Structured Data for AI Search

Give every key page JSON-LD that says exactly what the page is and nothing the page does not show, then prove it with a sourced check.

When to use

The user is adding schema to a site, fixing markup a validator complained about, deciding which types a SaaS, AI, EdTech or MedTech site needs, or asking whether schema will help them appear in AI answers. For a whole-page SEO/GEO/AEO score, use seo-geo-aeo-audit; for rewriting content to be cited, use geo-content-optimization. This skill is the markup itself.

Before you start

1. Read the brand/product context first. Pull the product name, pricing model, and site map from .agents/product-marketing.md. If none exists, ask for the key URLs.
2. Gather inputs: for each page to check, its HTML (best: JSON-LD and visible text in one) or its JSON-LD plus the page text.
3. Check what crawlers receive. If the site renders client-side, JSON-LD added by JavaScript may not be in the HTML a crawler fetches. Run the fetch test in seo-geo-aeo-audit first; markup nobody receives fixes nothing.
4. Set expectations honestly. Google says there is "no special schema.org structured data" needed for AI Overviews or AI Mode, and lists structured data matching the visible text among the fundamentals (Google, AI features and your website). Treat schema as an accurate description of the page, not a lever that wins AI citations. Do not promise an effect on any AI engine.

Method

Diagnose what each page is, then design one primary type per page plus the shared types (Organization once, BreadcrumbList on inner pages), and execute by checking every block against two sources: Google Search Central's documentation for the properties Google requires and recommends, and schema.org for the types Google no longer documents. The rule that overrides all others is Google's general guideline: structured data "must be a true representation of the page content" and you should not "mark up content that is not visible to readers of the page". A rating, price, question or step in the markup that the page does not show is the most common and most damaging error, so the check compares markup against the visible text whenever it is supplied.