Why LLM Optimization Guidance Does Not Transfer Like SEO Guidance Did

What changed

Search Engine Journal recently published a thoughtful opinion piece arguing that LLM guidance does not transfer the way SEO guidance did. The point is important because the market is rushing to package “LLM optimization,” “GEO,” “AEO,” “AI visibility” and “AI citations” as if they were simply the next version of classic SEO.

They are connected, but they are not identical. A strong technical SEO foundation still matters. Useful content still matters. Authority, entity clarity, links, structured pages, fast rendering and crawl access still matter. But the same recommendation may not produce the same result across different AI systems. A page that is easy for Google to crawl may not be selected by ChatGPT. A brand that appears in Perplexity may not appear in AI Overviews. A page that is cited for one query may be ignored when the prompt changes slightly.

This is a major shift for business owners because SEO used to be easier to explain at the operating level. You made pages crawlable, useful, relevant, internally linked, technically healthy and authoritative. You monitored rankings, impressions, clicks and conversions. You improved the website over time. The exact ranking formula was never public, but the practical direction was stable enough to build a durable workflow.

AI Search introduces a new problem: fragmentation. The user’s journey can start in Google, continue in AI Mode, happen inside ChatGPT, appear through a voice assistant, surface in a shopping agent, or be summarized in an answer engine that never looks like a traditional SERP. Each surface can have different retrieval behavior, different freshness, different citations and different tolerance for vague content.

Why traditional SEO guidance transferred better

Traditional SEO guidance transferred because the web had common mechanics. Search engines crawled URLs. They parsed HTML. They followed links. They evaluated canonical signals. They looked at content, metadata, internal linking, structured data, authority, spam patterns and user usefulness. Google was dominant, but many best practices were broadly sensible across search engines because they improved machine access and human usefulness at the same time.

If you fixed a broken canonical system, that usually helped more than one crawler. If you improved internal linking, that helped users and search engines understand relationships. If you compressed images and improved mobile speed, the page became better for people and easier to consume. If you removed thin tag archives or duplicate pages, crawl quality improved. If you wrote a genuinely useful guide, it could earn links, mentions, shares and recurring visibility.

Google’s own documentation still reinforces this foundation. Its guidance for generative AI features says that the same SEO best practices that help Google Search discover, crawl and index pages are relevant for AI Overviews and AI Mode in Google Search. Google also continues to emphasize helpful, reliable, people-first content, page experience, crawlability and structured data where it matches visible content.

That is why the practical SEO playbook became transferable: not because every search engine worked exactly the same, but because the work improved the website in ways that many discovery systems could use.

Why LLM guidance does not transfer as cleanly

LLM optimization guidance is harder to transfer because LLM-powered systems are not only ranking documents. They may retrieve passages, summarize multiple sources, use search indexes, use proprietary browsing tools, incorporate product feeds, rely on partner data, remember session context, interpret prompts differently and decide whether a citation is necessary. The output is not just “position 3.” It is an answer, a recommendation, a list, a synthesized explanation, a shopping suggestion or no citation at all.

Several factors create fragmentation.

First, retrieval systems differ. One AI product may use a web index. Another may use live search. Another may use a selected corpus. Another may combine search results with its own browsing layer. Another may rely heavily on structured product data. If the retrieval layer is different, the optimization strategy cannot be identical.

Second, citation behavior differs. Some systems cite sources prominently. Some cite selectively. Some use sources without always exposing them. Some are more likely to cite listicles, comparison pages, official documentation, review pages, forums or high-authority media depending on the query. A brand cannot assume that earning one citation in one product means broad AI visibility.

Third, prompts change the result. A classic keyword query is short and relatively stable. AI prompts are longer, contextual and multi-intent. “Best pediatric clinic in Bucharest” is different from “private pediatric clinic in Bucharest for a toddler with recurring fever, good parking, online booking and parent reviews.” The second query contains criteria, context and decision logic. The website has to be legible for that richer retrieval task.

Fourth, user context matters more. AI systems can preserve conversation context. They can refine answers through follow-up questions. They can compare options. This means visibility may depend on how well a brand answers connected questions, not only one keyword.

Fifth, different AI systems have different crawlers and controls. OpenAI documents different crawlers for different purposes, including search and user-triggered browsing. Anthropic has documented crawler behavior and separate user-agent identities. Blocking or allowing one bot is not the same as controlling all AI visibility.

Google is a special case

Google AI Overviews and AI Mode should not be treated as completely separate from SEO. They are part of Google Search. Google’s official AI optimization guidance says that site owners should keep following Search Essentials and SEO best practices. That matters. It means crawlability, indexability, helpful content, structured data, page experience and normal Google Search visibility are still the starting point for AI features inside Google.

But even inside Google, the journey is changing. AI Overviews can synthesize information above classic results. AI Mode can expand the user journey with follow-up questions. Multimodal search, deeper prompts and AI-assisted exploration can change which pages are useful. This means a page must do more than match a keyword. It must answer the task.

For example, a clinic page should not only say “pediatric clinic Bucharest.” It should help a parent understand services, age ranges, booking options, emergency boundaries, doctor credentials, reviews, parking, insurance, location and what to do next. A technical SEO audit page should not only define the term. It should explain the checks, risks, examples, prioritization and what happens after issues are found.

In Google’s ecosystem, classic SEO remains the base. But richer answer readiness, entity clarity and helpful page structure become more important as users ask more specific questions.

ChatGPT, Claude and other systems are not Google clones

ChatGPT Search, Claude, Perplexity and other AI assistants should not be treated as small Google replicas. They may search, browse, retrieve, summarize and cite differently. Some systems expose source links. Some may use external search providers. Some can browse pages when users ask. Some can process files, PDFs or page content. Some will become more commerce-oriented through product feeds and agentic shopping workflows.

This matters because advice like “add FAQ schema” or “write listicles” is too narrow. It may help in some cases, but it is not a universal AI visibility strategy. Google has even announced that FAQ rich results are being deprecated from its rich result reporting and support path. That does not mean FAQ sections are useless; visible question-answer content can still help users and retrieval systems. But chasing one markup pattern as the entire strategy is fragile.

The durable strategy is to make the business easy to understand, compare, trust and cite. That requires:

• clear service and product pages;
• specific answers to real user questions;
• structured sections and headings;
• visible evidence, examples and author context;
• internal links between related topics;
• crawlable HTML content;
• clean technical foundations;
• authority signals from relevant external sources;
• ongoing monitoring across AI surfaces.

The AI SEO playbooks that will age badly

The first bad playbook is “generate thousands of pages.” AI makes content production cheap, but cheap content is not automatically useful. If pages repeat the same generic explanations, add no original value and do not help a specific user make a decision, they are unlikely to build durable search or AI visibility. In some cases, they can create crawl waste and dilute quality.

The second bad playbook is “optimize for one LLM.” This is tempting because it feels concrete. But the market is moving too fast. If a business optimizes only for one assistant’s current citation behavior, it may miss Google, Perplexity, Gemini, Claude, vertical search, shopping agents and future interfaces.

The third bad playbook is “schema solves it.” Schema is useful when it accurately describes visible content, but it is not a magic citation switch. Structured data should support clarity, not replace substance.

The fourth bad playbook is “brand mentions are enough.” Mentions matter, but AI systems need context. A brand mention without clear topical relevance, entity relationships, useful pages and strong supporting content may not help much.

The fifth bad playbook is “reports are strategy.” AI visibility dashboards are useful, but a dashboard does not improve a page. The work still has to be prepared, approved and executed.

A better operating model for AI Search visibility

The better model starts with humility: nobody has a universal LLM ranking formula. The practical approach is to build a website that is useful to people, legible to machines and operationally easy to improve.

Step one is technical access. Important pages should be crawlable, indexable, fast, canonical, mobile-friendly and easy to render. If a page is blocked, slow, duplicated, hidden in JavaScript or buried in poor internal linking, AI visibility work starts on weak ground.

Step two is content usefulness. Each important page should answer the user’s real decision. Generic content is less defensible. Specific content, examples, comparisons, pricing logic, process explanations, FAQs, trust signals and expert context are more useful.

Step three is entity clarity. The website should make it obvious who the business is, what it does, where it operates, what services or products it offers, who it serves and how it differs. This matters for both classic search and AI retrieval.

Step four is semantic internal linking. Related pages should support each other. A service page should connect to guides, FAQs, examples, case studies, glossary definitions and related services. A blog article should not live alone.

Step five is authority building. Relevant publisher mentions, citations, partnerships, expert profiles, case studies, reviews and external references help the web understand the brand. This is one reason AYSA integrates with Adverlink for controlled authority-building workflows.

Step six is monitoring. A business must track not only rankings, but also mentions, citations, answer presence, pages ignored by AI systems, competitors cited instead, and topics where the brand is missing.

Step seven is execution. This is the part most teams underestimate. If the monitoring system finds a gap but nobody updates the website, nothing changes.

The AYSA view

AYSA’s position is simple: AI Search visibility is not solved by a one-time prompt test or a static report. It is an operating problem. The website must keep becoming clearer, more useful, more technically accessible and more authoritative as the search environment changes.

For SMEs, this is hard because they do not have the time or specialist knowledge to monitor Google, ChatGPT, Claude, Perplexity, AI Overviews, AI Mode, Search Console, content decay, internal links, technical SEO and authority opportunities manually. They also do not want blind automation changing the website without approval.

That is why AYSA uses an approval-first model. The agent learns the business, monitors opportunities, prepares SEO, AEO, GEO and AI visibility actions, explains why they matter, asks for approval and executes accepted changes inside the website workflow. The user’s job is not to become an AI Search specialist. The user’s job is to review important decisions and stay in control.

In my opinion, the winning businesses in the AI Search era will not be the ones chasing every new acronym. They will be the ones with the strongest execution rhythm: monitor, understand, improve, approve, publish, measure and repeat. The playbook will not transfer perfectly across every LLM. But the operating discipline will.

Sources and further reading

• Search Engine Journal: LLM guidance does not transfer the way SEO guidance did
• Google Search Central: Optimizing your website for generative AI features
• Google Search Central: Creating helpful, reliable, people-first content
• Google Search Central: Search Essentials
• OpenAI Platform: Crawlers and user agents
• Anthropic Support: Web crawling and crawler controls
• Schema.org: About structured data vocabulary
• AYSA: Google’s Generative AI Optimization Guide and what SMEs should do next
• AYSA: Why AI Search cites some websites and ignores others