Mike King, founder and CEO of iPullRank, today published a detailed critique of Google's recently updated guidance on optimizing for generative AI search, arguing the document reflects the company's platform interests rather than an honest account of how AI retrieval systems actually work.

The analysis, published on May 18, 2026, at ipullrank.com, arrives three days after Google released its updated "Optimizing your website for generative AI features on Google Search" guide through Google Search Central, last updated May 15, 2026. PPC Land covered the guide's publication on May 16, noting it was the first consolidated resource Google had produced specifically addressing how content surfaces inside AI Overviews and AI Mode.

King frames his objections around a recurring pattern. Every time Google ships a new Search Central document, practitioners split into two camps - one citing the text as confirmation that nothing has changed, the other citing it as proof of deliberate misdirection. His argument is that neither camp is asking the more useful question: what does Google gain by framing AI optimization the way it does?

The "it's just SEO" argument

Google's guide addresses the proliferation of new acronyms directly. According to the guide, "AEO" stands for "answer engine optimization" and "GEO" for "generative engine optimization" - terms the document describes as covering work focused on improving visibility in AI search experiences. The guide then states that "from Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

King pushes back against that framing with force. His objection is not semantic. SEO, he argues, is not merely a list of tactics but a set of organizational expectations, a budget line, and a reporting structure. When AI search work gets folded back into "SEO," it inherits the same underfunded, downstream-cleanup positioning that search practitioners have struggled against for years.

The pattern, King writes, has repeated itself for more than fifteen years. Mobile was declared "just SEO." Voice was "just SEO." AMP was "just SEO" - until Google quietly deprecated it after years of implementation work from publishers. Each new surface absorbed additional practitioner labor without proportional increases in headcount or compensation. Calling AI Search "still SEO" repeats that cycle.

There is also a structural argument. The skill set required for AI search work has grown well beyond traditional SEO. The conventional toolkit covers keyword research, technical auditing, internal linking, structured data, content optimization, link building, and rank tracking. The AI search layer adds information retrieval theory, vector distance measurement, RAG pipeline analysis, content engineering at the passage level, agent and protocol design including MCP, A2A, UCP, and ACP, brand citation tracking across LLM platforms, and synthesis evaluation. The overlap between those two sets is real but partial. Pretending the skills are identical, King argues, is how organizations underhire for the actual problem.

The audience has changed too. Traditional SEO optimizes for one machine and the humans clicking its results. AI search optimizes for a retrieval system, a synthesis pipeline, possibly an agentic browser, and a human reading an answer that may not link to a source website at all. Those are different consumers, with different measurement and different reporting requirements.

What the leaked documents showed

King grounds part of his skepticism in a specific historical event. In 2024, thousands of pages of Google's internal Search ranking documentation became public - the so-called Content Warehouse documents. PPC Land reported on the leak at the time, noting the documents outlined over 2,500 modules and 14,000 attributes related to how Google Search processes and stores information from web documents. The documents showed signals that Google had publicly stated did not exist, named and weighted inside its own engineering wiki. That is the baseline from which King evaluates every subsequent piece of public guidance from the company.

He is not claiming the new guide is entirely false. He is claiming that Google has a well-documented history of nudging the industry toward behaviors that serve the platform - formatting structured data, cleaning up technical debt, waiting for algorithm updates - rather than toward building expertise that operates across platforms Google does not control.

What Microsoft has been doing differently

King contrasts Google's posture with a series of publications from Microsoft's Bing team. The contrast is detailed and uses specific documents. Jordi Ribas, Corporate Vice President of Search and AI at Microsoft, published a post called "Elevating the Role of Grounding on the AI Web" that openly names what is changing: agents are doing the browsing, they are drawn to structured and verifiable content, and a new optimization discipline called Generative Engine Optimization is emerging in response. No dismissive air quotes.

A second post, "Introducing AI Performance in Bing Webmaster Tools," described what Microsoft itself called an early step toward GEO tooling. PPC Land covered that announcement in detail when it launched February 10, 2026. The dashboard gave publishers, for the first time, visibility into how often AI systems cited their content across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. It exposed four primary metrics: citation frequency, page-level activity, grounding query phrases, and temporal trends. Publishers had operated without any of that data. PPC Land also covered Microsoft's subsequent explanation of how grounding technology powers nearly every major AI assistant in the market, published February 12, 2026.

A third Bing post, "Evolving role of the index: From ranking pages to supporting answers," states according to King that "the unit of value shifts from documents to groundable information - discrete, supportable facts with clear provenance." It acknowledges that the metrics, the unit of analysis, and the responsibility of the system have all changed. King's point is blunt: read those three Bing posts, then read Google's mythbusting section, and the two documents do not appear to be describing the same technology.

llms.txt and multi-platform reality

Google's guide states plainly that publishers do not need to create new machine-readable files, AI text files, markup, or Markdown to appear in generative AI search. King's response is that this is true for Google and entirely misses the point.

According to King, llms.txt is genuinely useful for Claude and other systems that have explicitly committed to reading it. Anthropic has published documentation supporting the format, and observable benefits exist in environments where it is actually consumed. Telling publishers to ignore it because Google does not read it is, in his framing, exactly the kind of single-platform thinking that produces bad strategy. The honest version of the guidance, he writes, would state that Google does not process llms.txt in any special way, other systems may, and practitioners should decide accordingly. Instead, he argues, Google conflates "we don't use it" with "you don't need it."

The broader implication reaches beyond any single file format. Systems competing with Google have different opinions, different infrastructure, and different incentives. ChatGPT, Perplexity, Claude, Copilot, Gemini, and a growing tail of vertical agents are all making their own retrieval decisions. Some use Bing as their grounding layer. Some build their own indices. Some read llms.txt. Some are publishing the math behind their retrieval. The shared layer is shrinking, and the practice required to cover the full surface area is larger than what any Google Search Central document describes.

Chunking and passage retrieval

Google's guide states there is no requirement to break content into small pieces for AI to better understand it, adding that Google systems can understand the nuance of multiple topics on a page.

King spent 4,500 words on this topic in January 2026. The condensed version he offers here starts with a technical baseline: chunking is what RAG systems do to content regardless of whether publishers optimize for it. The question is not whether chunking happens but whether content survives the process with its meaning intact. A passage focused on one idea will retrieve better, in nearly every measurable case, than a passage covering three topics, because vector distance math does not care about Google's preferences on page structure.

He cites Bing's own documentation, which states directly that "chunking/transformations must preserve meaning and claims used in the answer." He also points to Google's own MUVERA research, its work on passage indexing, and its patents on pairwise passage selection. None of that internal work is consistent, he argues, with the public guidance that chunking does not matter. The systems retrieve passages. PPC Land has covered how Google's MUVERA technology played a role in content detection improvements during the June 2025 core update, adding technical context to the passage-level retrieval discussion.

How retrieval actually selects content

Perhaps King's most technically detailed objection involves Google's guidance that publishers do not need to write in a specific way for generative AI search because AI systems can understand synonyms and general meanings.

According to King's analysis, a retrieval system selects passages by computing vector distance against the query embedding. A synthesis pipeline then performs pairwise comparisons between candidate passages to decide which ones get sent to the model. The system is not "understanding" content in any human sense - it is computing a similarity score, ranking by it, and making committed selections. Specificity, entity salience, semantic coherence, and structural clarity all show up in those scores. Loose, generic, multi-topic prose loses those comparisons to tight, specific, self-contained passages.

King writes that empirical evidence exists showing that adjusting passages improves their retrieval scores, and that public APIs allow publishers to verify this on content before publishing. The guidance to ignore that and trust the systems to figure it out is, in his framing, asking practitioners to compete with one hand tied behind their back.

Where traditional SEO still holds

King is careful to avoid the opposite error. Technical structure matters. Crawlability matters. Page experience matters. Unique, non-commodity content matters. None of that is going away. Google's guide on those points, he writes, is fine. The section on creating valuable, non-commodity content - unique point of view, first-hand experience, content organized for human readers - contains nothing to dispute.

The problem is that "SEO best practices" has historically functioned as shorthand for "what Google likes." That proxy was adequate when Google was 90% of traffic and nothing else mattered. It is not adequate now. The discipline required to maintain visibility across ChatGPT, Perplexity, Claude, Copilot, Gemini, and emerging vertical agents requires a broader practice than any Search Central document describes - and that practice, King argues, is being built right now by practitioners who refuse to treat Google's opinion as the only one that counts.

This broader challenge has been building for some time. PPC Land reported in July 2025 that AI Overviews had already reduced organic clicks by 34.5% when present in search results, with zero-click searches reaching 69% since the May 2024 launch of AI Overviews. Google's own SVP Nick Fox told publishers in December 2025 that optimization for AI search requires no changes from traditional SEO - a message consistent with the May 2026 guide. The Google Search Relations team made the same argument in a December 17, 2025 podcast, and Danny Sullivan reinforced it on January 8, 2026, explicitly discouraging content chunking for LLMs.

Reaction

The LinkedIn post accompanying King's article attracted quick responses from across the industry. Krishna Madhavan, Principal Product Manager at Microsoft AI and Bing, left a brief reaction. King's reply to Madhavan read: "Keep fighting the good fight! We appreciate you for all you do!" Nikita Vlasyuk, CTO at DeepSEO, noted in the comments that AI Overviews are hitting over 1 billion users and that more than half of Google searches end without any clicks. Vlasyuk cited a Gartner prediction that traditional search volume could drop substantially. The "janitors of the web" phrase from King's post drew multiple responses, with Anthony Nichols of collystring calling it "savage and I approve."

Matthew Proctor, CEO at Narrative Bent, offered a more structural observation: the SEO versus GEO debate is confusing practitioners because, even if the disciplines are different and require different skills, the work will continue to fall on the people currently responsible for search - regardless of what the job title says.

Timeline

Summary

Who - Mike King, founder and CEO of iPullRank, published the analysis. The primary subject is Google's Search Central guidance on AI search optimization, and the contrast figure is Microsoft's Bing team, including Jordi Ribas and Krishna Madhavan.

What - King argues that Google's updated guide, which dismisses GEO and AEO as simply "still SEO" and tells publishers to ignore chunking, llms.txt, and passage-level writing optimization, reflects Google's interest in keeping practitioners focused on its own platform rather than the multi-platform reality of AI search. He cites technical contradictions between the guide and Google's own MUVERA research and passage indexing work, and contrasts Google's posture with the more technically transparent publications from Microsoft's Bing team.

When - Google published the guide on May 15, 2026. King published his critique on May 18, 2026, the same day the LinkedIn post and accompanying industry discussion circulated.

Where - The critique was published on ipullrank.com. The Google guidance is live on Google Search Central. The Microsoft posts King references appeared on the Bing Search Blog. Industry reaction appeared on LinkedIn.

Why - The analysis matters for the marketing and SEO community because it questions whether Google's guidance is a neutral technical document or a strategic communication designed to keep practitioners from developing expertise in platforms and channels Google does not control. With AI Overviews now active across 200 countries and AI Mode expanding, the question of how to maintain content visibility across a fragmented AI search landscape is among the most urgent open problems in digital marketing.

Share this article
The link has been copied!