Google’s AI search guidance is naive and self-serving

Every time Google ships a new Search Central document, two camps in our industry move at the speed of light. The first camp screenshots their favorite paragraph, posts it to LinkedIn with “SEE? IT’S JUST SEO” in the caption, and goes back to doing exactly what they were already doing. The second camp screenshots a different paragraph and posts it with “see, here’s the proof they’re lying to us.” Both camps treat Google’s guidance like scripture, depending on which verse confirms what they already believed.

Google’s recently updated guide on Optimizing your website for generative AI features on Google Search was a feast for the first camp. The “it’s just SEO” folks ate well that week. AEO and GEO got declared “still SEO.” Chunking got dismissed. llms.txt got dissed. Rewriting for AI got nullified. If you’ve spent the last two years on LinkedIn telling everyone that nothing has changed, Google handed you a gold star and a victory lap.

But I want to remind everyone of something the first camp likes to forget: two years ago, we held thousands of pages of Google’s internal Search ranking documentation in our hands. The leaked Content Warehouse documents showed, in Google’s own words, how the public guidance and the internal reality diverge. The same company that publicly insisted certain signals didn’t exist had them named, weighted, and documented inside their own engineering wiki. That wasn’t a leak from an enemy of search. That was Google’s own engineering documentation, and it showed exactly how much we should trust public guidance about what is and isn’t important.

I’m not saying every line of Google’s new guide is a lie. I am saying that Google has a long, well-documented history of nudging the industry in directions that benefit Google first and the open web maybe. It’s to Google’s benefit for SEOs to remain the janitors of the web cleaning up technical debt, formatting structured data, and politely waiting for the next algorithm update rather than evolving into a discipline that operates across multiple platforms and influences how content is engineered for systems Google does not control.

As I argued in my refutation of the misinformation about chunking, the influence Google has spent two decades accumulating is finally fragmenting. Competitive AI platforms are stealing attention. Referral traffic is shrinking. Investment is moving to channels Google doesn’t own. The leverage Google had to define what “good content” means is weaker than it has been in twenty years — and you can hear it in how protective the language has gotten.

Meanwhile, in Redmond

For a clean contrast, look at what’s been coming out of Bing.

Krishna Madhavan and his team have spent the last several months publishing posts that read like the opposite of Google’s guide. Keep in mind that there is near parity in both platforms’ offerings.

Where Google’s posture is “trust us, keep doing what you were doing,” Bing has been publicly explaining how their index is changing, what grounding actually requires, and giving publishers tools to measure how their content participates in AI answers.

In Elevating the Role of Grounding on the AI Web, Jordi Ribas openly names what’s happening: agents are doing the browsing now, they’re drawn to structured and verifiable content, and a new optimization discipline called Generative Engine Optimization is emerging in response. No dismissive air quotes. No “it’s all still SEO.” They just call it what it is.

Introducing AI Performance in Bing Webmaster Tools goes further. It is, in Microsoft’s own words, “an early step toward Generative Engine Optimization (GEO) tooling in Bing Webmaster Tools.” Citations across Copilot and Bing’s AI summaries. Page-level citation activity and  Grounding queries, you know, the actual phrases AI used when retrieving your content. The thing every working AI Search practitioner has been asking for, Bing shipped it.

Then, in Evolving role of the index: From ranking pages to supporting answers, Krishna’s team explains in plain detail that “the unit of value shifts from documents to groundable information — discrete, supportable facts with clear provenance.” They state directly that “chunking/transformations must preserve meaning and claims used in the answer.” They acknowledge that the metrics, the unit of analysis, and the responsibility of the system have all changed.

Read those three posts in order, then go re-read Google’s “mythbusting” section. You will struggle to believe you’re reading documents about the same technology.

Going point by point

With that framing, let’s walk through Google’s claims.

“Is SEO still relevant for generative AI search?”

“What about ‘AEO’ and ‘GEO’? ‘AEO’ stands for ‘answer engine optimization’ and ‘GEO’ for ‘generative engine optimization’. These are both terms you may see used to describe work specifically focused on improving visibility in AI search experiences. From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”

“It’s just SEO” is naive, and it’s naive for the same reason it has been naive every previous time someone trotted it out.

SEO as a discipline is not a list of tactics. It’s a mindset, a set of organizational expectations, a budget line, and a reporting structure. SEOs have been trying to expand that mindset for years to bring in content engineering, to influence product, to own technical architecture, to participate in video, brand, and design. We mostly haven’t won those fights, because the org charts in most companies treat SEO as a downstream cleanup function.

This is also the same trick the industry has played on us for fifteen years. Mobile was “just SEO.” Voice was “just SEO.” Schema was “just SEO.” AMP was “just SEO,” and we ate years of implementation work for a system Google quietly deprecated. Every time a new surface appears, the discipline absorbs the work, and every time, the line item that pays for it doesn’t grow proportionally. Folding AI Search into “SEO” isn’t a clarification. It’s the continuation of a pattern that has been excellent for Google and lousy for the people doing the work.

• The skill set has diverged whether the title has or not. The traditional SEO toolkit is keyword research, technical auditing, internal linking, structured data, content optimization tools, link building, and rank tracking. The work of AI Search adds information retrieval theory, vector distance measurement, RAG pipeline analysis, content engineering at the passage level, agent and protocol design (MCP, A2A, UCP, ACP), brand citation tracking across LLM platforms, and synthesis evaluation. There is overlap. There is also vast surface area that has never appeared in any SEO job description ever written. Pretending the skill set is the same is how organizations underhire for the actual problem.
• The audience 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 contain a link to your site at all. Those are different consumers with different criteria, different measurement, and different reporting. Pretending the audience hasn’t changed is how you end up running the wrong tactics against the wrong KPIs for the wrong stakeholders.
• The strategic cost of “just SEO” is concrete. When a brand asks “how do we show up in ChatGPT?” and you treat that as an SEO problem, you start optimizing pages and chasing indexing. The actual answer often has very little to do with your website. It involves your presence in Wikipedia, Reddit, third-party publications, and the licensed data partners that feed model training and grounding. That isn’t on-page work. That is brand, PR, third-party data, and information architecture across the open web. An SEO budget rarely funds that work. A GEO or AEO budget can.

When AI Search lands in an organization with a different name, it gets different expectations and a different budget. It gets cross-functional sponsorship. It gets executive attention. It gets the cross-discipline collaboration SEOs have been requesting since I started in this industry two decades ago. “AEO” and “GEO” are not magical incantations, but the labels create the room SEO has not been able to create for itself.

Meanwhile, the practitioners doing this work keep getting handed more responsibility. More platforms to optimize for. More systems to understand. More research papers to read. More tooling to build. None of that comes with new headcount or higher salaries when leadership sees it as “still SEO.” Google reframing this work as the same old discipline isn’t a neutral observation. It is the rhetorical move that keeps the work uncompensated.

And note: Google itself doesn’t actually run on “it’s just Search.” AI Mode, AI Overviews, and classic ranking are different systems run by different teams on different infrastructure with different evaluation criteria. The leaked Content Warehouse docs made those distinctions visible. Their public posture flattens the inside of their own org for the benefit of the outside narrative. We don’t have to accept the flattening.

That’s what “it’s just SEO” actually delivers to organizations: more scope, same budget, no new authority. That’s a fantastic outcome for the platforms that benefit from our unpaid labor. It is a terrible outcome for the people doing the work.

Non-commodity content

“Create valuable, non-commodity content for your audience”

This part is fine. Make good, unique content with a real point of view. Nobody serious disagrees. Moving on.

llms.txt files and other ‘special’ markup

“You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.”

True for Google. Also missing the point.

llms.txt is genuinely useful for Claude and a handful of other systems that have explicitly committed to reading it. Anthropic has documentation suggesting it. There are observable benefits to publishing it in environments where it’s actually consumed. Telling people to ignore it because Google doesn’t read it is exactly the kind of single-platform myopia I keep pointing at. Google’s guide describes one ecosystem. Your strategy needs to account for several.

The honest version of this guidance would be: “Google doesn’t process llms.txt in any special way. Other systems may. Make your own call.” Instead, Google quietly conflates “we don’t use it” with “you don’t need it.”

‘Chunking’ content

“There’s no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users.”

I wrote 4,500 words on this in January and I’d rather not relitigate the whole thing here. The short version is this: chunking is what RAG systems do to your content, whether you optimize for it or not. The question is whether your content survives the chunking process with its meaning intact, or whether it shatters into incoherent fragments. The vector math doesn’t care about Google’s preferences. A passage that focuses on one idea will, in nearly every measurable case, retrieve better than a passage that tries to cover three.

Bing acknowledges this directly: “chunking/transformations must preserve meaning and claims used in the answer.” Google’s own MUVERA research, their work on passage indexing, their patents on pairwise passage selection — none of it is consistent with the guidance that chunking doesn’t matter. The systems retrieve passages. Treat your passages like they matter, because the systems do.

Rewriting content just for AI systems

“You don’t need to write in a specific way just for generative AI search. AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words. This means you don’t have to worry that you don’t have enough ‘long-tail’ keywords or haven’t captured every variation of how someone might seek content like yours.”

This is the line that bothers me most, because it is antithetical to how these systems actually decide what to use.

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” your content in the human sense — it’s 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. Write loose, generic, multi-topic prose and your passages lose those comparisons to passages that are tight, specific, and self-contained.

“Just write naturally for humans” sounds like good advice until you realize the systems have a measurable preference, and you can win or lose on the margins by writing for both. We have empirical evidence that adjusting passages improves their retrieval scores. We have access to public APIs that let us verify this on the content we publish. The guidance to ignore all of that and trust the systems to figure it out is asking you to compete with one hand tied behind your back.

SEO best practices still help. They just don’t cover the whole map.

I want to be careful here because this discourse gets reduced to extremes:

• SEO best practices help.
• Technical structure matters.
• Crawlability matters.
• Page experience matters.
• Unique, non-commodity content matters.
• None of that is going anywhere.

But “SEO best practices” was always shorthand for “what Google likes.” That was a fine proxy when Google was 90% of the traffic, and the rest didn’t matter. It is not a fine proxy in a world where ChatGPT, Perplexity, Claude, Copilot, Gemini, and a long tail of vertical agents are all making their own retrieval decisions on different infrastructure with different priorities. Some of those systems use Bing as their grounding layer. Some build their own indices. Some lean on llms.txt. Some don’t. Some are shipping webmaster tooling. Some are publishing the math behind their retrieval. The shared layer is shrinking, and the surface area you have to actually optimize is growing.

The thing Google’s guide doesn’t say — because it can’t — is that the systems competing with Google have different opinions, different infrastructure, and different incentives. Optimizing for all of them at once requires a broader practice than what’s described in any Search Central document. That practice is being built right now, in public, by people who refuse to accept that the only opinion that counts is Google’s.

A new world, a lot of opinions

Google’s guidance on AI Search is one opinion. It is the opinion of the company with the most to lose from a multi-platform world. Read it. Take what is useful. Apply it where it applies. Don’t mistake it for the truth.

The truth is that we are in a new world. The infrastructure for how information is retrieved and presented is being rewritten across multiple platforms simultaneously, and the consensus we once had about how to optimize it no longer exists. Bing is publishing what they’re doing. Anthropic is publishing what they’re doing. The research community is publishing what they’re doing. Google is publishing what it wants you to do.

That last one is not the same as the others. Treat it accordingly.

This article was originally published on the iPullRank blog and is republished with permission.