Schema Markup Audit Prompt for ChatGPT & Claude

Why one prompt covers three different checks

Schema markup, heading structure and breadcrumb schema are three separate technical signals that never show up in a normal preview. Your page looks fine, so nothing prompts a check. A dedicated SEO audit tool bundles them together with dozens of other checks, usually behind a paid or freemium signup. If you're already building with Lovable, Bolt.new, Base44 or Claude Code, though, you almost certainly have an AI assistant open already, often one that can read your project directly. One well-specified prompt then gets you the same three answers, no new tool, no signup, no waiting for a crawl to finish.

Where your AI can see the page from

The three checks are the same everywhere. What changes is how your AI gets to the page, and that decides which of the checks it can actually do.

On a client-side rendered app, View Source can miss what scripts add. Open DevTools in Chrome, right-click the <html> line at the top of the Elements tab and choose Copy, then Copy outerHTML. That gives you the page after its scripts have run, which is the version to paste.

Platforms that prerender for crawlers (Base44, and older Lovable apps) can show a crawler something different from what your browser shows, so a "no JSON-LD found" result on your browser's version does not prove crawlers see none.

The prompt

Copy this into your AI and keep only the option that matches how it reaches the page:

I want you to audit a page for three structural issues: its schema markup, its heading stack, and its breadcrumb schema.

How you can see the page (keep the one option that applies to me, delete the other two):
- OPTION 1, you can read my project's files. Audit this page: [route or file name]. Read the files that build the page's <head> (index.html and any head or SEO component) and the page's own components. If something is generated by code at runtime, say so, because the live page may differ from the source.
- OPTION 2, I've pasted the page's HTML (after its scripts ran, if it's a JavaScript app): [paste the full HTML here]
- OPTION 3, I've only given you a URL: [paste the URL here]

If the way you're reading the page can't show <script> tags (a fetch tool may return plain text only), don't tell me a check found nothing. Tell me which checks you couldn't do and what you'd need to do them.

Check these three things and report each one separately and don't blend them into one paragraph:

1. Schema markup: find every JSON-LD block (a <script type="application/ld+json"> tag, or the code that writes one). For each one, state its @type, and note anything that looks wrong: no schema block at all, a @type that doesn't match what the page's actual content clearly is, two types that each claim to be the page's main subject (for example an Article and a Product; an FAQPage, HowTo or BreadcrumbList block alongside the main type is normal stacking, not competition), or a required field missing for that type (Google's Rich Results Test, for example, flags a software or web app type that has fewer than two of offers, aggregateRating, applicationCategory and operatingSystem).

2. Heading stack: list every H1 through H6 in document order, exactly as written. Flag it if there's no H1, if there's more than one H1, if a heading level is skipped (an H2 followed directly by an H4, with no H3 anywhere), or if a heading looks like it's only there for visual size rather than genuine document structure.

3. Breadcrumb schema: check for a BreadcrumbList JSON-LD block. If it exists, compare its hierarchy against the page's visible breadcrumb trail, if there is one, and note any place they disagree about the page's actual position in the site. If there's a visible breadcrumb trail but no BreadcrumbList schema, say so.

For each of the three checks, tell me plainly whether it's a real problem or just a minor style choice, and what the fix would actually look like. Don't invent an issue where none exists.

What "good" looks like, check by check

What this can look like in practice

Say a page is a free currency converter tool. This is an illustration of the kind of findings a run can turn up, not a single captured output:

Three separate, specific findings, each one something a quick glance at the rendered page would never surface. After a URL-only run, don't trust a "no JSON-LD found" answer for the schema check: re-run with file access or the pasted page HTML. Cross-check anything it flags as broken against Google's Rich Results Test before treating it as confirmed.

Fixing what the audit flags

For a wrong or missing schema block, including a missing BreadcrumbList, use this site's prompt for generating correct schema markup. For a heading problem, or a visible breadcrumb that's wrong, fix it in your project's code: your builder's chat agent or Claude Code can make the edit from the audit's findings, or you can change the file yourself. Then run the audit again to confirm the finding is gone.

What this prompt doesn't check

This prompt is deliberately narrow. It doesn't measure load speed, Largest Contentful Paint, Cumulative Layout Shift or any other real performance number. For that, use Google's own free PageSpeed Insights directly; an AI reading your page can't replicate what a real browser measurement engine does. It also doesn't check for broken links, content quality or whether you're targeting the right keywords. It covers three specific structural signals, nothing more, and it's meant to sit alongside a real performance check, not replace one.