SEO is dead. Learn AEO and GEO.

SEO is not dead. But the version of SEO that many companies are still buying is too small for the way people search now. If the entire plan is “pick a keyword, rank a page, get a click,” the plan stops before the customer journey does. People are increasingly asking complete questions, comparing options inside generated answers, following up conversationally and sometimes making decisions before they ever visit a website.

That is why I like the provocative title. “SEO is dead” gets attention, but the more useful truth is this: SEO grew up. Search visibility now includes classic rankings, answer boxes, AI Overviews, AI Mode, ChatGPT search, Copilot, Perplexity, multimodal search and the first wave of agents that can research and act. The job is no longer only to rank. The job is to become a reliable part of the answer.

The terminology is still messy. AEO, GEO, AI SEO, LLM optimization and AI search optimization are often used differently by different people. I do not think the label matters as much as understanding what changed and building for it without abandoning the fundamentals that still work.

Is SEO actually dead?

No. SEO is still the foundation of AI search visibility. Google says this directly: its generative AI features, including AI Overviews and AI Mode, are rooted in core Search ranking and quality systems. Pages still need to be crawlable, indexable, eligible to appear in Search and useful enough to retrieve.

Google now even has a dedicated guide titled Optimizing your website for generative AI features on Google Search. Its first answer to the question “Is SEO still relevant for generative AI search?” is yes. The important change is that SEO is no longer the whole visibility model.

The old mental model was linear: query → ranked result → click → website. The new model can be: question → multiple retrievals → synthesized answer → citations → follow-up question → comparison → brand shortlist → click, direct visit or action. A site can influence that journey without receiving the first click, and a brand can lose the decision before the click even happens.

SEO is not dead. Ranking as the only definition of search visibility is.

What are SEO, AEO and GEO?

There is no universal standards body that defines AEO and GEO, and the terms overlap. Google treats work aimed at its generative Search experiences as part of SEO. Microsoft openly uses GEO language for AI search visibility. Industry tools use AEO and GEO in slightly different ways. So the cleanest way to use the terms is as working layers, not competing religions.

My practical definition is simple: SEO gets you into the searchable web. AEO helps make you answerable. GEO helps make you usable inside a generated answer. In real life, the three overlap heavily. A strong page should not need three separate versions.

How does AI search actually find and use content?

Most AI search experiences do not simply “read the internet” in one step. They retrieve information, evaluate it, synthesize it and often show supporting sources. Understanding that flow is more useful than chasing hacks.

1. Discovery and indexing

The system first needs access to useful content. For Google, eligibility for generative Search features depends on being indexed and eligible to show with a snippet. For ChatGPT Search, OpenAI says publishers should allow OAI-SearchBot if they want their content to be discoverable and surfaced in search results. Bing AI visibility also depends on content that is eligible for indexing.

2. Retrieval and grounding

When the user asks something, the system looks for relevant evidence. Google describes retrieval-augmented generation, or grounding, as using its Search systems to retrieve current web pages that can support an answer. Microsoft describes grounding as the layer connecting generative systems to verifiable web information.

3. Query fan-out

The system may search for more than the words the user typed. Google describes query fan-out as generating multiple related searches in parallel to collect the information needed for a fuller answer. A question about entering a new market, for example, can trigger retrieval around pricing, regulation, competitors, local partnerships and customer behavior even if the original query did not list each one.

4. Synthesis, citation and recommendation

The answer is assembled from the retrieved evidence. A page can be cited for one useful passage, a brand can be mentioned because multiple sources support the same positioning, or a product can appear in a comparison because the system has enough reliable information to understand what it is and who it is for.

5. Follow-up and action

Search is becoming conversational and increasingly agentic. The user can refine the question, compare options and ask the system to act. That means websites are not only competing to be read by people. They increasingly need to be understandable to systems that retrieve, compare and interact with information on the user’s behalf.

What changed from ranking to answer visibility?

The biggest change is that visibility now has more stages. A page can be technically visible but not useful enough to retrieve. It can be retrieved but too vague to quote. It can be quoted but not strong enough to influence a recommendation. Thinking in stages makes the work much easier to diagnose.

How do you do SEO, AEO and GEO together?

Do not build three separate programs. Build one search and discovery system with a stronger standard for clarity, evidence and accessibility. The sequence below is the framework I would use with a real business.

1. Make the site crawlable before you make it clever

Start with the boring work because AI cannot cite what it cannot reliably reach. Audit robots.txt, noindex directives, canonical tags, redirects, 404s, XML sitemaps, internal links, JavaScript rendering and CDN or bot-protection rules. Make sure your important pages return clean 200 responses and are linked from places crawlers can find.

For Google, the same crawl and index requirements that support Search also support generative Search. For ChatGPT Search, OpenAI separates search discovery from model training: OAI-SearchBot controls search visibility, while GPTBot is a separate control for training. A publisher can allow search crawling while disallowing training.

Do not start GEO with prompts. Start with crawlability, indexing and a site that can be understood.

2. Separate search visibility from AI training controls

This distinction matters because companies sometimes block every AI-labelled crawler without realizing they are also blocking search visibility. Decide what you want, then configure it intentionally. OpenAI documents OAI-SearchBot and GPTBot as independent controls. If you want to be eligible for ChatGPT Search but not foundation-model training, those policies can be different.

Also check the infrastructure layer. A robots.txt file can allow a crawler while Cloudflare, Akamai or another bot-management system still returns 403. OpenAI explicitly recommends verifying crawler access through hosting and security layers, not only robots.txt.

3. Build a clean technical SEO foundation

Technical SEO is still the retrieval infrastructure. Keep canonical URLs stable, submit accurate XML sitemaps, reduce duplicate content, fix redirect chains, maintain good page experience and make sure important content is visible in rendered HTML. If a page depends heavily on JavaScript, test what search engines actually render instead of assuming the browser view is enough.

For fast-changing sites, freshness also matters operationally. Bing recommends IndexNow to notify participating engines when content is added, updated or removed. That does not guarantee citation, but it reduces the chance that retrieval systems are working from an old version of the page.

4. Make the entity obvious

AI systems need to understand who you are, what you do, what you are known for and how the pieces connect. Keep company name, founder name, service names, locations, product names, descriptions and key facts consistent across the website and credible external profiles.

Use structured data where it accurately describes visible content. Organization markup can help Google disambiguate a company, while Article or BlogPosting markup can connect an article to its author and publication details. But there is no special “AI schema” that unlocks generative answers. Google explicitly says structured data is not required for generative AI search and warns against overfocusing on it.

5. Build topics, not hundreds of keyword pages

A generative system can fan out a question into related retrievals, so shallow pages created for every small keyword variation are a weak long-term strategy. Build a strong topic architecture instead: one clear primary page, supporting articles that answer meaningful subquestions, and internal links that show the relationship between them.

This is where traditional content strategy becomes more important, not less. The goal is not to publish every possible question. It is to own the subject well enough that the site contains useful answers across the decisions a customer actually makes.

6. Write the answer early

For important questions, give the direct answer near the start of the section, then explain it. This is good for humans who scan, and Bing explicitly recommends clear headings, concise sections, tables and FAQ-style content because they make information easier to understand and reference in AI answers.

This is not a license to write robotic one-sentence fragments or to “chunk” every page into tiny pieces. Google says there is no requirement to chunk content for generative AI. The right standard is readability: each section should make sense on its own, while the whole article still reads naturally.

7. Add information that cannot be cheaply copied

Commodity content is the easiest content for an AI system to summarize without needing you. If twenty sites say the same thing, your version has very little reason to become the source. Original research, first-hand experience, strong examples, named frameworks, expert judgment, screenshots, tested processes and proprietary data create information gain.

Google’s 2026 guidance is unusually direct here: it recommends unique, non-commodity content and specifically says not to recycle what others have already said or what a generative model could easily produce. That is the content strategy shift I think matters most.

8. Make claims easy to verify

If a statement matters, support it. Link to the original research, show the date, identify the publisher, explain the methodology when it changes interpretation, and distinguish your opinion from external evidence. This makes the article better for a person and easier for a retrieval system to ground.

The original academic paper that formalized Generative Engine Optimization reported that its optimization methods could improve visibility by up to 40% in its benchmark, but also found that performance varied by domain. I would treat that as evidence that presentation and sourcing can matter, not as a promise that any checklist will produce a 40% gain on a live site.

9. Use formats that reduce ambiguity

Different questions deserve different formats. Use a comparison table when the user is comparing. Use numbered steps for a process. Use definitions where terminology is confusing. Use a diagram when the relationship matters more than the prose. Use FAQs for real recurring questions, not to manufacture extra keywords.

The goal is not to format for a machine. The goal is to remove the places where both the reader and the machine have to guess what you mean.

10. Treat images and video as search assets

Search is increasingly multimodal. Google says its generative Search features can surface relevant images and video, and in September 2026 Search Console added reporting for web multimodal searches such as Lens, Circle to Search and image-upload queries. That makes visual content part of discovery, not decoration.

Use original diagrams, screenshots, product images, comparison graphics and video when they explain something better than text. Give images descriptive alt text, useful surrounding copy and stable file/URL handling. If the visual contains important information, repeat the essential meaning in HTML text so the page is still understandable without the image.

11. Build off-site evidence, not fake mentions

Generative answers can pull from multiple sources, so what the rest of the web says about the brand matters. That does not mean manufacturing mentions. Google explicitly warns that seeking inauthentic mentions is not a useful strategy. The durable version is old-fashioned credibility: relevant press, industry references, expert contributions, reviews, partnerships, directories where appropriate and accurate profiles.

A brand claim is stronger when your own site says it clearly and credible third parties independently support it. This is one reason GEO is not only a content-team job. PR, reputation, partnerships, product quality and brand consistency all feed the same information environment.

12. Keep facts fresh and consistent

AI systems are especially sensitive to stale facts because generated answers often combine information from multiple places. Update dates when the content genuinely changes, remove outdated claims, maintain service/product details, and keep author and organization information consistent. Freshness is not changing the date on an unchanged article. It is maintaining the truth of the page.

How should an AEO/GEO article actually be written?

Use a structure that answers the reader before it asks the reader to work. The formula we use for CEO Corner is also a strong search structure because it balances direct answers with evidence and original thinking.

Question or topic → direct answer → explanation → evidence → example → original insight → actionable advice

• Start with the point. Open with the answer or the tension, not 500 words of history.
• Use one clear H1. Make the page about one primary subject, even if it covers many related questions.
• Use descriptive H2s. Question-style headings work when people genuinely ask the question; do not force every heading into a question.
• Answer first in important sections. Give the direct answer in the first sentence, then show why it is true.
• Add evidence where it changes trust. Link to primary sources, current official documentation and original research.
• Add something only you can say. First-hand experience, a framework, a process, a data point or an informed opinion gives the page a reason to exist.
• Use tables and lists when they simplify decisions. Do not use them just because someone said AI likes tables.
• End with action. Tell the reader what to check, change or measure next.

What should the development team implement?

Content work cannot compensate for a site that blocks the systems expected to find it. The technical implementation should be explicit and documented.

Example OpenAI policy decision: a publisher that wants ChatGPT Search visibility but does not want its crawled pages used for foundation-model training can allow OAI-SearchBot and disallow GPTBot. Always validate the latest official crawler documentation and published IP ranges before implementation.

Do you need llms.txt?

Not for Google Search, and it should not replace robots.txt, XML sitemaps, canonicalization or normal crawl/index controls. Google says it does not use llms.txt for Search, including generative AI features, and that adding one neither helps nor hurts Google visibility.

Other systems may experiment with or adopt llms.txt, so maintaining one can be a separate product decision. But I would treat it as optional infrastructure, not as the core of an AEO/GEO strategy. OpenAI’s current publisher guidance focuses on OAI-SearchBot access and crawlability for ChatGPT Search, not on requiring an llms.txt file.

What should you stop doing?

• Stop publishing commodity articles at scale. If the page says nothing beyond what any model can generate, it has very little defensive value.
• Stop creating one page for every tiny keyword variation. Build complete topic coverage around real intent instead.
• Stop hiding the answer below a long SEO introduction. Make the page useful immediately.
• Stop treating schema as a ranking hack. Use accurate supported markup to clarify what the page and entity are.
• Stop chasing fake brand mentions. Authority has to survive verification.
• Stop measuring only clicks. AI can influence awareness, comparison and brand choice before a referral ever appears.
• Stop calling every decline “AI.” Search performance can change because of demand, competitors, technical issues, content quality, SERP design, model changes or many things at once.

How do you measure AEO and GEO?

You need a layered dashboard because no single metric explains AI search. Traditional SEO metrics remain important, but citation and generative-visibility data now add another layer between discovery and conversion.

Google rolled out dedicated Search Generative AI performance reporting in Search Console worldwide in 2026, while Bing Webmaster Tools now reports AI citations, cited pages, grounding queries, topics, intents and citation share across supported AI experiences. ChatGPT Search referrals can be tracked in analytics, and OpenAI says its search links include a utm_source=chatgpt.com parameter.

Do not read one AI answer and call it a KPI. Generated answers are dynamic. Track a set of commercially relevant questions over time, compare average visibility, record which sources are cited, and look for trends rather than obsessing over a single prompt on a single day.

A practical 90-day SEO + AEO + GEO plan

If I were starting from zero, I would not begin by rewriting every page. I would build the foundation, choose the commercial topics that matter most and create evidence around them.

Days 1-15: establish the baseline

• Verify Search Console and Bing Webmaster Tools. Export current search queries, landing pages, indexing issues and AI-visibility data.
• Audit robots.txt, noindex, canonicals, redirects, sitemap quality, broken pages and bot/WAF access.
• Decide crawler policy for OAI-SearchBot and GPTBot based on the company’s search and training preferences.
• Choose 20-40 high-value questions the market asks across discovery, comparison, evaluation and decision stages.
• Record where the brand appears today across Google AI features, Copilot, ChatGPT Search and other relevant answer engines. Treat this as a baseline, not a ranking report.

Days 16-45: fix the information architecture

• Create or improve the core topic pages. Each major commercial theme should have one clear primary source of truth.
• Connect supporting articles with descriptive internal links instead of isolated blog posts.
• Normalize entity information: company, people, services, locations, product names, dates and claims.
• Implement accurate Organization, Person/author and Article markup where appropriate and validate it.
• Fix the pages that are already visible but weak: stale statistics, vague headings, duplicate intros, unsupported claims and unclear calls to action.

Days 46-75: create citable content

• Publish expert-led content around the questions that matter commercially, not just high-volume keywords.
• Add original evidence: screenshots, data, case examples, internal frameworks, comparisons and first-hand observations.
• Write direct answers near the start of key sections, then support them with depth.
• Create explanatory images or diagrams that can be understood on mobile and reinforce the same entities and facts as the text.
• Strengthen external credibility through legitimate PR, partnerships, expert contributions and third-party references.

Days 76-90: measure and iterate

• Compare Google generative AI visibility, Bing AI Performance and traditional search metrics against the baseline.
• Review which pages are cited and which grounding topics appear. Expand the themes where you have evidence of relevance.
• Check AI referral behavior and ask sales or lead forms to capture discovery source where appropriate.
• Sample the tracked question set again and record brand accuracy, citation sources and competitor presence.
• Update the next 90-day content plan based on missing information, not on whichever AI-search tactic is trending that week.

What does this look like in practice?

Three simplified examples show why SEO, AEO and GEO need to work together. These are illustrative, not Loor Connect client cases.

Example 1: a consultancy wants to be recommended for market entry

Traditional SEO might target “market entry strategy” and related service queries. AEO adds direct definitions, decision criteria, process steps and FAQs. GEO adds stronger entity signals, original experience across markets, cited evidence, a clear author identity and credible third-party references. The goal is not only to rank for the phrase. It is to become a source that can support an answer about how market entry should be done and which type of partner fits the problem.

Example 2: an ecommerce brand wants product comparisons

SEO makes category and product pages crawlable and relevant. AEO makes specifications, pricing logic, use cases, availability and comparison criteria explicit. GEO depends on consistent product facts across the site, merchant feeds, reviews, external references and enough trustworthy context for an AI system to compare the product accurately. Images also matter because search is increasingly multimodal.

Example 3: a B2B company wants to be shortlisted by AI

A blog alone is not enough. The company needs clear service pages, strong problem/solution language, identifiable expertise, customer evidence, current company information and outside signals that corroborate its position. AI visibility becomes a reputation problem as much as a content problem.

What changes when search becomes agentic?

The next shift is from answering to acting. Browser agents and task-oriented systems can already inspect page structure and accessibility information to understand forms, buttons, menus and actions. Google now explicitly recommends staying aware of agent-friendly website practices, while OpenAI documents the value of ARIA labels and accessible interactive elements for agent interaction in its browser experiences.

That means the future search question is not only “Can an AI cite our page?” It is also “Can an AI understand what can be done here, trust the information and complete the next step safely?” Clean information architecture, accessible interfaces, accurate inventory or service data and unambiguous calls to action become part of discovery.

What to do next

If the current search strategy was built around rankings alone, do not throw it away. Expand it.

• Protect the SEO foundation. Fix crawling, indexing, canonicals, redirects, internal links, sitemap quality and page experience first.
• Choose the questions that affect revenue. Build around real customer decisions, not just keyword volume.
• Create fewer, stronger sources of truth. Give each important topic a page that is complete, current and worth citing.
• Make expertise visible. Add authors, evidence, original insight, sources, examples and clear entity information.
• Measure the new visibility layer. Add generative AI reports, citations, grounding queries, AI referrals and brand visibility to the dashboard.
• Review quarterly, not once. Search interfaces, crawler rules and reporting are changing quickly; this article should be treated as a living operating model.

Final thoughts

SEO is not dead. The comfortable version of SEO is.

The part where a company could publish a page, target a keyword, watch a ranking and call the job finished is becoming less useful as a complete strategy. Search now happens inside answers, conversations, comparisons, images and increasingly actions. The brands that win will not be the ones that invent the most GEO hacks. They will be the ones that make their expertise easy to find, easy to verify, easy to understand and worth using.

If I had to reduce the whole article to one line, it would be this: build for the customer, make the truth machine-readable, and give every search or AI system a good reason to trust the source.

See you,

Tai Perez

Tai Perez