AI Models Ignore SEO Structured Data, Experts Debunk Vendor Claims

Image credit: Search Engine Journal

Global experts are refuting widespread vendor claims that traditional search engine optimization techniques, including structured data and specific content formatting, enhance how artificial intelligence language models process and connect information.

The consensus among AI specialists indicates that LLMs, such as Google’s Gemini, OpenAI’s ChatGPT, and Perplexity, parse text as sequences of tokens, rendering schema and other structured data largely irrelevant to their core comprehension mechanisms.

Companies including Semrush, AirOps, Peec AI, and Profound have been promoting these classical search engine optimization tactics, conflating their proven utility for traditional search engines with supposed benefits for AI engines, industry observers said.

Pedro Dias, a former Google employee, emphasized that Schema.org is valuable for generating rich search results, populating knowledge graphs, and assisting voice assistants but does not contribute to an LLM’s intrinsic understanding of text.

AI language models are inherently designed to process the unstructured public web, making publisher-side structured data unnecessary for fundamental parsing, analysts explained.

Aleyda Solis, an international SEO consultant, noted that while content chunking can be beneficial for AI retrieval, this process is controlled by the AI engine’s internal configurations, not by publishers’ efforts to pre-optimize content in specific ways.

The core architecture of LLMs relies on statistical relationships between tokens to derive meaning, a process distinct from how classical search engines interpret semantic markup.

Vendors’ assertions that headings, subheadings, and other content structuring elements directly improve an LLM’s ability to parse content overlook the fundamental operational differences between these technologies, experts said.

Source: Search Engine Journal