There is a particular kind of corporate sentence that sounds like a hug and behaves like a locked door. Google has become rather fluent in it. Not dishonest, not exactly, because that would be too easy and, frankly, too lazy an accusation. More like 'strategically complete in one direction and curiously silent in another'.
Search I/O 2026 was full of that language. The web is still central. Links still matter. Search is still Search. SEO is still SEO. All comforting enough, especially for the many poor souls still being sold 'AEO packages' and shiny little miracle files for machines that do not care about their little rituals. Yet when you put Google's Search I/O message beside Bing's recent primary-source writing, something more interesting appears. Google is telling publishers what is safe to do. Bing is describing more of the machinery underneath.
And that distinction matters.
Because Google is not saying SEO is dead. It is saying the opposite. Google's official guidance on generative AI features says AI Overviews and AI Mode are rooted in core Search ranking and quality systems, including retrieval-augmented generation, query fan-out and the ordinary, unglamorous requirements of crawlability, indexability, snippet eligibility and useful content.1 That part is true, and it is why I said before that Google's AI optimisation guidance was the funeral of AEO theatre, all that 'special AI markup', 'llms.txt will save us', content-confetti nonsense that has grown like mould around every shift in search.2
Google is right about Google. That does not mean Google has handed us the whole map.
But true is not the same as complete.
Google is right about Google. That does not mean Google has handed us the whole map.
The Search box is no longer just a Search box
At Search I/O 2026, Elizabeth Reid framed the whole thing as a new era for AI Search. AI Mode, according to Google, has surpassed one billion monthly users, its queries have more than doubled every quarter since launch, and overall Google queries reached an all-time high in the previous quarter.3 Sundar Pichai gave the wider I/O frame: AI Overviews now has more than 2.5 billion monthly active users, AI Mode has passed one billion monthly active users, and Search is becoming more like an ongoing conversation.4
That last phrase is the hinge.
Google is not simply putting nicer summaries above blue links. It is rebuilding the behaviour of search around conversation, memory, multimodal input, background agents, personal context and task completion. Search I/O 2026 included an AI-powered Search box, conversational follow-ups from AI Overviews into AI Mode, information agents that monitor blogs, news, social posts and fresh data, agentic booking, Google calling businesses on behalf of users, generative interfaces, custom dashboards, trackers, simulations and mini apps inside Search itself.3
Google's earlier AI Mode material had already told us how this works. AI Mode uses query fan-out, issuing multiple related searches across subtopics and data sources, then pulling the answers together into a single response.5 Google's own PDF says AI Mode uses real-time information from the web and from Google, including the Knowledge Graph, real-world information and shopping data for more than 50 billion products.6
So yes, 'SEO is still SEO', in the same way a house is still a house after you turn the ground floor into a train station. The foundations still matter. The traffic pattern is not the same.
Google's calm sentence and Bing's sharper one
Google's public line is continuity. Bing's public line is transformation. The interesting bit is that technically, they are often describing similar realities. The difference is the direction of the explanation.
Google says generative AI features on Search are grounded in core Search systems. Bing says the index is evolving from ranking pages to supporting answers.7 Google says helpful content and links remain central. Bing says the unit of value is shifting from documents to discrete, supportable facts with clear provenance.7 Google says users still click, and that AI Overview clicks can be higher quality. Bing says traditional analytics misses visibility inside summaries, citations and follow-up journeys, and that brands need to measure impressions, placement in AI answers and citations, not just clicks.8
That is where the Google Speak lives. Not in an outright contradiction, but in the soothing removal of consequence.
That is where the Google Speak lives. Not in an outright contradiction, but in the soothing removal of consequence.
| Question | Google's safe answer | Bing's less comfortable answer | What it means in practice |
|---|---|---|---|
| What is AI Search built on? | Core Search ranking and quality systems, crawlable content, indexable pages and helpful web results.1 | Crawling, understanding and ranking still matter, but grounding asks what information an AI system can responsibly use to construct an answer.7 | You still need proper SEO, but your content also needs to work as evidence. |
| What is being retrieved? | Relevant, high-quality pages from the Search index, used to support AI responses.1 | Supportable facts, current information, provenance and attribution quality.7 | Pages matter, but passages, claims, entities and evidence carry more weight than old page-only thinking allows. |
| What should publishers optimise for? | Non-commodity content, crawlability, indexability, snippet control and no special AI hacks.1 | Structured, verifiable, applicable content that can contribute to answers, citations, reasoning and outcomes.9 | The anti-hack message is correct. The operational layer is wider. |
| How should visibility be measured? | Google largely keeps the public advice close to existing Search behaviour and higher-quality clicks.6 | AI visibility includes answer placement, citations, zero-click influence and later-stage conversions.8 | If your reporting is click-only, you are measuring the old room while the conversation has moved next door. |
| How important is freshness? | Google mentions up-to-date web retrieval, real-time data and fresh sources across AI Mode.5 6 | Bing promotes IndexNow as immediate notification of added, updated and deleted content, reducing dependency on traditional crawling.10 | In AI answers, stale information is not just an SEO nuisance. It is a trust problem. |
This is not a childish 'Bing good, Google bad' argument. That would be its own kind of pantomime, and there is enough of that in search already. Microsoft has its own incentives. It wants the AI web to be described through grounding, IndexNow, Webmaster Tools and citation signals because that is where it can carve out influence. Fair enough. Business is business, and anyone pretending otherwise is either selling you something or has never had to make payroll.
The useful thing is not that Bing is pure. The useful thing is that Bing is more explicit.
What Bing says out loud
Bing's May 2026 article on the evolving role of the index is one of the more important search documents of the year, precisely because it says the quiet part with less varnish. Traditional search, Bing says, asks which pages a user should visit. Grounding asks what information an AI system can responsibly use to construct an answer.7
That is the line.
Read it again if you need to, because it is the little trapdoor under half the modern SEO conversation. In traditional search, the page is the destination. In grounded AI search, the page may be a source, a citation, a fragment of evidence, a corroborating passage, a conflicting view, or no more than raw material for a response the user may never click through to read.
In traditional search, the page is the destination. In grounded AI search, the page may be the evidence.
Bing goes further. It says retrieval for grounding has different error dynamics because mistakes can compound through reasoning steps. It says abstention is a valid outcome when evidence is insufficient, stale or conflicting. It says new measurements are required: factual fidelity, attribution quality, freshness, coverage of high-value facts and contradiction handling.7
That is not 'write a blog post and sprinkle schema on it'. That is an information architecture problem.
In February 2026, Bing described grounding as the layer connecting AI to current, authoritative information beyond the model's training, and said assistants and agents increasingly do the browsing. These systems are drawn to structured, verifiable and applicable content.9 In November 2025, Bing said AI Search changes conversion measurement because people often explore through conversations, build confidence before reaching a site, and click later with stronger intent.8
So when Google tells you SEO remains the foundation, believe it. When Bing tells you the index is becoming answer infrastructure, believe that too.
The trick is having the maturity to hold both ideas without running off to buy a new buzzword.
The old page is becoming a source object
For years, search people thought in pages. Title tags, headings, internal links, schema, topical depth, intent matching, all the usual things. Proper SEO was never as shallow as the cheap version sold on retainer invoices, but even serious SEO often treated the page as the main competitive unit.
AI Search does not remove the page. It changes what the page is asked to do.
A good page now has to perform twice. It must still satisfy a human who lands on it, reads it, trusts it and perhaps buys, books, subscribes or remembers the brand. But it must also contain passages that survive extraction. A definition must make sense when separated from the paragraphs around it. A claim must have enough context to be safely used. A recommendation must show its reasoning. An expert view must be visibly expert, not because you wrote 'expert' in a heading, but because the argument has fingerprints, experience, constraint and a bit of blood in it.
That is what Google calls non-commodity content. It wants information that adds something beyond the same reheated soup found everywhere else: first-hand experience, original insight, clear opinion, specialist knowledge, unique data and genuine usefulness.1
- Do not mistake 'AI Search' for permission to sell machine-facing tricks.
- Do not treat `llms.txt` as a magic visibility switch for Google Search.
- Do not fragment useful content into tiny chunks that lose meaning when separated from the page.
And this is where the theatre merchants get it wrong. They hear 'AI Search' and think the answer is machine-facing tricks. Google has been explicit that it does not require special AI markup, does not treat llms.txt as a ranking or visibility requirement, and does not tell publishers to forcibly chunk content into tiny fragments for AI systems.1
But the opposite mistake is nearly as bad. Some people hear Google's anti-hack guidance and conclude nothing has changed. That is the comfortable error. It lets everyone go back to the same thin content, the same listicle wallpaper, the same 'ultimate guide' with no ultimate anything in it, and pretend the old model will carry the new interface.
It will not.
Google Search is becoming an agentic layer
The most important part of Search I/O 2026 is not the model number, although Gemini 3.5 Flash becoming the default AI Mode model globally is obviously part of the performance story.3 The important part is the movement from answers to actions.
Google is putting agents into Search. Information agents can keep monitoring sources and fresh data in the background.3 AI Mode can use Project Mariner's live web browsing, partner integrations, Knowledge Graph and Google Maps to find restaurant reservations, check real-time availability and send users to booking pages.11 Travel planning in AI Mode uses real-time Search data for flights and hotels, Google Maps data such as photos and reviews, and information from across the web.12
That is no longer merely 'find me information'. That is 'mediate my decision and route my action'.
The visitor may not arrive at the beginning of the journey. The agent may have done the early browsing.
If you run a business, this should sober you up a little. The visitor may not arrive at the beginning of the journey. The agent may have done the early browsing. The comparison may have happened inside the answer layer. The trust formation may have happened before your analytics system even knows the person exists. By the time the user clicks, calls or books, the real persuasion may already be mostly over.
Bing has been more candid on this point. It says traditional analytics often fails to capture visibility in summaries, cited content and follow-up query engagement, and that zero-click visibility can shape awareness and preference before the click happens.8
This is not a reason to panic. It is a reason to stop measuring the wrong thing with great confidence.
The practical map: SEO, grounding and evidence
The grown-up answer is not to rename SEO as GEO, AEO, AISO, or whatever acronym someone with a webinar funnel is currently polishing. The answer is to build search assets that work across three layers: the page, the passage and the proof.
| Layer | What it must do | What weak content does instead |
|---|---|---|
| Page | Be crawlable, indexable, internally linked, fast enough, useful, clearly structured and worthy of a human visit. | Hides useful material behind poor UX, thin summaries, vague headings and generic advice. |
| Passage | Preserve meaning when retrieved as part of query fan-out, AI summaries or grounding. | Depends on surrounding fluff, uses unclear pronouns, buries the answer or fragments the idea into meaningless chunks. |
| Proof | Show evidence, provenance, freshness, experience, authorship, contradiction awareness and practical judgement. | Makes unsupported claims, repeats common knowledge and assumes 'we are experts' is evidence. |
That is the difference between optimising for the old SERP and preparing for the answer infrastructure. The old game rewarded being a good destination. The new game still rewards that, but it also asks whether parts of your page can be trusted as source material.
- Write things worth extracting, not just pages long enough to look substantial.
- Keep important facts current, especially prices, dates, availability, policies and local details.
- Make claims specific, attributed and supported by experience or evidence.
- Use schema where it genuinely clarifies entities, products, FAQs, reviews, organisations or local details.
- Maintain Google Business Profile and Merchant Center where relevant, because Google's own guidance calls them out for eligible businesses.
- Use IndexNow for Bing and participating engines where freshness matters, because immediate update notification is plumbing, not glamour.
This is why the advice remains boring and difficult, which is probably why most people will avoid it. Write things worth extracting. Keep important facts current. Make claims specific. Explain the edge cases. Use schema where it genuinely clarifies entities, products, FAQs, reviews, organisations or local details. Maintain your Google Business Profile and Merchant Center where relevant, because Google's own guidance calls them out for eligible businesses.1 Use IndexNow for Bing and participating engines where freshness matters, because immediate update notification is not glamour, it is plumbing.10
And above all, stop producing content that sounds as if it was assembled by a committee of timid ghosts.
AI systems may be machines, but they are increasingly being used to answer human questions. Humans do not only need facts. They need judgement. They need context. They need someone to say, 'this depends', and then have the decency to explain what it depends on.
The uncomfortable conclusion
Google's Search I/O 2026 message is not 'SEO is dead'. It is not even close. If anything, it is the opposite. The sites that cannot be crawled, cannot be understood, cannot be trusted and cannot offer something beyond commodity repetition will have even less room to hide.
But the message is also not 'carry on as before'. That is the part the great unwashed of the industry will miss, because it is easier to fight imaginary deaths than understand real mutations.
Google tells us Search is still Search. Bing tells us the index is becoming answer support. Both are useful. Both are partial. Together, they show the shape of the thing.
The web is still the source. The page is still needed. SEO is still the foundation. But the interface has changed, the journey has changed, the measurement has changed, and the unit of value is no longer only the ranking URL.
The work is not to chase AI hacks. The work is to build the kind of web asset that can be read by people, trusted by systems, cited by answers and carried into action without losing its meaning.
So the work is not to chase AI hacks. The work is to build the kind of web asset that can be read by people, trusted by systems, cited by answers and carried into action without losing its meaning.
That is not a new trick.
It is the old discipline, stripped of its theatre.
