GEO Myths Debunked: What SEOs Should Really Do About llms.txt, Schema and Freshness

Generative Engine Optimization (GEO) guidance has proliferated across the web, but not all recommendations are equally useful. Search Engine Land contributor Philipp Götza warns that “Bad GEO advice … can definitely cost money, cause unemployment, and lead to economic death.” That sharp line captures a real risk: following trendy, unproven tactics can do more harm than good. This post distills Götza’s key points, supplements them with additional data, and lays out practical steps for SEOs and site owners.

Why so much GEO advice feels misleading

Götza’s article argues that misinformation thrives because of cognitive bias, black-and-white thinking, and the compression culture around AI content. He recommends a “ladder of misinference”—moving from statement to fact, data, evidence, and finally proof—before accepting claims. That framework helps separate lightweight opinion from reproducible SEO practice.

Three myths he calls out

Götza focuses on three widely-recommended tactics: building an /llms.txt file, relying on schema markup to influence AI citations, and obsessing over freshness. His conclusions are pragmatic: llms.txt is a proposal without evidence of impact, schema is a hygiene factor that likely helps indirectly, and freshness has the strongest empirical support among the three.

What the llms.txt proposal actually says

The llms.txt proposal (the formal description is available at https://llmstxt.org/) proposes a root-level /llms.txt file that provides concise, LLM-friendly pointers and markdown versions of pages. As the proposal puts it, “We propose adding a /llms.txt markdown file to websites to provide LLM-friendly content.” The spec is a helpful idea for agentic workflows and developer documentation, but it remains a proposal—not a standardized signal used by major AI platforms.

Freshness: the data-backed signal

Of the three topics, freshness has the strongest empirical backing. Ahrefs’ analysis of 17 million AI citations concluded: “Compared to traditional search results, AI assistants prefer citing fresher content.” Their study shows AI-cited URLs are, on average, about 25.7% “fresher” than organic search results, with ChatGPT among the platforms favoring newer sources most strongly. That finding aligns with Götza’s advice to keep content updated where recency matters.

Practical implications for SEOs

Here’s what to do next—actionable steps that follow the evidence while minimizing wasted effort:

Don’t rush to implement llms.txt unless you have a clear agentic use case. Treat it as an experimental enhancement for complex APIs or developer docs, not a broad SEO fix.
Implement schema markup where it supports rich results and user experience. Götza calls schema a “hygiene factor”—it may not directly influence an LLM’s internal knowledge, but schema improves how search engines and downstream tools understand your content.
Prioritize freshness for queries where timeliness matters. Use a targeted update cadence: audit pages that cover news, product availability, or regulatory content and update them with substantive changes and consistent date metadata (on-page, in schema, and lastmod in sitemaps).
Log and measure. Check server logs and crawl activity to see if and how new files (like llms.txt) affect agent crawl volume before committing to broad rollouts.
Seek dissenting views and test. Use the ladder of misinference—seek experiments, reproducible data, and opposing viewpoints before turning tactics into policy.

Analysis: where to invest SEO resources

Not all SEO work returns the same value. Götza’s central warning—against following authority or consensus without evidence—suggests teams should be conservative about dev-heavy projects whose benefits are speculative. Invest first in durable assets: clear information architecture, on-page quality, schema for supported features, and measured freshness where it drives value. Experimental work (for instance, llms.txt deployment) should be limited-scope, A/B tested where possible, and monitored for measurable signals.