounted or flagged. Every value needs a visible counterpart.
Skipping video and image markup. VideoObject and ImageObject are the under-used levers now that engines read media directly — ignoring them leaves the easiest 2026 visibility on the table.
Adding FAQ or HowTo markup expecting a rich result. Google retired HowTo rich results and restricted FAQ rich results in 2023, then dropped FAQ rich results for every site on May 7, 2026. Keep FAQPage only where it matches real Q&A and only for machine parsing.
Never validating. Malformed JSON-LD parses to nothing, so a parser gets zero from a broken block. Run the validators after every template change.

Where Kompozy fits

This tutorial is technical, per-page work — entity blocks, nested JSON-LD, media markup, validation — and the reason most teams never finish it is not that the steps are hard. It is that the people who would do the schema work are the same people hand-producing every blog post, video, and carousel, so the technical layer stays permanently next-in-line. Kompozy removes that bottleneck from a different direction than the markup itself: it is a generation and multi-platform publishing engine, so one topic brief yields the whole set — Blog Articles, Text Posts, Carousel Posts, Infographic Photos, and [Persona Shorts](/glossary/persona-shorts) — behind a single review seat instead of a team of hands. That is the capacity you reassign to the entity and validation work this page describes.

It also does part of the page-level markup for you. Blog Articles generated in Kompozy already ship their own Article and FAQPage JSON-LD at render time for WordPress and Custom Webhook destinations, so the per-page layer in step three is partly handled on the pages Kompozy produces; on GHL Blog that markup is stripped before publish because GHL emits its own Article schema and two copies would only create duplicate-content noise. And the VideoObject and ImageObject lever in step four only matters if you publish media worth reading — Kompozy generates the [Persona Shorts](/glossary/persona-shorts) and HeyGen avatar video whose transcripts feed the markup, plus the Infographic Photos that carry your key facts as liftable visual units.

The one thing a JSON-LD block cannot do is keep your entity facts identical everywhere an engine looks — the sameAs corroboration in step one is a claim your cross-channel presence has to back up. The [Persona Brief](/glossary/persona-brief) fixes your name, positioning, and brand facts once so every output inherits them, and [autopilot](/glossary/autopilot) schedules the approved set across eight social platforms plus blog and email through a per-post review gate, so the entity your markup declares stays consistent on the surfaces schema never reaches. Starter ($199/mo, 5,500 credits) fits a solo creator hardening one brand; Pro ($499/mo, 18,000 credits) suits a business producing across every channel weekly; Enterprise is custom for agencies implementing this for many clients.

Frequently asked questions

Does structured data improve AI visibility?

Yes, indirectly and as a helper. Structured data makes a page machine-readable and its entity verifiable, which lets answer engines parse your facts and confirm who you are without guessing. Google confirms it is helpful but not required for generative AI features and that there is no special schema for AI Overviews. It removes ambiguity across Google's surfaces, Bing and Copilot, ChatGPT, and Perplexity — it does not rank you or rescue a thin or uncrawlable page.

What schema should I implement first?

Organization and Person with sameAs arrays, referenced site-wide. This entity backbone is the single highest-value structured data for AI visibility because it pins down who you are and corroborates your identity across the open web, and it benefits every page at once. Add Article or BlogPosting, Product, Service, or LocalBusiness, and VideoObject or ImageObject per page type after the backbone is in place.

Do I need VideoObject and ImageObject markup?

If you publish video or imagery, yes — it is the most under-used lever in 2026. As AI engines read media directly rather than only its surrounding text, VideoObject (with a readable transcript or caption track) and ImageObject tied to a real entity make your media eligible for visual and multimodal answers. It is not required, but it is low-competition visibility most sites have not claimed yet.

How do I verify my structured data is working?

Validate first: run each template through the Schema Markup Validator and Google's Rich Results Test, fix every error, and confirm the parsed types match your intent, because malformed JSON-LD parses to nothing. Then test the real outcome — ask your target questions in ChatGPT, Perplexity, Gemini, and Google AI Overviews over the following weeks and check whether they retrieve you with the right facts. Trace any wrong fact back to a stale value or a mismatch between markup and page.

Is FAQ schema still worth adding in 2026?

Only where it matches real question-and-answer content, and not for a search feature. Google dropped FAQ rich results for every site on May 7, 2026, completing a phase-out that began in 2023. FAQPage remains valid schema.org and Google says it will keep using the markup to understand pages — and AI systems like the clean Q&A pairs — so keep it for machine parsing, never for a SERP rich result that no longer exists.