---
title: 'A schema markup audit of our own site, before and after a rebuild'
description: >-
  We audited our own site's schema markup before and after a rebuild. Everything
  parsed both times; the move fixed one gap, kept one and added a new one.
author: elroot_labs
date: '2026-11-18'
categories:
  - digital-infrastructure
---
A schema markup audit answers two questions: does the structured data on your pages parse, and does it give Google what it asks for on that page type. The first is easy to pass. The second is where problems hide, and a site rebuild can move them around without anyone noticing. We ran the same audit on seven pages of elrootlabs.com twice: on 16 November 2026, on the old site, and on 17 November, the day after we moved to a new build. Every JSON-LD block parsed both times. The move fixed one gap, kept another and introduced a new one, which we fixed later that day. None of them would show up as an error in a validator.

## Why "rich results" is the search to pay attention to

In a Google Trends check on 16 November, the top rising queries under "schema markup" in India were "rich results", "google rich results test" and "rich results test". People are asking whether their markup produces anything visible in search, which is the right question. Google's [introduction to structured data](https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data) is plain that markup makes a page eligible for an enhanced result and does not guarantee one, and that the Rich Results Test, URL Inspection and the rich result status reports in Search Console are the tools for checking it.

An audit that only runs the validator stops at "no errors". The useful audit goes one step further and asks whether each property says something true and specific.

## How we ran it

We wrote a short Node script that fetches each page, pulls out every `<script type="application/ld+json">` block, parses it, and lists the `@type` of every object inside, including nested ones. For BlogPosting objects it prints `headline`, `image`, `datePublished`, `dateModified` and `author`. It reads the robots meta tag and canonical link while it is there. The pages were the home page, services, work, the blog index, one blog post, about and contact.

A script beats clicking through a validator page by page because you can run it again after every deploy, and because it shows the whole site's markup side by side, which is how repeated values and changed types become obvious. Running it before and after a rebuild is the case it was made for.

## Before and after the rebuild

| Check                                         | Old site, 16 Nov                             | New site, 17 Nov                             |
| --------------------------------------------- | -------------------------------------------- | -------------------------------------------- |
| All JSON-LD parses                            | Yes                                          | Yes                                          |
| Organization and WebSite                      | On every page                                | On the home page only                        |
| Services described                            | Four Service objects                         | Four Service objects, nested in an item list |
| Blog post image                               | The same site-wide share image on every post | A cover per post, 1200 by 630 pixels         |
| Blog post author                              | The Organization, by its @id                 | A Person named "Elroot Labs"                 |
| Dates                                         | Plain dates, no time or timezone             | Unchanged                                    |
| dateModified never earlier than datePublished | Yes, after a fix                             | Yes                                          |

Google's [Article structured data page](https://developers.google.com/search/docs/appearance/structured-data/article) says there are no required properties for articles, only recommended ones: author, author.name, author.url, dateModified, datePublished, headline and image. By the letter of that page, both versions of our post markup are complete. The differences are in what the values say.

## Fixed by the rebuild: every post now has its own image

On the old site, every post's `image` was the site-wide share image, which still showed an old tagline. Google's guidance asks for images that are relevant to the article and represent the marked-up content, and it specifically says to avoid logos. A shared brand image passes a validator and fails that.

The new site draws a cover for each post at build time, and the markup points at it. That closes the main gap. One recommendation is still open: Google suggests high-resolution images in three aspect ratios, 16x9, 4x3 and 1x1, and our covers come in a single 1200 by 630 size. Adding the other two ratios is a small change to the cover generator.

## Kept by the rebuild: dates without a time or timezone

Both versions write plain calendar dates, such as 2026-11-01. Google asks for ISO 8601 date-times and recommends including timezone information, otherwise it uses Googlebot's timezone. For a site in India publishing at midnight, a date read as UTC can land on the previous day. Writing `2026-11-01T00:00:00+05:30` removes the ambiguity, and it is a one-line change in the template. A rebuild is exactly when a fix like this should have gone in, and it did not, because nobody had written it down as a requirement.

## Introduced by the rebuild: the company marked up as a person

The old site set `author` to the Organization by its `@id`. The new site sets it to a Person whose name is "Elroot Labs". Google's guidance asks you to use the Person type for people and the Organization type for organisations, to give `author.name` with only the name, and to point `author.url` or `sameAs` at a page that identifies the author. A company described as a person breaks the first of those rules, and it is the kind of change that happens quietly when a new template ships with a default author type.

Our pages show "By Elroot Labs", so the visible byline is right. The fix is to mark the author as the Organization again, or, if a named engineer starts signing posts, as a real Person with a profile page, with the organisation kept as `publisher`. We made the first change on 17 November, after the audit, and the live posts now name the Organization as author again.

## The bug a validator would never catch

A few days before the first audit we fixed a real error in the old site's markup. Every post's `dateModified` was taken from the time the site's content store was last saved, so posts published on 13 November claimed to have been modified on 30 October, before they existed. The JSON-LD was perfectly valid. The [schema.org BlogPosting type](https://schema.org/BlogPosting) accepts any date in that field. Only reading the values against each other showed it. On the new site the two dates match, so that bug did not come back.

So an audit has to read values against each other as well as against the guidance. These are the cross-checks we run:

- `dateModified` is never earlier than `datePublished`.
- `headline` matches the visible title.
- `image` is unique per post.
- `author` uses the right type for what it names.
- Every URL in the markup returns 200.

## Run the same audit on your site

1. List one URL per template: home, a service or product page, a listing page, an article, contact.
1. Extract every JSON-LD block from each and parse it, including nested objects. A short script or the Rich Results Test, one URL at a time, both work.
1. For each page type, open Google's guidance for that type and list the recommended properties.
1. Compare the values across pages. Repeated images, identical descriptions, wrong types and impossible dates show up only side by side.
1. Check that everything in the markup is visible on the page, as Google's general guidelines require.
1. Save the output, and run the audit again after any redesign, migration or template change, then compare the two runs line by line.

None of this decides whether a page is worth citing. It makes sure the page describes itself accurately. For what Google says AI features need from a page, see [our post on Google AI Mode](/blog/google-ai-mode-website).
