How-to

🧩 Mark up your site so AI can read it, not just crawl it

Schema won't guarantee an AI citation, but it removes doubt about what your page is. Generate it with one prompt, from real page facts only.

🎯 Why it matters

AI engines don't need a special AI schema. Google's 2026 guidance says AI Overviews run on the same index and signals as Search. But structured data still tells an AI system, unambiguously, what it's reading instead of inferring it from prose, and that matters most for content the engine hasn't seen before. (This note is about AI engines parsing your entities; the classic-SEO schema-structured-data note is about Google rich results.)

💬 My take

Mark up your site so AI can read it, not just crawl it

Google's May 2026 guidance says no special markup is needed for AI Overviews. At the same time, independent studies find pages with clean structured data get cited measurably more, especially the first time an engine sees them.

Both are true: schema is not a ranking trick, but it's how you remove ambiguity about what your page is. AI can write the JSON-LD for you in seconds. Its one danger is inventing facts, so it must pull every field from what's actually on your page.

Fastest path: one prompt, end to end

Paste your real page details into this prompt in ChatGPT or Claude:

🤖 AI prompt — paste into ChatGPT / Claude

You are a structured-data engineer writing JSON-LD for AI-search legibility.
Page type: [Organization / Product / FAQ / Article]
Here are the REAL facts visible on the page:
[paste: brand name, product name, price, description, any FAQ Q&As, review count/rating if shown]
My verified profile URLs (for sameAs): [Wikipedia / Wikidata / LinkedIn / Crunchbase URLs I actually have]

Do this:
1. Generate valid JSON-LD (schema.org) for the page type above, in a single <script type="application/ld+json"> block.
2. Every field must match the text I pasted WORD FOR WORD. Do not invent a price, rating, review count, GTIN, or sameAs URL. If I didn't give you a value, leave the property out and list what's missing at the end.
3. For Organization, include a sameAs array using ONLY the URLs I provided.
4. Tell me exactly where to paste it and how to validate it (validator.schema.org and Google's Rich Results Test).

Output: the JSON-LD block, then a short "missing / please confirm" list.

Mismatched schema (markup claiming something the page doesn't say) gets a page ignored or flagged, not boosted. That's why the prompt forbids invented values and asks for a confirm list instead.

Or do it in 5 steps

Start with Organization + sameAs. Before an AI trusts a page it must trust who's speaking. Organization schema with a sameAs array (Wikipedia, Wikidata, LinkedIn, Crunchbase) feeds Google's Knowledge Graph and confirms you're one real, verifiable entity. Highest-leverage schema on the site, and most brands skip it.
Layer Product / FAQPage / HowTo / Article on your highest-intent pages. FAQPage and BreadcrumbList are among the best-performing types for AI retrieval.
Match the visible text word for word. Every field must appear on the page. No orphan markup.
Use JSON-LD, not microdata or RDFa. It's Google's recommended format, sits in one script tag, and drifts out of sync less.
Validate and re-validate. Paste the URL into validator.schema.org and Google's Rich Results Test. Re-check after every CMS or theme update, silent breakage is common.

Why AI-first: hand-writing JSON-LD is fiddly and error-prone; the model nails the syntax instantly. The catch is it will happily invent a rating or GTIN, so pin it to your real, visible facts and make it list what's missing.

What schema won't do

Not a ranking factor. A controlled test across ~2,000 pages found near-zero citation lift from adding JSON-LD to pages an engine already knew well. The benefit shows up on content the engine hasn't seen yet.
Results are platform-specific. One 2,000+ URL rollout hugely lifted Google AI Overview and AI Mode citations but produced no measurable change on ChatGPT, Gemini, Copilot, or Perplexity.
It's table stakes, not a growth lever. Schema earns a clearer seat at the table; pair it with distinctive content and off-site authority.

Re-validate after every CMS or theme change.

⚖️ Do & Don't

Do

Add Organization schema with a complete `sameAs` array (Wikipedia, Wikidata, LinkedIn, Crunchbase) before anything else — it's the fastest entity-clarity win on the site.
Layer Product, FAQPage, HowTo, or Article schema onto your highest-intent pages, matching the visible text word for word.
Use JSON-LD rather than microdata or RDFa — it's the format Google recommends and the easiest to validate and maintain.
Re-validate schema after every CMS or theme update — silent breakage from template changes is common and easy to miss.

Avoid

Don't mark up content that isn't visible on the page in the same words — mismatched schema gets pages ignored or penalized, not boosted.
Don't expect schema alone to lift citations on pages an AI engine already knows well — controlled tests found close to zero uplift there; it helps most on content the engine hasn't seen yet.
Don't add schema describing your whole business on every page regardless of topic — 2026 eligibility rules tightened to schema matching a page's primary content only.
Don't treat one platform's results as universal — a rollout that dramatically lifted Google AI Overview citations produced no measurable change on ChatGPT, Gemini, or Perplexity in the same test.

💡 Quick tips

If you can only do one page, do your homepage or About page's Organization schema with a full `sameAs` list — it's what AI systems use to confirm who you are before trusting anything else you say.
Free spot-check: paste any page URL into Google's Rich Results Test — it flags exactly which fields are missing or mismatched.

🏢 Brand in focus

BrandGoogle traffic share fell from ~60% to ~24-33% despite full Recipe schema

Allrecipes — deep schema, but it didn't stop the zero-click hit

Dotdash Meredith · Why it fits: Allrecipes is one of the most schema-marked-up sites on the internet — Recipe, nutrition, rating, and video schema on nearly every page, a textbook implementation. Did well: that structured data made its content easy for search engines, and now AI answer engines, to parse and extract, exactly what thorough schema is supposed to do. Watch-out: exhaustive schema didn't stop AI Overviews from doing the extracting and answering directly — parent company Dotdash Meredith's CEO said AI Overviews appear on roughly a third of its searches and acknowledged the traffic hit, and the real offset came from AI licensing deals with OpenAI and Microsoft and its own ad tech, not from schema itself. Structured data earns you a seat at the table; it doesn't guarantee anyone still visits your site.

📚 Go deeper — resources

article

AI features and your website

article

Does adding schema markup actually increase AI citations?

article

Schema markup: the real impact on AI search

🏷️ Tags

geoai-searchschema

🔗 Related