TL;DR

Mike King, founder of iPullRank, published a detailed critique of Google's May 15, 2026 AI Search optimization guide, calling it naive and self-serving. His core argument: Google dismisses Generative Engine Optimization (GEO), llms.txt, and passage-level chunking not because they don't work, but because they benefit platforms Google doesn't control. Meanwhile, Microsoft's Bing openly embraces these practices. The real takeaway for brands: stop waiting for Google to tell you what matters. Start measuring who actually cites you across ChatGPT, Perplexity, Claude, and Gemini.

The Debate That Broke SEO Twitter

On May 18, 2026, Mike King dropped a 4,500-word essay titled "Google's Guidance on AI Search is Naive and Self-Serving" on iPullRank.com. Within hours, PPC Land ran full coverage. LinkedIn threads from Krishna Madhavan (Principal PM at Microsoft AI/Bing), Anthony Nichols, and Nikita Vlasyuk (CTO, DeepSEO) piled on.

King's target: the new Google Search Central guide published May 15, which told publishers that AI search optimization is still just SEO, that llms.txt files aren't needed, and that chunking content for passage retrieval doesn't matter.

His counter-argument is sharp. Google has a 15-year pattern of labeling every new surface (mobile, voice, AMP) as "just SEO" while practitioners absorb extra work without extra budget. The skill set for AI search visibility (information retrieval theory, vector distance math, RAG pipeline analysis, brand citation tracking) barely overlaps with the traditional SEO toolkit of keywords, link building, and rank tracking.

Three Arguments That Matter Most for Brands

1. The "Still SEO" Framing Hides a Skills Gap

King documents how Google's framing lets companies underhire for the problem. If AI optimization is "just SEO," nobody gets budget for citation tracking, passage-level content engineering, or MCP agent design. But these skills determine whether ChatGPT or Perplexity surfaces your brand in a synthesized answer. Pretending the skills are identical means pretending the problem doesn't exist.

2. llms.txt: True for Google, False for Everyone Else

Google's guide says llms.txt files aren't necessary. King's response: that's only true for Google, because Google doesn't process them. Anthropic documents llms.txt support for Claude. Other AI systems read them. An honest guide would say "Google doesn't use llms.txt; other systems do; decide based on your multi-platform strategy." Instead, Google conflates "we don't use it" with "you don't need it."

3. Chunking Happens Whether You Optimize for It or Not

King spent 4,500 words in January 2026 explaining why chunking occurs in every RAG pipeline regardless of whether authors optimize for it. He cites Bing documentation stating that chunking and transformations must preserve meaning and claims used in answers. He cites Google's own MUVERA research and passage indexing patents. A passage focused on one idea retrieves better than a passage covering three topics. That's vector distance math, and it doesn't care about Google's public guidance.

Why Bing's Position Changes Everything

While Google downplays GEO, Microsoft takes the opposite approach. Jordi Ribas, CVP of Search and AI at Microsoft, uses "GEO" without scare quotes in official posts. In February 2026, Bing launched an AI Performance dashboard in Webmaster Tools showing citation frequency, page-level activity, grounding query phrases, and temporal trends.

Bing's "Evolving Role of the Index" post states directly that the unit of value shifts from documents to groundable information, meaning discrete, supportable facts with clear provenance. Google's guide and Bing's documentation describe fundamentally different realities.

For brands, this matters because measurement already exists. You don't need to wait for Google to acknowledge multi-platform AI visibility. The data is available today.

What to Actually Do: 5 Steps for an AI Brand Visibility Audit

Instead of debating whether GEO is real, measure it. Here's a practical checklist:

Audit your AI citations now. Check who mentions your brand in ChatGPT, Perplexity, Claude, and Gemini responses for your target queries. Tools like Be Recommended track this across platforms automatically.
Compare AI citations to Google organic. Research from 5W Public Relations found the overlap between Google rankings and AI citations dropped from 70% to roughly 20%. Your Google position increasingly doesn't predict your AI visibility.
Implement llms.txt regardless of Google's opinion. If non-Google AI systems read it (and they do), the cost of a curated llms.txt file is near zero and the upside is real.
Structure content at the passage level. Write each H2 section as a standalone answer block of 40 to 80 words. AI engines extract passages without full-page context. A focused passage on one topic retrieves better than a paragraph covering three.
Track citation changes weekly. AI search results shift faster than traditional rankings. Monthly checks miss the signal. Set up automated monitoring and compare week over week.

The Content Warehouse Leak Precedent

King invokes the 2024 Google Content Warehouse leak as a trust baseline. Over 2,500 modules and 14,000 attributes were documented in an internal engineering wiki, including signals Google had publicly denied using. When guidance and internal practice diverge, trust measurement over PR.

FAQ

Is GEO actually different from SEO?

Yes. While they share some foundations (crawlability, structured data, content quality), GEO requires skills in information retrieval theory, citation tracking across AI platforms, passage-level content engineering, and entity optimization that traditional SEO doesn't cover.

Should I still follow Google's AI optimization guide?

Read it, but don't treat it as the full picture. Google's guide optimizes for Google's ecosystem. If your audience uses ChatGPT, Perplexity, or Claude, you need a multi-platform strategy that Google's guide doesn't address.

Does chunking content for AI search actually help?

Yes. RAG pipelines chunk content regardless of whether you optimize for it. A passage that covers one clear idea retrieves more accurately than one that mixes multiple topics. This is measurable on any public embedding API.

What is llms.txt and should I use it?

llms.txt is a curated list of your most important pages placed at your domain root, formatted for LLM consumption. Google doesn't process it, but other AI systems do. The implementation cost is minimal and the multi-platform benefit is real.

How do I measure my brand's AI search visibility?

Use a cross-platform monitoring tool like Be Recommended to track which AI engines cite your brand, for which queries, and how that changes over time. Compare this data to your Google Search Console organic performance.

Sources

Mike King, "Google's Guidance on AI Search is Naive and Self-Serving," iPullRank, May 18, 2026
Luis Rijo, "Mike King: Google's AI search guide serves platform over the open web," PPC Land, May 18, 2026
Google Search Central, "Optimizing your website for generative AI features," May 15, 2026
Bing Webmaster Tools AI Performance Dashboard, launched February 2026
5W Public Relations, AI citation overlap research, 2026

Mike King Says Google's AI Search Guide Is Platform Strategy, Not Technical Advice. Here's What to Do Instead.