How can I measure whether schema markup changes affected AI search citations?

Write down the exact query strings, the AI surface, the country or city context, the device and account context, the treatment URLs, any comparison URLs, the schema type, the start date, and the planned end date. Keep Google AI Overviews, Google AI Mode, and any non-Google surface in separate columns. Do not combine them into one citation percentage.

For every treatment URL and comparison URL, record:

• the answer returned for each fixed query and surface, whether the URL was cited, the exact cited URL and source title, and the capture date and time;
• the rendered facts visible on the page and the exact JSON-LD or other structured-data block;
• the schema type, required properties, relevant recommended properties, and the complete before-state representation;
• the Rich Results Test and URL Inspection results, with their dates; and
• the Generative AI performance report period and page-level impressions, with country, date, and device dimensions where they help explain the comparison.

Google recommends a before-and-after test on a few pages with several months of Search Console data, and it recommends validating the markup and using URL Inspection to check that Google found it.² That is testing guidance, not a prescribed sample size for AI citations.

The existing LocalBusiness schema answer covers the property boundaries and implementation details. This page uses those checks only to establish what changed and whether Google could find the resulting page.

Record the change timestamp, page URL, changed properties, rendered markup, and validation output. Keep content, internal links, templates, Google Business Profile facts, robots.txt, and indexing changes out of the test window when practical. If another change is unavoidable, record it as a confounder instead of assigning its effect to schema.

Structured data should accurately represent visible, current page content. Google says correctly implemented structured data does not guarantee that it will appear in search results, even when the Rich Results Test accepts it.³ A passing validation result is therefore a technical check, not evidence of an AI citation.

Run the Rich Results Test and URL Inspection again. Check that the markup is still visible-page content, accurate, current, and complete for the properties you use. Google says a newly published page may take several days to be found and crawled, so a successful test immediately after deployment is not the end of the observation.⁶

Google's Generative AI performance report defines an impression as a link to a site shown to a user in a generative-AI feature. The report supports page, date, country, and device dimensions.⁴ A manual citation observation is narrower: it records what one specified answer returned for one query, surface, location context, device, account context, and date.

Export or record the same page and dimensions for the pre-change and post-change periods. A zero in either log means that the selected test observed no event; it does not prove that the page was never retrieved outside the test.

Choose pages that are similar enough to make the comparison useful, record why each page was selected, and keep the query panel, surfaces, context, dates, and logging method the same. If no credible comparison pages exist, report the result as a pre-and-post observation without control evidence.

Once the planned window is complete, stop claiming an effect from a test that shows no differential change. Preserve the null observation, its conditions, and its limits so the next test can be compared with it.