Google and Bing now run opposite GEO playbooks

Four days before Google I/O, Google Search Central published an optimisation guide for generative AI features. It told the industry that AEO and GEO are "still SEO," that llms.txt isn't needed, that chunking strategies are a distraction, and that content rewriting for AI is unnecessary. Trust us. Keep doing what you were doing.

In the same period, Microsoft's Bing team published the opposite. Jordi Ribas wrote *Elevating the Role of Grounding on the AI Web*, naming Generative Engine Optimisation as a real emerging discipline. Krishna Madhavan announced AI Search reporting features for Bing Webmaster Tools — actual page-level citation data, the kind Google has refused to give publishers. At SEO Week, Madhavan stood on stage and dropped not-yet-announced tooling live, walking through how Bing's index actually treats content.

Two of the largest search platforms in the world, with near-parity in their AI offerings, are now telling website owners completely contradictory things about what to do. That is the actual story of this week. Not the I/O keynote. Not the panic. The divergence.

And it tells you almost everything you need to know about who's actually confident in their position.

The honest read of Google's guidance

Google's guide isn't lying. It's just self-serving in a way that's easy to miss if you read it as neutral documentation.

The system-level claim — that AI Overviews and AI Mode are grounded in the same index, ranked by the same systems, and respond to the same fundamentals — is correct. Mike King has made this point repeatedly. The infrastructure is one infrastructure. AI features are a presentation layer over the existing retrieval stack. In that narrow sense, "it's still SEO" is technically true.

But system-level correctness underestimates the interface problem. The thing that determines whether your business gets traffic, citations, or zero-click obliteration isn't the index. It's the presentation layer Google sits on top of it. And the presentation layer has changed enormously. Telling publishers "the fundamentals are the same" while quietly redesigning the surface every quarter is the most Google move imaginable.

The dismissals are where the guide moves from "self-serving" to "actively misleading." Google's own Lighthouse tool includes an llms.txt audit. Chrome's developer documentation suggests considering it. Search Central tells you to skip it. That isn't strategic clarity — that's three teams inside the same company contradicting each other, with the team that talks to publishers picking the option most convenient for Google.

What Bing is doing differently

Microsoft has, for now, decided to be honest. That's the simplest way to read it.

Bing's posture is: agents are doing the browsing, structured and verifiable content is what they reach for, a new optimisation discipline is emerging, and here are the tools to measure how your content participates. They call it GEO. They don't air-quote it. They built citation reporting into Webmaster Tools because publishers asked for it. They're publishing technical explanations of how grounding works rather than vague reassurances that it's all fine.

There's a reason for this, and it isn't that Microsoft has discovered the moral high ground. Bing has roughly nothing to lose by being transparent about AI search mechanics. Their market share floor is already at a level where additional candour can only help. Google has enormous incentive to keep the publisher class calm, compliant, and convinced nothing has structurally changed — because the moment publishers start optimising aggressively for non-Google surfaces, Google's leverage erodes.

When two competitors with similar technology adopt opposite communication strategies, the cheaper move is almost always the one made by the company with less to defend.

Why this matters for what you actually do

If you take Google's guidance at face value, your roadmap is: keep doing solid SEO, ignore llms.txt, don't worry about chunking, write good content, trust the system.

If you take Bing's at face value, your roadmap is: structure your content so machines can ground claims to specific passages, expose verifiable signals, monitor citation behaviour across AI surfaces, treat GEO as a discipline with measurable inputs and outputs.

These are not the same roadmap. They are not even adjacent. They imply different content architectures, different measurement frameworks, different budget allocations.

The pragmatic position is the one I've held in the machine-readability piece earlier this month: the fundamentals genuinely haven't changed, but the *structural* requirements have. Semantic HTML, clean information architecture, retrievable passages, schema that actually describes entities rather than ticking a box — these are SEO fundamentals, but they're fundamentals that most sites have been allowed to ignore for a decade because Google was forgiving enough to compensate. AI retrieval is less forgiving. Bing is telling you that out loud. Google is telling you not to worry about it.

One of them is being more useful to you. It isn't the one with the larger market share.

The leak nobody seems to remember

Two years ago we had thousands of pages of Google's internal Content Warehouse documentation in our hands. The documents named signals Google had publicly denied existed. They weighted things Google had insisted were irrelevant. They confirmed what experienced SEOs had suspected for years: the gap between public guidance and internal reality at Google is not narrow.

That isn't a conspiracy theory. That's the documentation. And the people now telling you Google's GEO guidance is the authoritative word are the same people who, two years ago, were posting screenshots of the leak with stunned-face emojis.

The lesson from the leak wasn't that Google lies in some Bond-villain sense. It's that Google's public guidance is communications output, not engineering documentation. It's written by a team whose job is to manage publisher behaviour, not to give publishers the most accurate possible picture of how the system works. That's true of every large platform. The mistake is treating any of it as scripture.

What to actually do this week

Read both guides. Read Google's optimisation document. Read Ribas's grounding piece. Read Madhavan's posts. Notice where they agree — which is more than you'd expect — and notice where they diverge.

Where they agree, that's probably the genuine technical reality. Structured data matters. Verifiable claims matter. Authoritative entities matter. Speed and crawlability matter. Brand strength matters more than anything else, which neither platform says explicitly enough but both keep implying.

Where they diverge, ask yourself which company has more reason to obscure the answer. The dismissals in Google's guide — llms.txt, chunking, rewriting — are the places to look harder, not the places to stop looking.

And if you're paying an agency that's currently quoting Google's guidance as proof you don't need to change anything: get a second opinion. The agencies most aggressive about telling you nothing has changed are the agencies with the most incentive for nothing to change.

The honest limit

I can't tell you with certainty that Bing's framing is correct and Google's is misleading. Both companies have commercial reasons for what they're publishing. Both contain real information. The conversation will look different in six months as more data emerges from publishers running controlled tests on AI surfaces.

What I can tell you is that when two competitors describe the same technology in opposite terms, the asymmetry is the signal. The party with less to lose is usually the party telling you something closer to what's actually happening. That's the loop. And right now Bing is on the right side of it.

Whether that lasts is a different question. Microsoft has been the honest one before, found it commercially convenient for about eighteen months, and then quietly reverted to platform-incumbent behaviour. Don't mistake this moment for a permanent character trait. Use the candour while it's on offer.