Microsoft AI Search Guide That Google Couldn't Write
In the first half of May 2026, the two largest companies in search published guidance on AI search that contradicts at the level of first principles. The post that landed on every SEO LinkedIn feed was Google's. The post the SEO industry quietly missed was Microsoft's, and Microsoft published first.
On May 6, three senior engineers at Microsoft AI published a post on the Bing blog titled "Evolving role of the index: From ranking pages to supporting answers." It is the most important piece of writing on AI search this year. Almost no one in our community shared it.
Nine days later, on May 15, Google followed with its AI Optimization Guide. "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Ignore AEO. Ignore GEO. Ignore llms.txt, chunking, structured data, inauthentic mentions. Just keep doing SEO. Yesterday we argued that this framing was incomplete.
The Microsoft post, written before Google's guide was even published, reads like a rebuttal of it anyway. That is what happens when one company is describing the technical reality and the other is positioning around it. The reality does not care about the publication order, and Microsoft is not coincidentally the company that actually powers the retrieval layer behind ChatGPT, the engine sending 95% of AI-referred traffic into the clients we work with.
These two posts cannot both be right. One is positioning. The other is telling the truth. The interesting question is which is which, and why the industry kept treating Google's version as the default while the Microsoft post sat there for two weeks unread.
Microsoft says the quiet part out loud
The thesis of the Microsoft post is in its subtitle: "Same Foundations. Different Optimization Problems."
Same crawlers. Same quality signals. Same deep understanding of the web. But the optimization problem on top of that foundation is fundamentally different depending on whether you are ranking pages for a human to click or grounding facts for an AI to commit to.
This is the sentence Google's guide spent four thousand words avoiding: "A common misconception is that grounding replaces search. It does not. Grounding builds on the same foundational infrastructure but it adds a new optimization layer on top."
Google never used the word grounding. Google never named the new optimization layer. Google's entire rhetorical move was to collapse the distinction Microsoft just drew in writing.
To make sure no one missed the point, Microsoft laid the two systems out side by side. The table is a quiet demolition of the "AI search is just SEO" position.
Traditional search asks which pages a user should visit. Grounding asks what information an AI system can responsibly use to construct an answer. Those are not the same question. The first one is solved by ranking. The second one is solved by something Google does not have an established discipline for, which is why Google's guide pretends the second question is just a flavor of the first.
The unit of value shifts too. Traditional search optimizes around the document. Grounding optimizes around "groundable information: discrete, supportable facts with clear provenance." A page that ranks well can still be useless for grounding if its facts cannot be cleanly extracted, attributed, and verified. Conversely, a page that ranks badly can be heavily cited by AI engines if its facts are dense, attributed, and unambiguous.
The role of the user changes. In ranking, the human in the loop scans, skips, and self-corrects. In grounding, the user sees a synthesized answer and has to actively check sources to verify it. The error tolerance collapses. Imperfect ranking is recoverable. Imperfect grounding produces a confidently wrong sentence.
And the valid outcomes change. Ranking returns options. Grounding either answers when supported, or abstains when the evidence is insufficient. That last word, abstain, does not appear anywhere in Google's guide. Microsoft puts it in the middle of the table as a first-class outcome, alongside answering. Because in a grounded system, the right thing to do when the index does not contain enough evidence is to not answer.
You cannot read that table and still believe AI search is "just SEO." That table is the entire argument.
Why this matters: the index is being rebuilt, not just reused
Microsoft's second table is even more uncomfortable for the SEO industry, because it spells out what gets measured differently when the index is feeding AI answers instead of ranking pages.
Factual fidelity becomes critical. In ranking, an indexed representation that distorts the page slightly is a ranking nuisance. In grounding, that same distortion produces a wrong answer. The chunking and transformation processes the index uses, the same chunking SEOs have been told to ignore by Google, are exactly where Microsoft says the index can break grounding silently.
Source attribution quality stops being a nice-to-have and becomes a core signal. Not all indexed content carries equal evidentiary weight. An AI system grounding an answer needs to know which sources can be relied on for what kinds of claims. That is provenance, citation lineage, and entity authority working together as a measurement system that simply does not exist on the ranking side.
Freshness gets a new failure mode. Stale ranking is a degraded experience. Stale grounding produces a misleading answer. The cost of stale facts inside a grounded AI response is categorically higher than the cost of a slightly outdated organic result. This is why Perplexity, the engine that puts freshness at the center of its retrieval logic, cites recent content at roughly twice the rate of older content. It is not a quirk. It is the right design for the problem.
Coverage of high-value facts replaces coverage of pages. The question is no longer "is this URL in the index" but "are the facts our users ask about actually retrievable and groundable in this index." A web that is broadly indexed but full of paraphrased commodity content is not a useful grounding substrate. A web with denser primary-source coverage on the specific questions people ask AI engines is.
And conflict detection becomes a new responsibility for the index itself. When two indexed sources contradict each other, a ranking system can put one above the other and let the user arbitrate. A grounding system cannot, because the user does not see both. The index now has to register the contradiction explicitly, or the AI on top of it will silently pick one and confidently assert it.
None of these dimensions are in Google's guide. None of them. Google's guide is a list of things you do not need to do. Microsoft's post is a description of what is actually being built.
The strategic read: who has the incentive to be honest
Read both posts back to back and the strategic motivation behind each becomes very hard to miss.
Google's incentive is to keep SEO anchored to Google. Their entire business model depends on the discipline of search optimization being functionally identical to "what works on google.com." If practitioners conclude that AI search is a different problem requiring different work, the gravitational pull of Google's documentation, Google's tools, and Google's surfaces weakens. The AI Optimization Guide is, in part, a defensive document. It is Google reminding the industry that the playbook is still Google's playbook.
Microsoft's incentive runs the opposite direction. Bing's organic search is the second-place ranking system in a market where second place has not mattered much for a decade. But the Bing index powers ChatGPT, the largest AI search interface by query volume after Google itself, and Microsoft has every commercial reason to convince the industry that grounding for AI answers is a separate, important, technically demanding problem that requires its own attention. Because the more the industry treats grounding as its own discipline, the more relevant Microsoft becomes inside it.
Both companies are arguing their book. That is normal. The difference is that Microsoft's commercial interest happens to line up with the technical reality, and Google's commercial interest happens to require flattening that reality into "still SEO."
When you have to choose between a marketing document from the incumbent who needs the old framing to survive, and a technical paper from the upstart who is being rewarded by telling you how the new system actually works, the technical paper is the one to bring into client conversations.
The practical implications nobody is talking about
If Microsoft is right (and the evidence sitting in our client GA4 dashboards says they are), the optimization work for the next two years splits into two tracks, not one.
The ranking track is what Google's guide describes. Strong technical SEO, helpful and non-commodity content, clean indexability, top-ten organic positions, structural HTML, E-E-A-T. This is the foundation. Microsoft says it explicitly: grounding builds on this. Skip it and nothing downstream works.
The grounding track is what Microsoft's post describes and Google's guide refuses to name. It is a different discipline, and it requires different work:
Information design at the passage level. Not page-level optimization, but sentence-level extractability. A claim that can be lifted verbatim from your page, with a clear source attached, is groundable. A claim buried in a 600-word section that requires summarization is not. Most enterprise content fails this test today, including content that ranks in the top ten on Google.
Provenance as a discipline. Microsoft says provenance is now a core grounding signal. That means named authors, named sources, dated statistics, and clear citation lineage stop being aesthetic choices and start being retrieval signals. An article with a strong author byline, internal expertise indicators, and inline citations is more groundable than a stronger but unattributed article from a larger brand.
Freshness as a fact-level commitment, not a page-level one. The question is not "when was this page last updated." The question is "when was each specific claim on this page last verified." Pages that put dates on facts, not just on the article, will be cited more reliably by grounded engines.
Contradiction discipline. If your brand or your content asserts something different from the broader web, you need to know it. Microsoft is now telling you, in writing, that the next generation of grounding indexes will register contradictions explicitly. Brands that have contradictory information across owned and earned surfaces will be penalized in ways the old SERP did not capture.
Bing-first thinking for ChatGPT visibility. The Bing index is the gate to ChatGPT, the largest AI search engine. Bing Webmaster Tools is not a backup tool. It is a primary discovery surface for an engine that does not retrieve from Google. Teams that treat Bing as a second-class deployment target are invisible to ChatGPT, and Google's guide will not tell them that.
None of these moves contradicts good SEO. All of them go beyond it. Microsoft just published the technical justification for why that beyond layer matters.
What we are doing with this internally
Search Agency has been running the layered model for two years. We work the SEO foundation. We work the grounding layer. We measure them separately. We optimize for them separately. We do not collapse them, because the engines themselves do not collapse them.
Yesterday we told you Google's guide was incomplete. Microsoft's post, published nine days before Google's and ignored by most of the industry, fills in the part of the picture Google left out. The two posts together form the actual map. Google drew the part of the territory closest to Google. Microsoft drew the rest, and in the process drew the part Google overlaps with too, in technical terms Google's guide could not use without admitting the problem was bigger than its document.
If you are a practitioner reading both, the takeaway is not "pick a side." The takeaway is that the side that needs you to think one way is the side telling you it is still SEO. The side that benefits from you understanding the actual mechanics is the side telling you it is grounding. Track the incentives. Read both. Build for the world Microsoft just described, because that is the world ChatGPT, Perplexity, Claude, and increasingly Gemini itself are operating inside.
The AI search wars just got a lot more interesting. Anyone still telling you it is "just SEO" is either reading one document, or selling you the position that benefits the company that wrote it.
