SEO, GEO and AEO: What NZ Marketers Actually Need to Know

Depending on who is selling something, you may hear that SEO is dead, GEO has replaced it, AEO is the future, or that your organisation urgently needs an entirely new optimisation strategy for ChatGPT. You may also be offered a score showing how visible your brand is in AI answers.

I would be cautious about all of it.

The terminology is still evolving and the boundaries are blurry. Google's own guide to its generative AI search features says that, from Google's perspective, optimising for generative AI search is still SEO.

But there is a real change underneath the terminology. People are increasingly asking AI systems for answers rather than asking search engines for websites. The organisation that is easiest to find is no longer necessarily the one that is most visible. More and more, it is the one that is easiest to understand and trust.

Most GEO tools measure the end of that process: whether an AI mentioned or cited you. The more valuable work sits further upstream, and some of it can now be observed.

A simple way to think about it

Rather than seeing SEO, GEO and AEO as competing disciplines, think of them as overlapping parts of a journey:

• SEO helps machines find you.
• GEO helps them understand you.
• AEO helps them answer with you.

The three overlap heavily. A technically sound website helps SEO, clear structure helps GEO, useful and authoritative answers help AEO, and good content helps all three.

So marketers should resist anyone suggesting these are separate activities requiring separate strategies. They aren't. But the distinction still helps, because it shows where things can go wrong.

If you would rather work through this on your own website than in the abstract, our AI workshop for marketing teams covers SEO, GEO and AEO on real campaign material.

SEO is still the foundation

SEO is not disappearing.

Google's guide says SEO best practice remains relevant because its generative AI features are rooted in its core Search ranking and quality systems. To be eligible for those features at all, a page must be indexed and eligible to appear in Google Search with a snippet.

So the familiar fundamentals still matter: crawlability, a technically healthy website, good internal linking, helpful original content, clear page structure, appropriate metadata, and authority and credibility.

The same principle applies beyond Google. OpenAI tells publishers that if they want their content to appear in ChatGPT search, they must not block its OAI-SearchBot crawler.

SEO remains part of the infrastructure. But discovery is no longer the whole problem.

GEO is about understanding

This is where I think the idea of Generative Engine Optimisation earns its place. I would define GEO as:

Making your organisation and its information accessible, understandable, trustworthy and usable by generative AI systems.

A traditional search engine might only need to decide whether a page is relevant to a search phrase. A generative system may need to work out what your organisation does, how a product works, who it suits, how it compares with an alternative, what your research says, whether a statement can be trusted, and which pieces of information across your website belong together.

That takes more than keyword relevance. The system may have to build its understanding from several pages, PDFs, structured data, external references and other sources. That makes GEO a broader problem than appearing in an AI citation. Our GEO guide covers the entity and structured-data side in depth.

AEO comes afterwards

Answer Engine Optimisation is about whether your information becomes part of an answer.

Google's AI Overviews are the easiest example. Someone asks a question and, instead of ten links, an AI system retrieves information, synthesises it and generates an answer. For your organisation to appear in that answer, your content has to give the system something to work with: a clearly stated fact, an authoritative explanation, original research, an eligibility criterion, a product specification, a fair comparison or a well-supported expert position. Our AEO guide covers how to structure those answers.

This is why I see GEO as a prerequisite for effective AEO with most AI assistants. An AI system cannot confidently answer with you if it cannot first understand you. Google is the partial exception: it has spent two decades building its index, so it often already knows who you are, and well-structured answers can win an AI Overview citation before the rest of your GEO work is done.

So the sequence is roughly discovery, then understanding, then answer. It isn't a neat linear funnel and the stages overlap considerably, but it is a useful mental model.

Why this matters more than the acronyms

Consider a New Zealand organisation with deep expertise. Its website has dozens of PDFs, several product pages, technical information, customer FAQs, research, some JavaScript-heavy interactive pages and years of institutional knowledge.

To someone who already understands the organisation, the website might work perfectly well. An AI system may see something quite different. The important information might sit mostly inside PDFs, with critical product details rendered by JavaScript. Several pages may describe the same service in different terms. The site may assume visitors already know the sector's jargon, and its most authoritative material may be buried three levels deep.

For that organisation, the question shifts from "Are we ranking?" to "Can an AI system build an accurate picture of us?" That second question is GEO. If the answer is no, AEO gets considerably harder.

Accessibility comes before understanding

This gives marketers a practical place to start. Before asking whether you appear in ChatGPT, ask whether AI systems can reliably reach the information you want them to understand.

Google's guide tells site owners to make sure their content is crawlable, because its generative AI models rely on publicly accessible, crawlable content. It also notes that SEO on a website built with JavaScript frameworks is generally more complex than on other kinds of site.

OpenAI's guidance on allowing its crawlers, written for advertisers, lists the usual blockers: robots.txt rules, firewall and CDN settings, bot protection services, CAPTCHAs and JavaScript challenges, and rate limiting.

This matters most for organisations that rely heavily on PDFs, JavaScript-rendered or embedded content, interactive tools, gated resources, complex navigation, bot protection or inconsistent site structures. The content can exist online and still be hard for machines to retrieve or interpret.

We can now observe some of this

So far, most GEO discussion has focused on output. Does ChatGPT mention us? Do we appear in Perplexity? Are we cited in Google's AI results? Those are fair questions, but they only show the end result.

We can now observe some of what happens further upstream. PostHog, for example, classifies crawlers, AI agents and other automated traffic in its web analytics. That lets an organisation look for patterns:

• Which pages are AI crawlers accessing?
• Which sections receive little AI crawler activity?
• Do different AI systems access different content?
• Does crawler behaviour change after we restructure or republish something?
• Are HTML pages accessed differently from JavaScript-heavy sections?

One catch: the main AI crawlers don't run JavaScript, so a standard analytics tag never sees them. You need your server or CDN logs feeding the analytics tool.

None of this tells you that an AI has understood a page, and it certainly does not prove a page has been cited. But it does tell you something real about AI accessibility.

Think of GEO as a chain

For marketers, I think this is the most useful model:

Accessible → discoverable → understandable → retrievable → trusted → used → cited → visited or acted upon.

Most GEO tools focus on the last few links. They tell you that ChatGPT mentioned your brand, or that you appeared in a particular set of prompts. That information has value, but if you want to improve performance you need to understand what happens further upstream. A citation problem may turn out to be an accessibility problem, an information architecture problem, an authority problem or a content problem.

Be careful with AI visibility scores

A growing number of GEO and AEO tools run large numbers of prompts and produce a visibility score for a brand. That can be useful, if you understand what is being measured.

Many of these tools use an AI provider's API to run repeatable tests at scale. The API gives them a controlled environment, so the same prompt can be sent repeatedly and the results compared over time.

But an API request is not necessarily the same as a customer using ChatGPT, Gemini, Claude or another AI product. A live session may carry far more context: the user's location or market, earlier questions in the conversation, personalisation, account state, live web search, different search or browsing tools, different model routing and other signals supplied by the product.

Consider a New Zealand customer asking, "What are some good providers for this?" They may already have told the AI they live in Christchurch, run a small business and have a particular budget. Their answer may differ materially from the one a standalone API prompt produces with none of that context.

API-based testing is synthetic testing. A visibility score derived from thousands of API calls can still be a repeatable benchmark. It should not be read as "this is what New Zealand customers see when they use AI". Treat it as one part of a broader measurement system. We cover one such benchmark in Share of Model.

Measure from several angles

No single metric can currently tell you whether your GEO strategy is working. Each measure answers a different question.

Crawler data tells us something about access. Synthetic testing tells us something about repeatable visibility. Real sessions tell us something about customer experience. Referrals and conversions tell us whether it matters. No single GEO score does all of those things.

A practical GEO audit for marketers

You do not need an expensive platform to start asking useful questions.

One experiment to run

Choose ten strategically important pages. Record a baseline for crawler activity, synthetic AI visibility, real-world AI visibility, citations and referral traffic.

Then improve the pages. Make the information clearer. Bring information out of PDFs where it makes sense, improve internal linking, reduce ambiguity, add credible sources and evidence, and make important information available without depending on JavaScript.

Then watch what changes. You are unlikely to prove causation immediately, but over time you can start to see patterns. That is far more valuable than blindly following someone's "15 GEO hacks".

Don't optimise for AI at the expense of people

Do not rewrite every webpage into robotic question-and-answer blocks, and do not create hundreds of near-identical pages to capture AI query variations. Google's guide says that creating separate content for every query variation, mainly to manipulate rankings or AI responses, breaches its scaled content abuse spam policy.

Good GEO should make information more understandable to people as well: clearer structure, better explanations, explicit terminology, better accessibility, stronger evidence, less jargon and more useful content.

These are not particularly exotic ideas. AI gives us another reason to do them properly.

Where AI agents come in

AI systems are starting to act on behalf of users as well as answer their questions. An agent might need to investigate products, compare providers, find eligibility information, navigate a website, complete a process, make a booking or interact with a service.

At that point your website is no longer communicating only with a person behind a browser. It may also be communicating with software acting for that person.

On that trajectory, GEO becomes the work of preparing your organisation's information for a machine-mediated web. Today's ChatGPT citation is one small part of it.

So where should NZ marketers start?

Probably not by buying another platform.

There is no need to panic, and little value in chasing every new acronym. But doing nothing because the terminology is still messy would be a mistake. The behaviour underneath it is real.

Start with the list of things AI must understand about your organisation, and check that machines can reach them. Then pick your ten pages, record the baseline and see what changes.