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

Each time Google ships a brand new Search Central doc, two camps in our trade transfer on the pace of sunshine. The primary camp screenshots their favourite paragraph, posts it to LinkedIn with “SEE? IT’S JUST search engine optimisation” within the caption, and goes again to doing precisely what they had been already doing. The second camp screenshots a unique paragraph and posts it with “see, right here’s the proof they’re mendacity to us.” Each camps deal with Google’s steering like scripture, relying on which verse confirms what they already believed.

Google’s just lately up to date information on Optimizing your website for generative AI features on Google Search was a feast for the primary camp. The “it’s simply search engine optimisation” of us ate nicely that week. AEO and GEO acquired declared “nonetheless search engine optimisation.” Chunking acquired dismissed. llms.txt acquired dissed. Rewriting for AI acquired nullified. When you’ve spent the final two years on LinkedIn telling everybody that nothing has modified, Google handed you a gold star and a victory lap.

However I need to remind everybody of one thing the primary camp likes to overlook: two years in the past, we held hundreds of pages of Google’s inner Search rating documentation in our palms. The leaked Content Warehouse documents confirmed, in Google’s personal phrases, how the general public steering and the interior actuality diverge. The identical firm that publicly insisted sure alerts didn’t exist had them named, weighted, and documented inside their very own engineering wiki. That wasn’t a leak from an enemy of search. That was Google’s personal engineering documentation, and it confirmed precisely how a lot we must always belief public steering about what’s and isn’t necessary.

I’m not saying each line of Google’s new information is a lie. I’m saying that Google has a protracted, well-documented historical past of nudging the trade in instructions that profit Google first and the open internet possibly. It’s to Google’s profit for SEOs to stay the janitors of the online cleansing up technical debt, formatting structured information, and politely ready for the following algorithm replace relatively than evolving right into a self-discipline that operates throughout a number of platforms and influences how content material is engineered for programs Google doesn’t management.

As I argued in my refutation of the misinformation about chunking, the affect Google has spent twenty years accumulating is lastly fragmenting. Aggressive AI platforms are stealing consideration. Referral site visitors is shrinking. Funding is shifting to channels Google doesn’t personal. The leverage Google needed to outline what “good content material” means is weaker than it has been in twenty years — and you’ll hear it in how protecting the language has gotten.

In the meantime, in Redmond

For a clear distinction, have a look at what’s been popping out of Bing.

Krishna Madhavan and his staff have spent the final a number of months publishing posts that learn like the alternative of Google’s information. Needless to say there’s close to parity in each platforms’ choices.

The place Google’s posture is “belief us, preserve doing what you had been doing,” Bing has been publicly explaining how their index is altering, what grounding really requires, and giving publishers instruments to measure how their content material participates in AI solutions.

In Elevating the Role of Grounding on the AI Web, Jordi Ribas brazenly names what’s taking place: brokers are doing the searching now, they’re drawn to structured and verifiable content material, and a brand new optimization self-discipline referred to as Generative Engine Optimization is rising in response. No dismissive air quotes. No “it’s all nonetheless search engine optimisation.” They simply name it what it’s.

Introducing AI Performance in Bing Webmaster Tools goes additional. It’s, in Microsoft’s personal phrases, “an early step towards Generative Engine Optimization (GEO) tooling in Bing Webmaster Instruments.” Citations throughout Copilot and Bing’s AI summaries. Web page-level quotation exercise and  Grounding queries, you already know, the precise phrases AI used when retrieving your content material. The factor each 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 staff explains in plain element that “the unit of worth shifts from paperwork to groundable data — discrete, supportable information with clear provenance.” They state instantly that “chunking/transformations should protect that means and claims used within the reply.” They acknowledge that the metrics, the unit of study, and the accountability of the system have all modified.

Learn these three posts so as, then go re-read Google’s “mythbusting” part. You’ll battle to consider you’re studying paperwork about the identical know-how.

Going level by level

With that framing, let’s stroll by Google’s claims.

“Is search engine optimisation nonetheless related for generative AI search?”

“What about ‘AEO’ and ‘GEO’? ‘AEO’ stands for ‘reply engine optimization’ and ‘GEO’ for ‘generative engine optimization’. These are each phrases you may even see used to explain work particularly centered on enhancing visibility in AI search experiences. From Google Search’s perspective, optimizing for generative AI search is optimizing for the search expertise, and thus nonetheless search engine optimisation.”

“It’s simply search engine optimisation” is naive, and it’s naive for a similar cause it has been naive each earlier time somebody trotted it out.

search engine optimisation as a self-discipline isn’t a listing of ways. It’s a mindset, a set of organizational expectations, a funds line, and a reporting construction. SEOs have been attempting to broaden that mindset for years to usher in content material engineering, to affect product, to personal technical structure, to take part in video, model, and design. We principally haven’t gained these fights, as a result of the org charts in most corporations deal with search engine optimisation as a downstream cleanup perform.

That is additionally the identical trick the trade has performed on us for fifteen years. Cellular was “simply search engine optimisation.” Voice was “simply search engine optimisation.” Schema was “simply search engine optimisation.” AMP was “simply search engine optimisation,” and we ate years of implementation work for a system Google quietly deprecated. Each time a brand new floor seems, the self-discipline absorbs the work, and each time, the road merchandise that pays for it doesn’t develop proportionally. Folding AI Search into “search engine optimisation” isn’t a clarification. It’s the continuation of a sample that has been wonderful for Google and awful for the folks doing the work.

• The talent set has diverged whether or not the title has or not. The standard search engine optimisation toolkit is key phrase analysis, technical auditing, inner linking, structured information, content material optimization instruments, hyperlink constructing, and rank monitoring. The work of AI Search provides data retrieval concept, vector distance measurement, RAG pipeline evaluation, content material engineering on the passage degree, agent and protocol design (MCP, A2A, UCP, ACP), model quotation monitoring throughout LLM platforms, and synthesis analysis. There’s overlap. There’s additionally huge floor space that has by no means appeared in any search engine optimisation job description ever written. Pretending the talent set is identical is how organizations underhire for the precise drawback.
• The viewers modified, too. Conventional search engine optimisation optimizes for one machine and the people clicking its outcomes. AI Search optimizes for a retrieval system, a synthesis pipeline, probably an agentic browser, and a human studying a solution that won’t include a hyperlink to your website in any respect. These are completely different customers with completely different standards, completely different measurement, and completely different reporting. Pretending the viewers hasn’t modified is how you find yourself working the improper ways towards the improper KPIs for the improper stakeholders.
• The strategic price of “simply search engine optimisation” is concrete. When a model asks “how can we present up in ChatGPT?” and also you deal with that as an search engine optimisation drawback, you begin optimizing pages and chasing indexing. The precise reply usually has little or no to do together with your web site. It includes your presence in Wikipedia, Reddit, third-party publications, and the licensed information companions that feed mannequin coaching and grounding. That isn’t on-page work. That’s model, PR, third-party information, and knowledge structure throughout the open internet. An search engine optimisation funds hardly ever funds that work. A GEO or AEO funds can.

When AI Search lands in a company with a unique title, it will get completely different expectations and a unique funds. It will get cross-functional sponsorship. It will get govt consideration. It will get the cross-discipline collaboration SEOs have been requesting since I began on this trade twenty years in the past. “AEO” and “GEO” aren’t magical incantations, however the labels create the room search engine optimisation has not been capable of create for itself.

In the meantime, the practitioners doing this work preserve getting handed extra accountability. Extra platforms to optimize for. Extra programs to grasp. Extra analysis papers to learn. Extra tooling to construct. None of that comes with new headcount or larger salaries when management sees it as “nonetheless search engine optimisation.” Google reframing this work as the identical previous self-discipline isn’t a impartial statement. It’s the rhetorical transfer that retains the work uncompensated.

And observe: Google itself doesn’t really run on “it’s simply Search.” AI Mode, AI Overviews, and basic rating are completely different programs run by completely different groups on completely different infrastructure with completely different analysis standards. The leaked Content Warehouse docs made these distinctions seen. Their public posture flattens the within of their very own org for the good thing about the skin narrative. We don’t have to simply accept the flattening.

That’s what “it’s simply search engine optimisation” really delivers to organizations: extra scope, identical funds, no new authority. That’s a unbelievable final result for the platforms that profit from our unpaid labor. It’s a horrible final result for the folks doing the work.

Non-commodity content material

“Create helpful, non-commodity content material in your viewers”

This half is ok. Make good, distinctive content material with an actual viewpoint. No one critical disagrees. Transferring on.

llms.txt information and different ‘particular’ markup

“You don’t must create new machine readable information, AI textual content information, markup, or Markdown to seem in generative AI search.”

True for Google. Additionally lacking the purpose.

llms.txt is genuinely helpful for Claude and a handful of different programs which have explicitly dedicated to studying it. Anthropic has documentation suggesting it. There are observable advantages to publishing it in environments the place it’s really consumed. Telling folks to disregard it as a result of Google doesn’t learn it’s precisely the sort of single-platform myopia I preserve pointing at. Google’s information describes one ecosystem. Your technique must account for a number of.

The trustworthy model of this steering could be: “Google doesn’t course of llms.txt in any particular method. Different programs could. Make your personal name.” As a substitute, Google quietly conflates “we don’t use it” with “you don’t want it.”

‘Chunking’ content material

“There’s no requirement to interrupt your content material into tiny items for AI to raised perceive it. Google programs are capable of perceive the nuance of a number of subjects on a web page and present the related piece to customers.”

I wrote 4,500 words on this in January and I’d relatively not relitigate the entire thing right here. The quick model is that this: chunking is what RAG programs do to your content material, whether or not you optimize for it or not. The query is whether or not your content material survives the chunking course of with its that means intact, or whether or not it shatters into incoherent fragments. The vector math doesn’t care about Google’s preferences. A passage that focuses on one thought will, in almost each measurable case, retrieve higher than a passage that tries to cowl three.

Bing acknowledges this instantly: “chunking/transformations should protect that means and claims used within the reply.” Google’s personal MUVERA research, their work on passage indexing, their patents on pairwise passage choice — none of it’s in keeping with the steering that chunking doesn’t matter. The programs retrieve passages. Deal with your passages like they matter, as a result of the programs do.

Rewriting content material only for AI programs

“You don’t want to put in writing in a particular method only for generative AI search. AI programs can perceive synonyms and common meanings of what somebody is looking for, with a purpose to join them with content material which may not use the identical exact phrases. This implies you don’t have to fret that you simply don’t have sufficient ‘long-tail’ key phrases or haven’t captured each variation of how somebody may search content material like yours.”

That is the road that bothers me most, as a result of it’s antithetical to how these programs really resolve what to make use of.

A retrieval system selects passages by computing vector distance towards the question embedding. A synthesis pipeline then performs pairwise comparisons between candidate passages to resolve which of them get despatched to the mannequin. The system isn’t “understanding” your content material within the human sense — it’s computing a similarity rating, rating by it, and making dedicated alternatives. Specificity, entity salience, semantic coherence, and structural readability all present up in these scores. Write unfastened, generic, multi-topic prose and your passages lose these comparisons to passages which can be tight, particular, and self-contained.

“Simply write naturally for people” seems like good recommendation till you notice the programs have a measurable choice, and you’ll win or lose on the margins by writing for each. We’ve got empirical proof that adjusting passages improves their retrieval scores. We’ve got entry to public APIs that allow us confirm this on the content material we publish. The steering to disregard all of that and belief the programs to determine it out is asking you to compete with one hand tied behind your again.

search engine optimisation greatest practices nonetheless assist. They simply don’t cowl the entire map.

I need to watch out right here as a result of this discourse will get decreased to extremes:

• search engine optimisation greatest practices assist.
• Technical construction issues.
• Crawlability issues.
• Web page expertise issues.
• Distinctive, non-commodity content material issues.
• None of that’s going wherever.

However “search engine optimisation greatest practices” was all the time shorthand for “what Google likes.” That was a fantastic proxy when Google was 90% of the site visitors, and the remainder didn’t matter. It’s not a fantastic proxy in a world the place ChatGPT, Perplexity, Claude, Copilot, Gemini, and a protracted tail of vertical brokers are all making their very own retrieval choices on completely different infrastructure with completely different priorities. A few of these programs use Bing as their grounding layer. Some construct their very own indices. Some lean on llms.txt. Some don’t. Some are delivery webmaster tooling. Some are publishing the maths behind their retrieval. The shared layer is shrinking, and the floor space you need to really optimize is rising.

The factor Google’s information doesn’t say — as a result of it might’t — is that the programs competing with Google have completely different opinions, completely different infrastructure, and completely different incentives. Optimizing for all of them directly requires a broader follow than what’s described in any Search Central doc. That follow is being constructed proper now, in public, by individuals who refuse to simply accept that the one opinion that counts is Google’s.

A brand new world, lots of opinions

Google’s steering on AI Search is one opinion. It’s the opinion of the corporate with essentially the most to lose from a multi-platform world. Learn it. Take what is beneficial. Apply it the place it applies. Don’t mistake it for the reality.

The reality is that we’re in a brand new world. The infrastructure for a way data is retrieved and introduced is being rewritten throughout a number of platforms concurrently, and the consensus we as soon as had about methods to optimize it now not exists. Bing is publishing what they’re doing. Anthropic is publishing what they’re doing. The analysis neighborhood is publishing what they’re doing. Google is publishing what it needs you to do.

That final one isn’t the identical because the others. Deal with it accordingly.

This text was initially revealed on the iPullRank blog and is republished with permission.