How to Add Schema Markup That Passes Google's Test the First Time

Why does schema markup keep failing Google's test?

Schema markup fails Google's test mostly because it was copied from somewhere old, not written for the page it sits on. A missing comma breaks the whole block. A property from 2019 is no longer supported. A rating that is not on the page breaks Google's rules.

Here is where the time goes:

• Broken code. JSON-LD is strict. One extra comma at the end of a list and the block stops working.
• Old advice. Google keeps changing which types earn a special look in results. It cut HowTo and most FAQ results in 2023, removed the sitelinks search box in 2024, and retired FAQ rich results for every site in 2026. Its Search Central changelog dates that last change to May 7, 2026.
• Markup that does not match the page. Google's structured data policies say not to mark up content readers cannot see. A price in the code that differs from the price on the page is a problem.
• Starting from zero every time. Each new site gets the same Organization block typed out again.

I learned this on my own sites. On 2026-09-11 Search Console flagged "offers, review or aggregateRating must be specified" on 227 product families across 33 catalogue categories of isiklisusleme.com. The cause was Product markup on items that had no price on the page. Switching those list items to the plain Thing type cleared it. An earlier fix on 2026-08-09 had been worse: it pasted the company's 5 Google reviews onto the products as a rating. I rolled that back, because it counts as misleading markup.

The other trap is invented facts. On 2026-09-18 I found that 8 of the 9 Wikidata IDs in the schema of agentskillpacks.com were wrong. An AI had written them. The Stripe ID pointed at a human-rights group, the Gumroad ID at a plant, and 3 IDs did not exist at all. Since then no ID goes live until I have looked it up on Wikidata myself.

How to add schema markup by hand, step by step

Adding schema markup by hand takes seven steps: pick the type, write the block, fill in real facts, place it in the page, test it twice, publish, then watch Search Console. Do this once by hand, even if you later use templates, so you can read and fix any block you meet.

1. Pick the right type. A type is the kind of thing the page is about. Use Article for a blog post, Product for something you sell, Organization for your company and LocalBusiness for a shop with an address. The full list lives at schema.org, but most sites need only four or five types.

2. Write the block in JSON-LD. JSON-LD is a way of writing data as name and value pairs inside curly brackets. Google recommends JSON-LD over the older formats because it sits in one place and does not mix with your page design. A small Organization block looks like this:

   <script type="application/ld+json">
     {
       "@context": "https://schema.org",
       "@type": "Organization",
       "name": "Example Studio",
       "url": "https://example.com",
       "logo": "https://example.com/logo.png",
       "sameAs": ["https://www.linkedin.com/company/example-studio"]
     }
   </script>
   

3. Use only real, visible facts. Every value in the block should also be on the page. Do not add ratings you do not have. Google says a business that hosts reviews about itself cannot get star snippets on its Organization or LocalBusiness markup.

4. Place it in the page. Paste the script block into the head of the page, or into the custom code field your CMS offers. One block per type is enough. Keep it on the page it describes, not site-wide, unless it is your Organization block.

5. Test it in the Rich Results Test. Paste your URL into Google's Rich Results Test. It shows which Google features the page can get and lists errors and warnings. Fix every error. Warnings are optional fields you may skip.

6. Test it in the Schema Markup Validator. The Schema Markup Validator checks your code against all of schema.org, not only the parts Google uses. It catches types that Google ignores but other tools, including AI crawlers, may still read.

7. Publish and watch Search Console. After Google crawls the page again, the Enhancements section of Search Console shows valid items and errors per type. Check it a week or two after launch. If errors show up, test that URL again to find the exact line.

A word on AI search. Google says there is no special schema needed to appear in AI Overviews. Schema still helps machines know who you are. I treat it as a clear label, not a ranking trick. For the AI side, our free AI crawler checker confirms bots can reach your pages at all, and GEO for small shops covers what else moves the needle.

No access to the page head? Other places the block can go

When the block is added by script, test the live URL in the Rich Results Test, not pasted code.

The faster way: start from tested JSON-LD templates

The faster way is to start every block from a template that already has the right fields in place, then change only the values. Structured Data Pro Pack is that set of six templates. A separate field guide lists the required and recommended fields for each type, so you know what you can delete.

An older description of this pack promised FAQ rich results. That promise stopped being true in May 2026, and this page says so plainly.

How to use it, step by step

1. Buy and download. Polar or Gumroad sends the files right after payment.
2. Open the template for your page type, for example Product.
3. Replace the sample values with your real name, price, links and dates. Delete any field you cannot fill with a true value. The Product template has rating and review fields; remove them unless real reviews are shown on that page.
4. Paste the block into your page head or your CMS custom code field. The pack has no CMS-specific notes, so use your platform's own docs for where that field lives.
5. Run the test workflow. Test the live URL in the Rich Results Test and the Schema Markup Validator, in that order.
6. Use the skills if you work in Claude Code. Run /schema-markup with a page address to audit the page and draft markup for it. Use geo-schema for entity markup aimed at AI search, with sameAs links to your real profiles. Run schema_validator.py on a saved page to pull out each JSON-LD block and score it from 0 to 100 on required and recommended fields.

What the output looks like

The output is a single script block per page type, filled with your own facts. The templates are plain JSON, which cannot hold comments. Each one marks the values you replace in capitals, such as PRODUCT_NAME, and a separate field guide lists the required and optional fields for each type. The README describes these as bracketed placeholders, but in the files they are capitals.

One more honest note: the field guide is out of date in three rows. It still lists HowTo steps, FAQ dropdowns and the sitelinks search box as rich results you can earn. Google retired all three, as the links above show. The WebSite template also carries a SearchAction for that retired search box, so treat it as optional.

I do not have a filled pack template with a saved test result to show you yet. What I can show is a real schema check from my own audit setup, run on a lighting-decoration site I worked on, on 2026-09-25 (English original, trimmed):

Schema & Structured Data: 61/100 (Fair) … The live HTML of 8 pages was fetched as GPTBot … No JSON parse errors. … HowTo was dropped by Google in 2023. LocalBusiness is a second entity with no link to #organization.

And one real Rich Results Test result: the home page of textwordcount.com, another of my sites, had 4 ItemLists and got "Multiple ListItem specified on the page". One fix took the page from 13 JSON-LD blocks to 9, with no ItemList left.

How much time it saves

I have not timed a Product block from scratch against one from the template, so I will not give you a number. The one timing I do have is the schema check above: the agent fetched and checked 8 live pages in 3 minutes 46 seconds. That agent is part of my own audit setup, not this pack.

The bigger win is on the tenth site. You stop rebuilding Organization blocks from memory and start from code you already trust.