Removing JSON-LD that contradicted visible page content lifted Rich Results Test pass rates and removed manual-action risk in the 9 cases we audited.
Across 9 client sites, we removed inflated AggregateRating, FAQ entries with no visible answer, and Service entries for services not actually offered. Subsequent Search Console manual-action notices that had been pending dropped to zero.
Methodology
Manual schema-content audit; pre/post Search Console manual-action review.
Limits of this finding
Removing risk is not the same as adding lift. Citation rates did not measurably change in the same window.
More Technical findings
Pages whose primary content was JavaScript-rendered (visible only after hydration) were retrieved less often by RAG-based LLMs than equivalent server-rendered pages.
We mirrored 6 SPA pages as prerendered HTML on the same domain. Retrieval inclusion (the LLM citing or summarising the content at all) rose by roughly 2x on the prerendered variants over a 3-week window.
Canadian sites with INP (Interaction to Next Paint) above 200ms see roughly 22% lower SERP click-through compared to sub-200ms peers in our matched-pair sample.
We matched 64 Canadian SMB sites into pairs by industry, query intent, and serp position, then segmented by INP performance. The sub-200ms cohort had a mean CTR of 4.1%; the >200ms cohort 3.2% — a roughly 22% relative gap. INP became Google's third Core Web Vital in March 2024; this is the first full year where its ranking weight has been observably material in our data.
Only 18% of top-50 Canadian agency websites we audited implement Organization, Person, and Article schema together — the configuration we observe most strongly correlates with AI engine citations.
Hand-audited the top 50 ranking Canadian SEO agency websites for schema coverage. 84% had Organization schema. 47% had FAQPage. Only 18% had the trifecta of Organization + Person (author) + Article schema deployed consistently. In our citation-probe data, the trifecta cohort received roughly 2.6× the LLM citations of the partial-coverage cohort.
Run experiments like this on your site.
Senior-led measurement plans, hand-counted, no laundered numbers.
Book a strategy call