Google published updated AI search optimization guidance, and iPullRank founder Mike King’s verdict, published May 22 in Search Engine Land, is direct: the document tells practitioners what benefits Google’s ecosystem, not what the underlying technology actually requires.
King points to the 2024 Content Warehouse leak as the baseline for how seriously to take Google’s public guidance. That leak showed internal engineering documentation naming signals Google publicly denied were in use. The gap between internal practice and external communication has a documented prior, and King argues that context belongs in every reading of every Search Central document Google releases.
The core dispute is over three claims in Google’s guide. First, that AEO (answer engine optimization) and GEO (generative engine optimization) are still just SEO. Second, that chunking content is unnecessary because Google’s systems understand multi-topic pages. Third, that rewriting for AI systems is not required because those systems handle synonyms and semantic variation automatically.
King rejects all three. On the GEO=SEO framing, his argument is organizational. SEO carries a budget line and an org-chart position historically placed downstream of product and content decisions. When AI search work is filed under “SEO,” it inherits those constraints. When it gets a distinct label, it gets distinct headcount, cross-functional access, and executive sponsorship. Calling GEO “just SEO” keeps the work underfunded.
On chunking, King’s case rests on how retrieval-augmented generation systems actually process content. RAG pipelines chunk source documents into passages regardless of publisher intent. The practical question is whether those passages survive chunking with coherent meaning or fragment into disconnected text. King notes that Bing’s own published guidance acknowledges this directly, stating that “chunking/transformations must preserve meaning and claims used in the answer.” Google’s public guidance says the opposite, while Google’s own MUVERA research and passage-indexing patents point toward passage-level retrieval being the operative mechanism.
The Bing contrast is the sharpest analytical move in King’s piece. Microsoft’s Bing Webmaster Tools team has published posts explaining how grounding works, announced page-level citation data in their AI Performance tools, and acknowledged GEO as a distinct discipline. Bing’s documentation names the shift from document ranking to “groundable information, discrete, supportable facts with clear provenance.” Google’s documentation says the shift hasn’t changed what you need to do. Reading both in sequence makes it hard to believe they describe the same technology.
On writing for AI retrieval, King’s argument is technical. A retrieval system selects passages by computing vector similarity against a query embedding, then runs pairwise comparisons for synthesis. Specificity, entity salience, and semantic coherence affect those similarity scores in measurable ways. Loose, multi-topic prose puts passages at a disadvantage relative to tight, focused passages. King says empirical testing through public APIs confirms the effect. Google’s guidance, telling practitioners the systems handle semantic variation automatically, is accurate in a narrow sense and misleading in practice: it describes what the systems can do, not what they prefer.
The llms.txt section makes a point worth noting for teams working across platforms. Google’s guidance says publishers do not need to create machine-readable files for AI systems. King’s reading: true for Google, silent on every other system. Anthropic has documented support for llms.txt. Dismissing the file because one platform ignores it is the single-platform reasoning the rest of the piece is criticizing.
King is careful to say SEO fundamentals remain valid: technical structure, crawlability, page experience, and original content all continue to matter. His claim is not that SEO is dead. His claim is that the skill set required to operate across ChatGPT, Perplexity, Bing Copilot, and Google simultaneously has grown well beyond what any Search Central document describes, and that Google’s guidance does not acknowledge that growth because doing so would not serve Google’s interests.
For search teams fielding executive questions about LLM visibility, the gap King identifies is the actionable one. Showing up in ChatGPT, Perplexity, or Copilot depends heavily on Wikipedia presence, third-party publication coverage, licensed data partnerships, and brand citation across the open web. Those are PR, brand, and data-architecture problems. An SEO budget rarely funds them. Treating AI search as a distinct practice, with its own budget and its own measurement, is the organizational move that creates space for the actual work.
Originally published by Search Engine Land on 2026-05-22, based on a column by Mike King of iPullRank, republished with permission.
