There is a very old habit in search, and like most old habits, it does not look dangerous until the world changes around it. A platform publishes guidance, the industry reads it, and within a few days the guidance becomes doctrine. Not because it is complete. Not because it is universal. Simply because it came from the platform with the loudest gravitational pull.
That is the mistake to avoid with Google’s guidance on generative AI features in Search. The document is useful. It is worth reading. It contains sensible advice about crawlability, technical structure, non-commodity content, page experience, structured data and the continuing value of foundational SEO. But it is not a complete map of AI visibility.
The most important phrase in the whole thing is easy to walk past:
“From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
Google Search Central [1]
That phrase, from Google Search’s perspective, is the hinge. Google is not describing every AI search, answer, assistant, retrieval, citation or agentic system. Google is describing Google. Fair enough. The trouble starts when the industry takes a scoped statement and turns it into a universal law, because that is what the industry does, it finds a neat phrase, wraps a deck around it, and calls the thing 'strategy'.
SEO is still the foundation
Let us give Google its due, because the argument falls apart if we pretend the fundamentals have vanished. Google says its generative AI features are rooted in its core Search ranking and quality systems, and it names retrieval-augmented generation, also called grounding, and query fan-out as part of how those features work.[1] In that context, traditional SEO work still matters. If a page cannot be crawled, rendered, indexed, understood or trusted, there is no clever AI wrapper that magically turns it into a useful source.
Google’s advice to create helpful, reliable, people-first content also remains right. Its own guidance asks whether content provides original information, reporting, research or analysis, whether it offers something beyond the obvious, whether it is clearly sourced, and whether there is evidence of the expertise behind it.[2] That is not old-fashioned. That is still the bones of the thing.
| Foundational practice | Why it still matters in Google AI Search |
|---|---|
| Crawlability and indexability | Google’s AI features draw from content available through Google Search systems.[1] |
| Helpful, reliable content | Google’s ranking systems are designed to prioritise content made to benefit people, not manipulate rankings.[2] |
| Clear authorship and sourcing | Google’s helpful content guidance asks whether users can see who created the content and why it should be trusted.[2] |
| Structured data | Google says structured data gives explicit clues about the meaning of a page, although it should describe visible and accurate page content.[3] |
| Page experience | Google’s guidance continues to connect useful content with an overall good page experience.[2] |
So no, this is not a funeral for SEO. Anyone telling clients that SEO no longer matters is taking the mickey, usually while trying to sell them the same thing under a shinier acronym. The foundation is still there. The question is whether the foundation is now the whole house.
It is not.
The mistake is not listening to Google. The mistake is allowing Google’s view of Google to become your entire view of the web.
The page is no longer the only unit of value
Classic SEO trained us to think in pages. A query maps to a result. A result maps to a URL. A URL earns impressions, clicks, links, conversions and, if we are lucky, a little applause from a dashboard that pretends to know more than it does.
AI retrieval changes the shape of the problem. The page still matters, but it is no longer the only unit being judged. AI systems need extractable, supportable, attributable information. They need claims that survive being retrieved, transformed, compared, summarised and cited. That is a different pressure on content, and pretending otherwise is not pragmatism, it is just the comfort of an old map.
Microsoft’s explanation of grounding makes the distinction unusually clear. It says traditional search asks which pages a user should visit, while grounding asks what information an AI system can responsibly use to construct an answer.[4] That is not a small wording change. It moves the centre of gravity from the document to the evidence inside the document.
| Traditional search | Grounded AI answer systems |
|---|---|
| The user scans options. | The system commits to an answer. |
| The page is the main unit of value. | The claim, passage, source and entity become units of value. |
| Imperfect ranking can be corrected by the user. | Bad evidence can become a confident wrong answer. |
| Clicks are a primary success signal. | Citations, mentions, evidence use and answer inclusion become important. |
| Content competes to be visited. | Content competes to be used. |
Microsoft goes further by saying the unit of value shifts from documents to groundable information, meaning discrete, supportable facts with clear provenance.[4] That phrase should make every serious SEO stop for a moment, because it names the work more accurately than most of the industry’s buzzword soup.
The practical consequence is simple. A page can be well optimised for search and still be poorly prepared for grounded AI use. It can rank, but fail to provide cleanly extractable evidence. It can be comprehensive, but too muddy. It can be authoritative in tone, but thin on provenance. It can look fine to a human reader, but fragment poorly when retrieved in passages.
Google is not wrong about chunking, it is scoped
Google says there is no requirement to break content into tiny pieces for AI to understand it, and that Google systems can understand nuance across multiple topics on a page.[1] For Google Search, that may be a perfectly reasonable instruction. It is also a useful warning against the sort of mechanical nonsense where people turn every paragraph into a lonely little island because someone on LinkedIn found a new word to overuse.
But there is a better way to say it. Do not write in fragments. Do not vandalise good pages into atomised sludge. But do write passages that preserve meaning when separated from their neighbours.
That is the distinction. The goal is not tiny content. The goal is coherent evidence.
Microsoft’s grounding guidance says that chunking and transformations must preserve the meaning and claims used in an answer.[4] That is the more useful operational lens. If an AI system retrieves a section of your page, that section should not depend on five vague paragraphs before it to make sense. It should contain the entity, the claim, the context and, where appropriate, the evidence.
This is not writing for robots instead of people. It is writing with enough discipline that both people and machines can follow the thread. There is nothing noble about ambiguity when the task is explanation.
The new visitor may not be human
The next uncomfortable truth is that websites are no longer visited only by human beings with eyes, thumbs, patience and an inexplicable willingness to close cookie banners. Some visits will be delegated. Some journeys will be performed by agents acting on behalf of users.
Google’s own web.dev guidance says websites have a new type of visitor: AI agents that can interpret input, plan and execute actions on behalf of a user.[5] Those agents do not experience a site as a human does. They may use screenshots, raw HTML and the accessibility tree to understand what a page contains and what actions are available.[5]
That changes the meaning of a usable website. A beautiful interface that hides important actions behind vague divs, shifting layouts, ghost elements or visual tricks may be fine for a designer’s presentation deck, but a poor map for an agent. Semantic HTML, labelled inputs, stable layouts and accessible interactive elements are not merely compliance theatre. They are machine-readable affordances.
| Agent-readiness factor | Practical implication |
|---|---|
| Semantic HTML | Use real buttons, links, headings, labels and form elements wherever possible.[5] |
| Stable layouts | Avoid interfaces where important actions move unpredictably between states.[5] |
| Accessibility tree quality | Treat the accessibility tree as a functional map, not an afterthought.[5] |
| Visible action areas | Make critical interactive elements visible, named and large enough to identify.[5] |
| Clear data relationships | Ensure product, service, price, location, author and evidence relationships are explicit. |
This is where the old SEO department starts to look too small for the job. Technical SEO overlaps with this work, yes. Accessibility overlaps with it. UX overlaps with it. Development, product, content, analytics and brand all overlap with it. That is exactly the point. AI visibility is not a new label slapped on the old checklist. It is an operating layer across the digital ecosystem.
Evidence is becoming the competitive layer
The phrase E-E-A-T has been abused badly enough to deserve a quiet room and a week off. Too often it becomes a checklist of author boxes, stock credentials and schema markup, as if trust were a decorative element you could staple to a page after the copy is written.
Google’s own helpful content guidance is more serious than that. It asks whether content presents information in a way that makes people want to trust it, such as clear sourcing, evidence of expertise, background about the author or site, and easily verified factual accuracy.[2] It also says trust is the most important member of the E-E-A-T family.[2]
In AI visibility, trust becomes even more interesting because the system may be triangulating across sources. It may not take your website’s word for who you are. It may compare your claims with third-party references, structured data, reviews, forums, profiles, publications, citations, social proof, knowledge bases and whatever else the retrieval system can access.
That means the work moves beyond the page. Your owned site still matters, but the entity has to make sense in the wider web. A clinic, consultant, SaaS company or professional service business cannot simply declare authority. It has to be corroborated. The web has to agree, or at least provide enough consistent evidence for systems to treat the entity as real, relevant and reliable.
This is where SEO becomes less like decorating pages and more like building an evidence ecosystem. Not a little badge of trust. Not a manufactured author box. An ecosystem, the messy, living sort of thing that either exists or does not.
A better framework: five layers of AI visibility
The sensible position is neither “AI search is just SEO” nor “SEO is dead”. Both are lazy in their own way. The better frame is this: SEO remains the foundation, but AI visibility adds new layers of retrieval, evidence, entity and agent readiness.
| Layer | The question it answers | Typical work |
|---|---|---|
| Google AI Search readiness | Can Google crawl, index, understand and confidently use this content in its Search experiences? | Technical SEO, helpful content, non-commodity insight, Search Console hygiene, structured data where appropriate. |
| Retrieval readiness | Can an AI system extract useful passages without losing the meaning? | Clear sections, self-contained explanations, explicit entities, concise definitions, stable terminology. |
| Evidence readiness | Can the claim be checked, trusted and cited? | Sources, dates, author expertise, methodology notes, proof, comparison tables, original data. |
| Entity readiness | Does the wider web corroborate who you are and what you are known for? | Third-party mentions, professional profiles, reviews, associations, publications, consistent sameAs references. |
| Agent readiness | Can delegated AI systems navigate and act on the site reliably? | Semantic HTML, accessibility tree quality, labelled inputs, stable layouts, visible actions. |
This framework is more useful than arguing over acronyms. Call it SEO, GEO, AEO, AI visibility or, if you are feeling particularly sedulous, ‘retrieval-era digital architecture’. The name is not the work. The work is making useful information discoverable, extractable, verifiable and usable across systems that no longer behave like a blue-link results page.
What to do now
The practical response is not to panic and rebuild every page as if a machine were the only reader. That would be another form of stupidity wearing a clever hat. The right response is to audit content and architecture against the new unit of value.
Start with your important pages. Ask whether each page contains original information or only polished common knowledge. Ask whether its sections can stand alone when retrieved. Ask whether its claims are sourced. Ask whether the author, organisation, product, location and service entities are clear. Ask whether a third party can verify what the page says. Ask whether an agent can understand the page’s structure and complete the intended journey.
Then look beyond the website. If the business depends on trust, and most real businesses do, ask where that trust is evidenced outside the site. Professional bodies, reviews, case studies, talks, publications, partnerships, directory profiles, local citations, interviews, data sources and community discussions all form part of the entity record. Some of that work used to sit awkwardly outside SEO. It no longer does.
The map has split
Google’s guidance is useful, but it is not the whole map. It tells us how to think about Google’s generative AI features in Google Search. It does not settle the wider question of how information is retrieved, grounded, cited, trusted and acted upon across the AI ecosystem.
The old map was built around rankings and pages. The new map still includes those things, but it adds passages, claims, entities, agents, citations and evidence. That is not a minor expansion. It is a change in the unit of competition.
So yes, keep doing SEO properly. Fix the crawl traps. Improve the internal links. Write useful content. Add structured data where it describes visible reality. Make pages fast, accessible and comprehensible. But do not confuse the foundation with the building.
The next advantage will not belong to the businesses that shout ‘we do SEO’ the loudest. It will belong to the ones whose knowledge can be found, trusted, extracted and used when the user is no longer the one doing the searching.
SEO is still the foundation. Evidence is becoming the architecture.
References
Google Search Central: Optimising your website for generative AI features on Google Search
Google Search Central: Creating helpful, reliable, people-first content
Google Search Central: Introduction to structured data markup in Google Search
Bing: Evolving role of the index, from ranking pages to supporting answers
