Generate Schema Markup with ChatGPT: Prompt, Checks and Pitfalls

Laptop showing large JSON-LD code, with the headline: Schema markup from ChatGPT, draft it, test it Laptop showing large JSON-LD code, with the headline: Schema markup from ChatGPT, draft it, test it

Here is a mistake that costs you. You paste a page into ChatGPT, ask for schema, and get back a neat block with a star rating and a review count your page never showed. Google’s guidelines say not to mark up content readers cannot see. They also say reviews and ratings must come from real users, and a breach can lead to a manual action [2]. That action removes rich result eligibility. Google says it does not affect ranking. This guide shows how to get clean JSON-LD from ChatGPT. It builds on technical SEO fundamentals, and it does not need any code.

If you only do one thing

Give ChatGPT only facts copied from your page, and tell it to leave out anything missing.

Editor’s note, 10 November 2026: we rebuilt this page. We removed statistics we could not trace to a source. Every Google statement below was checked against Google’s own pages.

Yes, for the drafting. It can turn facts you give it into JSON-LD. It cannot see your live page, and it can state wrong values with confidence. So you supply the facts, tell it to drop anything missing, and check the result against the page.

Google recommends JSON-LD where your setup allows it. Microdata and RDFa are also supported. Article markup has no required properties. Google recommends author, dates, headline and image [1] [3]. And markup may not show a rich result even when the Rich Results Test passes [2].

How schema reaches Google: four layers

Schema is built in layersLayer one is the visible page. Layer two is the JSON-LD that repeats it. Layer three is the tests. Layer four is the Search Console reports.aiseojournal.netSchema is built in layers1. What readers seeThe visible page: headline, author, dates, images. This is the source of truth.2. What you tell GoogleJSON-LD that repeats those facts and adds nothing the page does not show.3. What tests checkThe Rich Results Test before launch. URL Inspection on the live URL.4. What Search Console reportsRich result status reports show whether the markup stays valid.Source: Google Search Central structured data documentationaiseojournal.net
Read the workflow as text
  1. 1. What readers see: The visible page: headline, author, dates, images. This is the source of truth.
  2. 2. What you tell Google: JSON-LD that repeats those facts and adds nothing the page does not show.
  3. 3. What tests check: The Rich Results Test before launch. URL Inspection on the live URL.
  4. 4. What Search Console reports: Rich result status reports show whether the markup stays valid.
Built from the steps in this article.aiseojournal.net

Each layer depends on the one above it. If layer one does not show a fact, layer two should not state it.

Professional tip

Use one JSON-LD block per page, and put it on the page it describes. Google says to place markup on the page it describes [2]. Several competing blocks are harder to check.

Build the prompt

Fill in the fields. The prompt writes itself. Copy it into ChatGPT and read the answer line by line.

Prompt builder: BlogPosting JSON-LDaiseojournal.net

Type the facts from your page. The prompt below updates as you type. Nothing leaves your browser.

Copy it exactly.

Name only. No job title.

A page about the author.

ISO 8601 with timezone, for example 2026-03-04T09:00:00+00:00.

Same format.

One or more full URLs, separated by commas.

The full address.

Your prompt

It builds a prompt. It does not call ChatGPT and does not check your facts. aiseojournal.net
  • Dates. Use ISO 8601 with a timezone. Without one, Google assumes the timezone Googlebot uses [3].
  • Author name. Put the name only. Leave out job titles, honorifics and “posted by” [3].
  • Images. Google asks for images of at least 50,000 pixels, ideally in 16×9, 4×3 and 1×1 ratios [3].
  • Type. Google’s Article page supports Article, NewsArticle and BlogPosting [3].

Worked example 1: a clean answer

Hypothetical page, with example.com addresses. Every value is one you supplied.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "How to Repot a Snake Plant",
  "author": {
    "@type": "Person",
    "name": "Jane Doe",
    "url": "https://example.com/about/jane-doe/"
  },
  "datePublished": "2026-03-04T09:00:00+00:00",
  "dateModified": "2026-03-04T09:00:00+00:00",
  "image": [
    "https://example.com/img/repot-16x9.jpg",
    "https://example.com/img/repot-4x3.jpg",
    "https://example.com/img/repot-1x1.jpg"
  ]
}
</script>

Worked example 2: invented ratings, and the fix

Say your page shows no reviews, and the answer contains this block:

"aggregateRating": {
  "@type": "AggregateRating",
  "ratingValue": "4.8",
  "reviewCount": "127"
}

The fix. Delete the whole block and run the test again. Then add the prompt rule “no ratings or reviews I did not give”, so it stays out.

Quick check

Hold your page next to the JSON-LD. Can you point to every value on the visible page? If not, delete that value.

Worked example 3: test it, step by step

  1. Open Google’s Rich Results Test and paste the page URL, or the code.
  2. Fix every error first. Then read each warning and decide whether to act.
  3. Publish. Run URL Inspection in Search Console on the live URL to confirm Google found the markup.
  4. After launch, check the rich result status reports. They show whether templating or serving broke the markup [1].

The rules Google sets

  • Visible content only. Do not mark up information users cannot see [2].
  • Relevant markup. It must represent the page. Google’s example: do not label woodworking steps as a recipe [2].
  • Access. Do not block pages that carry structured data from Googlebot [2].
  • Fresh content. Time-sensitive content that is no longer relevant will not get a rich result [2].

Watch out

Do not ask the model to "add rich result markup" for FAQ or HowTo. The search gallery page we read lists 24 features and does not list either. It gives no deprecation dates, so check the gallery before you plan around a feature [4].

If the test shows errors

  1. Read the first error only. Many later ones come from it.
  2. Check that the JSON parses. A stray comma or comment breaks the whole block.
  3. Compare each value with the page. Change the page or the markup until they agree.
  4. Re-test. Stop when there are no errors and each warning has a reason.

What went wrong on our own page

From our own site

On 9 November 2026 a bulk schema job on this site replaced the whole article on this page with a schema block. The page showed no body text until 10 November. The markup was valid. The page was empty.

How we found it: a scan of all 559 published posts for pages whose only content was a schema block. It found three, this post and two others. Limit: the scan only catches pages that hold nothing but schema.

After you add JSON-LD, open the live page and read it.

Which tools to use, and what they cannot do

  • ChatGPT drafts the JSON. It cannot see your live page unless you paste it.
  • Rich Results Test validates markup for supported features. A pass does not guarantee display [2].
  • URL Inspection and rich result reports show whether Google found the data and whether it stays valid [1].
  • Schema Markup Validator (validator.schema.org) checks syntax against schema.org. We did not open its documentation for this update.

Frequently asked questions

It can draft the JSON-LD from facts you give it. Check each value against the page, because it can state wrong values.

Google recommends JSON-LD where your setup allows it. Microdata and RDFa are also supported.

No. Google’s Article page lists no required properties. It recommends author, dates, headline and image.

Not always. Google says markup may not show even when the Rich Results Test passes.

The search gallery page we read on 10 November 2026 does not list them. Check the gallery before you rely on either.

Test the draft, then read the live page

Give the model facts, not freedom. Then test the markup and open the live page. Next, read how Google finds and renders pages in our technical SEO fundamentals guide.

Sources

3 comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Keep Up to Date with the Most Important News

By pressing the Subscribe button, you confirm that you have read and are agreeing to our Privacy Policy and Terms of Use