Semrush’s Technical GEO Claims Schema Won’t Help AI Engines – A Critical Look
Focus on high‑quality, evidence‑rich content rather than over‑optimising schema for AI engines, as LLMs do not parse schema tags.
Implement schema where it delivers real search benefits, but prioritize clear, evidence‑based writing to improve AI visibility.
Summary
Semrush’s latest infographic claims that schema, structured data, and clean architecture—collectively called Technical GEO—ensure AI engines can parse and connect content. The article argues that this claim is a category error because large language models read text token by token and do not parse <schema> tags; the retrieval layer that feeds the model is controlled by the engine, not the publisher.
AirOps’ graphic cites a 13 % citation lift from schema markup, but the methodology is opaque and the numbers are not reproducible. A 2024 KDD paper on GEO tested nine optimisation methods and found that adding credible citations, quotations, statistics, and improving fluency raised visibility, while keyword stuffing and schema had no measurable effect. The piece explains that chunking advice—short, self‑contained paragraphs—actually reflects good writing practices rather than a new technical layer. It also notes that LLMs’ chunking configurations vary across systems and are tuned behind the scenes, making publisher‑controlled optimisation ineffective.
The conclusion is that schema continues to serve classical search functions like rich results and entity disambiguation, but it does not influence how LLMs understand prose. Therefore, technical SEO teams should focus on high‑quality, evidence‑rich content and only implement schema where it delivers real search benefits.
Key changes
- Semrush’s Technical GEO claims schema ensures AI engines can parse content, but LLMs read text tokens, not schema tags.
- AirOps reports a 13 % citation lift from schema, but methodology lacks reproducibility.
- A 2024 KDD paper found that credible citations, quotations, statistics, and fluency improve visibility, while keyword stuffing and schema do not.
- Chunking advice (short paragraphs) reflects good writing, not a new technical optimisation layer.
- LLM retrieval uses internal chunking configurations that publishers cannot control.
- Schema still benefits classical search features like rich results and entity disambiguation.
- Publishers should prioritize content quality over schema for AI visibility.
- Technical SEO teams should implement schema only where it delivers real search benefits.