What Google’s AI search guidance changes for marketing education

On Friday, Google published official guidance on how to optimise for the generative AI features inside Google Search. The most-quoted sentence is also the most useful: "Optimising for generative AI search is optimising for the search experience, and thus still SEO." The argument that follows is that AEO (answer engine optimisation) and GEO (generative engine optimisation) are not new disciplines, that several proposed "AI optimisation hacks" are nonsense, and that the core practice of SEO continues to be the right frame.

There is real substance in that position. There is also a piece of the picture missing.

For a marketing professor planning the autumn syllabus, the document is useful but incomplete. It tells you what to keep teaching. It does not tell you what to add. And it limits itself, deliberately, to Google’s own surfaces, which is a fair editorial choice for Google but a partial picture of the world your students will be working in.

What Google gets right

The fundamentals of SEO still apply to AI search. Retrieval-augmented generation, the mechanism behind AI Overviews, depends on Google’s index. If you do the foundational work (credible authors, original perspective, technically clean pages, real expertise) you end up well-positioned for citation inside AI-generated answers. The publishers who were doing EEAT properly are now the publishers being quoted inside AI Overviews. The publishers chasing keyword density and content velocity are having a bad year. This is not new ground. It is Google saying out loud that the discipline they have been pushing for a decade was always pointing at this.

Most of the "GEO hacks" circulating on LinkedIn are unsupported. llms.txt files were a community proposal that never had standards-body backing, and there is no public evidence they change citation behaviour. Schema markup is good SEO practice, and Google has now confirmed it is not a magic key to AI citation. Wholesale rewriting of pages "for AI" is, in most cases, optimising for a problem you cannot prove exists.

The "scaled content abuse" line in Google’s document is the one that should travel furthest. The Xero earnings-call narrative from last week (60 SEO pieces a quarter scaled to 50 a day with AI) has now been formally placed inside Google’s spam policy territory. The teaching question writes itself. What is the ethical and strategic position of a marketing team adopting that volume play?

If you were a really good SEO practitioner, and most were not, you were already doing much of what AI search now rewards. That is what Google is saying, and that is the part to keep.

What Google does not say

Two things missing from the document, and both of them matter more than the things the document affirms.

The behaviour on the user’s side of the search box has changed radically. A user asking Google five years ago typed "flights Brazil November." A user asking ChatGPT in 2026 types a paragraph. "I’m planning a beach trip to Brazil in November with two kids under ten, we have ten days, what’s the right combination of city and beach we should fly into?" The intent is unpacked. The constraints are stated. The history is included. The user expects synthesis, not a list of blue links, and the LLM does work that used to belong to the user: research, comparison, recommendation. The agent is doing the searching on the customer’s behalf, and it is making brand recommendations the user often acts on without seeing the underlying pages at all.

This is the part Google’s guidance treats as a footnote and your students will treat as the whole game. The question that follows the behavioural shift is not "how do I rank for [keyword]." It is "how does the agent doing the research on my customer’s behalf decide which brands to recommend." That question overlaps with SEO; it also includes things classical SEO often did not. Brand entity presence in trusted sources. Citation density across the open web. Conversational query coverage in long-form content. Agent-readable trust signals that are visible to systems beyond Google.

Google’s guidance is Google’s. It is silent, by design, on ChatGPT, Claude, and Perplexity. This might be an obvious point, because is hardly likely to give guidance on its rival platforms. However, we need to acclimate to a new world where "Google" and "SEO" are not synonymous.

Those systems are not retrieving solely from Google’s index. Each one constructs its own picture of "authority" from corpora and signals Google does not control. The differences are observable: run the same query through each of them and the citations, the framings, and the brands surfaced come back different every time. For a marketing student asking "how does a company show up when a customer asks Claude for a recommendation," Google’s document is not the answer to the question.

Put the two together and the honest position is one most senior SEO practitioners would already endorse. SEO is still the right umbrella term. The practice has broadened. The things really good SEOs should always have been doing (entity strategy, brand perception, citation work, deep audience understanding) are now baseline rather than advanced. And several things genuinely new (multi-engine optimisation, conversational query coverage, content architecture that supports retrieval without falling into the chunking trap) have to sit inside the umbrella too.

Calling all of that "GEO" overclaims the novelty. Calling it "just SEO" underclaims the shift. The accurate description is that SEO has had to grow.

What this changes for the autumn syllabus

The practical move for a marketing professor is to widen the SEO unit rather than rename it.

Keep. The fundamentals. How search works, indexing, ranking signals, EEAT, keyword research as a foundational exercise. These are not obsolete. Students still need them, partly because LLMs are still grounded in them.

Add. Five things, in order of how teachable they are in a single semester.

1. The new mechanisms Google did name. Retrieval-augmented generation and query fan-out are concrete, teachable, and most syllabi do not yet cover them.

2. The behavioural shift in how people search. The "flights to Brazil" query versus the paragraph query, and what the agent does in between. The exercise that works in class: have students run the same paragraph-shaped intent through Google AI Mode, ChatGPT, Claude, and Perplexity, and compare what each returns and why.

3. The four-engine landscape. Google AI Overviews, ChatGPT, Claude, Perplexity. Different corpora, different citation patterns, different commercial logic. Students need this for the same reason they once needed to know the difference between Google and Bing: the channels do not behave the same way.

4. Brand perception as a search variable. A brand’s reputation in trusted third-party sources increasingly determines its visibility in AI recommendations. This is brand strategy plus PR plus SEO blurring into one discipline. The classical SEO course did not teach this; the 2026 one has to.

5. The ethics line. The "scaled content abuse" position Google has now formalised, framed in class as a discussion about AI-assisted content production and the limits of the productivity narrative.

That widened unit can still be called SEO. The students who graduate from it will be better prepared with AI skills than the students who graduated from a course that taught "GEO" as a separate discipline, and they will be much better prepared than the students whose course did not change at all.

What this means for what we are building at Novela

The AI Search simulation Novela is building is not an attempt to teach a new discipline. It is an attempt to teach the broader version of SEO that the practice has had to become. Foundations of search, the new mechanisms, the behavioural shift, the multi-engine landscape, the brand perception layer, and the ethics of AI-assisted production, all inside one environment where students run campaigns and watch the consequences.

Google’s guidance is the official position from one player in a four-engine world. It is correct about the foundations, narrow about the surface, and worth respecting where it speaks and pressing where it is silent.

The marketing professor of 2026 needs a way to teach both the part Google describes and the part Google does not. That is what Novela has been building. Get in touch for early demo access.

Glossary

Reference terms used in this piece, in alphabetical order. Useful for class handouts or for orienting students before the broader discussion.

AEO: Answer Engine Optimisation

A term for optimising content to be cited inside AI-generated answers. Google’s recent guidance pushes back on AEO as a separate discipline. Useful as a label for the practices that have been added to the SEO playbook rather than as a standalone field.

Agentic search

A pattern where an AI system performs multi-step research on behalf of the user, rather than returning a list of pages for the user to evaluate. Follows links, compares sources, drafts summaries, makes recommendations. Increasingly, the customer’s first interaction with a category happens inside an agent’s research process, not on a search results page.

AI Mode

Google’s separate, conversational AI search experience inside Search, distinct from AI Overviews. Allows follow-up questions and longer interactions rather than the one-shot summary of an Overview.

AI Overviews

The AI-generated summary Google shows at the top of many search results pages. Built on Google’s existing search index plus a generative model that summarises retrieved content. The link in the upper-right corner of an Overview lists the sources the model cited.

AI Search

The umbrella term for using generative AI to surface and synthesise answers to user queries, replacing or sitting on top of the traditional blue-link results page. Covers Google’s AI Overviews and AI Mode plus the standalone chat products from ChatGPT, Claude, Perplexity, and others.

AI Skills (for marketing)

The capabilities a marketing professional needs to work effectively with generative AI tools. Includes prompting, AI workflow design, output evaluation, and judgement about when AI is and is not the right tool. Prompting in particular has commoditised quickly and is now baseline literacy rather than a competitive edge.

Brand entity / Entity SEO

A page-independent concept of "what a brand is" that search engines and LLMs construct from mentions, citations, and structured information across the web. A brand can be recommended by an AI even when its own pages do not appear, because the system has built an internal model of the brand from external sources. Increasingly important in AI search, where the recommendation often happens before the user sees any specific page.

Chunking

The practice of breaking content into small fragments specifically intended to be easier for AI systems to parse and cite. Mostly unnecessary in 2026: Google has explicitly said the AI systems can handle nuance across multiple topics on a page. Writing for human readers in clear sections is sufficient.

Conversational query

A query expressed in natural-language paragraph form, with context, constraints, and intent included. Replaces the keyword-string queries of the previous decade ("flights Brazil" becomes "I’m planning a beach trip to Brazil in November with two kids under ten, where should we fly into"). Content that covers full topic clusters rather than single keywords is more likely to be surfaced.

EEAT: Experience, Expertise, Authoritativeness, Trustworthiness

Google’s framework for evaluating content quality, formalised around 2014 and expanded in 2022 with the extra E for Experience. Not a direct ranking signal; the meta-framework Google’s quality raters use to assess pages. Sites that genuinely achieve EEAT (named expert authors, original perspective, verifiable expertise) are over-represented in AI Overview citations.

GEO: Generative Engine Optimisation

The term most often used in 2025 and 2026 industry discussion to describe optimisation for generative AI search experiences. Closely related to AEO. Google argues GEO is not a separate discipline; senior SEO practitioners view it as a useful label for the additional practices the SEO playbook has had to absorb.

Index / Indexing

The process by which a search engine crawls a web page, processes its content, and adds it to a database for retrieval. A page must be indexed to be eligible for any kind of search visibility, including in AI-generated answers. Google’s AI Overviews and AI Mode both depend on the existing search index.

llms.txt

A community-proposed file format intended to give large language models structured instructions about how to use a website’s content. Has no formal standards-body backing and is not implemented by major AI search providers. Google has now formally said llms.txt is not required for AI search visibility.

LLM: Large Language Model

A class of AI system trained on large volumes of text, capable of generating fluent responses to a prompt. The underlying technology behind ChatGPT, Claude, Gemini, and the generative features of AI search. The model is trained on a fixed corpus at training time; for current information it relies on retrieval (see RAG) or live search integrations.

Multi-engine optimisation

The practice of optimising for visibility across multiple AI search systems (Google’s AI Overviews, ChatGPT, Claude, Perplexity) rather than only one. Each system has different retrieval logic, different corpus weights, and different commercial dynamics. The skills overlap with SEO but the surfaces are different.

Query fan-out

The technique by which a single user query is decomposed by an AI system into a cluster of related sub-queries, run in parallel to gather wider context. "How do I fix a weedy lawn" might fan out into "best herbicides," "removing weeds without chemicals," and "preventing weeds in lawn." Pages that cover a topic cluster rather than a single phrase are more likely to be picked up by fan-out retrieval.

RAG / Grounding: Retrieval-augmented generation

The technical mechanism by which a generative AI system retrieves relevant pages from a search index before generating an answer. Google uses the term "grounding" for the same process inside its AI features. Pages that rank well for the underlying query are more likely to be retrieved and cited.

Scaled content abuse

Google’s policy framing for content produced at high volume, often using AI, with the primary intention of gaming search rankings rather than serving users. Google’s recent guidance has explicitly placed high-volume AI content production inside this spam policy. A useful classroom discussion topic for the ethics of AI-assisted marketing.

Schema markup / Structured data

Code added to a web page (in JSON-LD or microdata format) that describes the page’s content in a machine-readable way. Useful for rich results in conventional search. Google has now confirmed it is not a magic key to AI search citation. Worth doing for the SEO benefits it always had; not a GEO unlock.

SEO: Search Engine Optimisation

The discipline of building content, structure, and signals on a website such that search engines surface it to the right users. Has broadened over the last two years to cover citation inside AI-generated answers as well as ranking in conventional search results. Still the right umbrella term for the practice as a whole.