ChatGPT & Perplexity Returned Fake Data

E956: Do large language models actually use schema the way SEOs think they do?

A recent test by Mark Williams-Cook set out to answer that question. He created a fictional company and added completely fabricated schema types to the page - things like "FlockName," "WaddleStyle," and "MigrationPattern." None of it was valid structured data.

When ChatGPT and Perplexity were asked for the company's address, they returned the made-up data from the schema anyway. What does that mean?

It means that LLMs are not interpreting structured data in the way many "generative engine optimization" advocates claim. Instead, they treat schema like any other text in the HTML.

In this episode, I cover: - The full breakdown of the Duck Yeah T-Shirts schema experiment - Why invalid schema was still surfaced by ChatGPT - Whether LLMs are parsing structured data in real time - Why body content appears to matter more than tags or markup - Charles Floate's take on schema and entity reinforcement - How "sameAs" schema actually works in practice - When schema still makes sense for traditional Google search - What John Mueller has publicly said about structured data and rankings - Whether schema can influence AI citations - The risks of trying to "hide" text in schema for LLM visibility

We also discuss: - Consensus schema in competitive SERPs - Entity trust and validation signals - YMYL niches and domain-level trust - Why click-through rate may be the only indirect ranking benefit of structured data

If you care about SEO, AI search visibility, or understanding how LLMs actually process web pages, this episode walks through the logic step by step.

⭐️ Mark Williams-Cook's experiment - https://www.linkedin.com/posts/markseo_seo-share-7424067359088455680-vNFv/ ⭐️ John Mueller's schema comment - https://bsky.app/profile/johnmu.com/post/3lmoenzsfwc2c

💎 Compact Keywords - My SEO Course - Get paying customers through SEO - Clear step-by-step video breakdowns - SEO templates to be copied and adapted for your products and services: https://compactkeywords.com/

👨‍💻 Edward Sturm on Grokipedia - https://grokipedia.com/page/Edward_Sturm

00:00 Do LLMs Use Schema? 00:25 Mark Williams-Cook's "Fake Schema" Experiment 02:23 The Shocking Results 02:47 Schema Treated Like Regular HTML Text 03:12 Charles Floate's Take: What Schema Is Really For 04:47 "SameAs" Schema For Entity Linking & Trust 07:29 Entity Trust, YMYL, and Why SameAs Can Be a Growth Hack 08:14 When Schema Helps (SERP Features & CTR) 09:27 John Mueller: Structured Data Doesn't Boost Rankings 10:11 Final Verdict: Schema Won't Boost LLM Presence (and Hidden Text Risks) 10:58 Wrap-Up

The Edward Show. The #1 search engine optimization podcast: https://edwardsturm.com/the-edward-show/

#searchengineoptimization #answerengineoptimization #generativeengineoptimization #seoschema