A high-impact technical change, for several reasons
Schema.org is the shared vocabulary; JSON-LD is the format
Schema.org is a vocabulary maintained jointly by Google, Microsoft, Yahoo, and Yandex since 2011. It defines machine-readable types — Product, Article, Recipe, FAQPage, HowTo, LocalBusiness, Organization, Event, Course, Review, Person, and hundreds more — and the properties each type carries. JSON-LD (JSON for Linking Data) is the recommended format for embedding that vocabulary in a web page. It lives in a single script tag in the head or body and is invisible to humans. Search engines and AI agents read it and use it to understand exactly what the page is about. There is no SEO penalty for using it; there is a substantial visibility cost for not.
Pick the types that match the page, and use them honestly
A product page should carry a Product type with name, description, brand, image, sku, offers (with price, priceCurrency, availability), and ideally aggregateRating with reviewCount. A blog post or news article should carry an Article or NewsArticle with headline, datePublished, dateModified, and author. A frequently asked questions page benefits from FAQPage with each Question and acceptedAnswer. A how-to guide uses HowTo with step objects. A small business should publish an Organization or LocalBusiness with address, telephone, openingHours, and sameAs links to social profiles. The rule of honesty matters: marking up a non-product as a Product, or fabricating a five-star rating, is a flag that AI systems and search engines penalise heavily.
Every page needs at least the basics
At a minimum, every public page should carry a BreadcrumbList showing where it sits in the site hierarchy, a canonical URL, an Organization block on the home page, and a type-specific schema (Article, Product, FAQPage, etc.) on each content page. Vercel's agent-readability specification adds the requirement that the JSON-LD include dateModified — agents heavily favour pages that prove they have been updated. Google's structured-data testing tool and Schema.org's validator both flag missing required fields; running new pages through one of them before publishing is a fifteen-minute habit that catches almost all errors.
JSON-LD has compounding benefits
A page with good JSON-LD serves three different audiences at once. Classic search engines display rich results (product cards, FAQ accordions, recipe tiles, review stars) that lift click-through rates. AI assistants pull the structured facts directly into their answers, often with a citation. Agents tend to use the JSON-LD as a first port of call because parsing it is faster and more reliable than scraping the rendered HTML. One important caveat. Google's May 2026 generative-AI guidance states explicitly that structured data is not strictly required for visibility in AI Overviews; the underlying language models can read paragraphs. Google still recommends using it as part of the overall SEO strategy, because the rich-results pathway and the AI-extraction pathway both pay back. The same fifteen lines of markup therefore improve traditional SEO, generative-engine extraction, and agent-readable structure off the same effort.
Pick a single page on your site that you most want an AI agent to recommend. Could a stranger reading only its source code answer five basic questions about it — what is sold, who it is for, how much it costs, who has reviewed it, and when it was last updated?