Use schema to describe an existing article, product page, or organization in machine-readable form. It needs actual content and facts. The skill can organize JSON-LD, but it cannot manufacture reviews, prices, or eligibility for a search feature Google does not support.
Install and invoke
Run this in your project terminal to install only this skill for Claude Code:
npx skills add coreyhaines31/marketingskills --skill schema -a claude-code
Node.js and npx must be available. For Codex, replace -a claude-code with -a codex. In the agent conversation, explicitly ask it to use schema and supply your material. The installation command belongs in the terminal, not in your content brief.
Supply the page before choosing a type
Give the agent the URL, visible copy, dates, author information, and existing JSON-LD. Establish what the page describes before selecting markup. A company homepage and a specific product page should not receive identical properties just because they belong to the same business.
Example brief:
Use schema to inspect article.html and existing-schema.json.
This tutorial has a real author and update date. List supported entities and fields before generating JSON-LD.
Flag missing information; do not invent ratings, credentials, or dates.
Distinguish valid Schema.org syntax from current Google rich-result requirements and provide validation steps.
The skill’s example types are starting points, not a current eligibility list for every Google enhancement. Consult the documentation for the intended feature, including required properties and restrictions. Remove unsupported ratings, offers, and stock information rather than deploying sample values.
Keep code consistent with visible facts
After adding JSON-LD to a template, compare it with the title, author, price, and other facts readers can see. Linked entities should have stable identifiers. The same organization should not accidentally become unrelated objects across pages. Inspect existing plugins or CMS output for duplication, conflicts, and outdated addresses.
Use a Schema.org validator for vocabulary and syntax checks, then Rich Results Test for supported Google features. These tools answer different questions: correct vocabulary does not establish support for an enhanced search display. Inspect the deployed rendered page as well; a correct development file does not prove the live template emits it.
Google explicitly states that compliance and successful testing do not guarantee rich results. Record the checks and date, then sample the affected templates after release. When product or article facts change, update the markup too. Do not add information absent from the reader-facing page merely to pursue a more prominent search appearance.
Generate the smallest markup from visible page facts
Give the skill a real page and its content, not just “add Schema.” For an article, supply title, author, publication date, body URL, and breadcrumbs. If a product page has no public price or availability, do not invent those values to fill Product fields. Ask first for a mapping of visible fact to proposed property, plus missing fields; only then generate JSON-LD. A page may use several relevant types, but names, URLs, and entity identity must stay consistent.
Keep markup synchronized with the rendered page. A static template may emit it directly; on a dynamic site, confirm that the final page really includes the intended JSON-LD. Checking a source variable is not enough, and comments or placeholders are not published structured data. After a migration or pricing change, recheck for old URLs and stale product details.
Validate syntax with a structured-data tool and compare the result with the search platform’s eligibility guidance for the relevant feature. A passing test shows that the syntax and some requirements are met; it does not guarantee a rich result. The most useful manual check is whether a visitor can find the same facts on the page that the JSON-LD claims.
Example: repair stale author data on an article
Suppose an article visibly has a new author, but a shared JSON-LD template still emits the former author. Provide the current rendered page, template fragment, and editorial record. Ask: “Use schema to compare the visible article with its Article markup. List mismatched properties and show a template correction. Change only confirmed author, date, and URL data; do not add ratings, FAQ, or entities absent from the page.”
Inspect the rendered page afterward. Do the visible byline, author name, and author URL in structured data agree? Check other articles using the same template so the fix is not limited to one JSON instance. A syntax validator can pass while the author is still wrong; format validity does not correct a factual mismatch. Retest representative pages after publication. Whether a search platform displays a rich treatment is outside this fix’s completion criteria.
One integration pattern for a multi-page site
A public SEO pipeline loads the predecessor schema-markup alongside site-structure, copywriting and AI-search Skill text as non-interactive system-prompt guidance, then drafts JSON-LD for each route. It shows one way to coordinate markup across a multi-page site, not proof that rich results appeared. Review each draft against visible page facts and current Google eligibility guidance before publication.