GEO advice gets a fact-check framework: separating data from hype

Search Engine Land published “GEO myths: This article may contain lies” on Jan. 19, 2026, presenting a 14‑minute read that fact-checks common AI-search optimization claims.

It frames GEO advice through Alex Edmans’ “May Contain Lies”, using a five-step “ladder of misinference” to separate what’s merely stated from what’s actually supported: statement → fact → data → evidence → proof.

Key points the article highlights:

• How to grade GEO claims: apply the ladder (statement → fact → data → evidence → proof) before treating advice as true.
• User signals as an example: it connects “user signals” to organic performance and layers support from experiments, the 2024 Google leak’s notes on evaluation, and DOJ trial court records as stronger forms of backing.
• Where the idea came from: it credits early advocacy of user signals to Rand Fishkin and Marcus Tandler, then warns (as of Jan. 2026) how tactics spread faster than understanding.
• Practical guidance: it says you don’t need an llms.txt file right now, but still recommends schema markup even if chatbots aren’t using it consistently today.
• Freshness myth: it flags content freshness as important only for queries where recency matters, and treats “fresh updates” as one of the myths under review.
• Summarization reliability: it points to the Hugging Face Vectara summarization leaderboard and a Hugging Face PHARE analysis reporting higher hallucination rates for brief-summary prompts.