--- title: "LLM SEO: how to optimize for AI search and citations" description: "A practical LLM SEO system for crawl access, source selection, citable passages, entity clarity, and measurement across ChatGPT, Perplexity, and Google." canonical: "https://nikoalho.fi/writing/llm-seo/" language: "en" --- > Canonical source: [https://nikoalho.fi/writing/llm-seo/](https://nikoalho.fi/writing/llm-seo/) [← writing](https://nikoalho.fi/writing/) GEO Published 2026 · 07 · 17 Updated 2026 · 07 · 18 # LLM SEO: how to optimize for AI search and citations A practical LLM SEO system for crawl access, source selection, citable passages, entity clarity, and measurement across ChatGPT, Perplexity, and Google. ![Niko Alho](https://nikoalho.fi/assets/niko-alho-headcut.webp) **Niko Alho**Operator in Turku · firsthand systems ![Editorial illustration for LLM SEO: how to optimize for AI search and citations](https://nikoalho.fi/visuals/llm-seo-960.webp) ON THIS PAGE [01 LLM SEO vs. SEO vs. GEO](#llm-seo-vs-seo-vs-geo) [02 How AI Search Systems Find Sources](#how-ai-search-systems-find-sources) [03 The Six-Stage LLM SEO System](#the-six-stage-llm-seo-system) [04 What Not to Do](#what-not-to-do) [05 A 30-Day Implementation Order](#a-30-day-implementation-order) [06 The Standard to Aim For](#the-standard-to-aim-for) PROGRESS 0% ON THIS PAGE 6 sections [01 LLM SEO vs. SEO vs. GEO](#llm-seo-vs-seo-vs-geo) [02 How AI Search Systems Find Sources](#how-ai-search-systems-find-sources) [03 The Six-Stage LLM SEO System](#the-six-stage-llm-seo-system) [04 What Not to Do](#what-not-to-do) [05 A 30-Day Implementation Order](#a-30-day-implementation-order) [06 The Standard to Aim For](#the-standard-to-aim-for) TL ; DR tl;dr. 13-min read 5 takeaways 1. 01 LLM SEO is the work of making your pages discoverable, retrievable, and worth citing in AI-generated answers. It is an extension of SEO, not a replacement for it. 2. 02 Crawler access is engine-specific. OpenAI separates OAI-SearchBot for search from GPTBot for training, so one robots.txt decision does not cover both. 3. 03 Google says no special AI markup or llms.txt file is required for its generative Search features. Core SEO, visible text, and valid structured data still do the work. 4. 04 The best unit of optimization is a supported claim : a direct answer, the evidence behind it, and enough context to quote it without distortion. 5. 05 Measure mentions, citations, cited URLs, and referrals separately. A brand mention is not a citation, and a citation does not guarantee a click. DIRECT ANSWER Q. What is LLM SEO? A. LLM SEO is the practice of making web content easy for AI search and answer systems to discover, retrieve, understand, and cite. It combines technical SEO, clear passage-level answers, reliable evidence, consistent entity information, and engine-specific measurement. It does not require a separate duplicate site or a new schema vocabulary. EVIDENCE Google's official guidance says optimization for its generative Search features remains SEO and requires no special AI files or markup. OpenAI separately documents OAI-SearchBot for ChatGPT search and GPTBot for model training, confirming that access and training are distinct decisions. **LLM SEO** is the practice of making a page discoverable, retrievable, understandable, and citable inside AI-generated answers. That definition matters because most advice starts at the wrong end. It jumps straight to short paragraphs, FAQ blocks, or an `llms.txt` file. None of those helps if the engine cannot fetch the page, cannot identify the canonical URL, retrieves a better-matching source, or does not trust the claim enough to cite it. LLM SEO is a system. The system starts with access and ends with measurement. Content formatting sits in the middle. ## LLM SEO vs. SEO vs. GEO Classical SEO tries to earn visibility in a ranked search result. LLM SEO tries to earn inclusion in an answer: as a source, a linked citation, or a brand recommendation. The disciplines share the same foundation: - useful pages that satisfy a real intent; - crawlable HTML and stable URLs; - clear site architecture and internal links; - external signals that validate the source; - accurate, current information. The difference is the unit of selection. A search engine can rank a whole document. An answer system often retrieves passages from several documents, then decides which claims are useful enough to synthesize or cite. [Generative engine optimization](https://nikoalho.fi/writing/generative-engine-optimization/) is the broader distribution strategy around these systems. LLM SEO is the operating layer: access, retrieval, passage design, entity consistency, and measurement. The labels are less important than knowing which stage failed. CRITERIA search result visibility Classical SEO answer and citation visibility LLM SEO WIN Primary surface Ranked result Generated answer, mention, or citation Optimization unit Page and query Passage, claim, entity, and source Technical floor Indexable HTML Indexable HTML plus intentional AI crawler policy Evidence Supports trust and links Also makes claims safer to extract and cite Measurement Rank, impressions, clicks Prompt coverage, mentions, citations, cited URLs ## How AI Search Systems Find Sources There is no single “LLM index.” The major surfaces use different crawlers, search indexes, retrieval systems, and source-selection logic. Google’s official position is the simplest: its generative Search features use the existing Search foundation. A page must be indexed and eligible to show a snippet. Google says there are no extra technical requirements, no special AI schema, and no need for an `llms.txt` file to appear in those features. The [AI Overview optimization guide](https://nikoalho.fi/writing/ai-overview-optimization/) turns that eligibility requirement into a surface-specific test plan. OpenAI exposes separate controls. `OAI-SearchBot` is used to surface sites in ChatGPT search, while `GPTBot` is associated with potential model-training use. `ChatGPT-User` represents user-triggered visits. OpenAI states that these controls are independent, so a site can allow search discovery while declining training use. The [ChatGPT SEO playbook](https://nikoalho.fi/writing/chatgpt-seo-playbook/) covers the corresponding crawler, citation, and referral checks. Perplexity documents a continuously refreshed search index in its Search API. The practical implication is familiar: source discovery, relevance, freshness, and an extractable answer still matter. The exact source-selection weights are not public. For a surface-specific implementation, the [Perplexity optimization playbook](https://nikoalho.fi/writing/perplexity-optimization/) separates what you can verify—crawler access, source retrieval, citations, and referral visits—from claims about unpublished ranking weights. Treat each engine as a separate distribution surface. A rule written for one crawler is not a universal AI policy. ## The Six-Stage LLM SEO System ### 1\. Access: make the crawler policy intentional Start with `robots.txt`, but do not stop there. Decide separately whether you want: - classical search crawling; - AI search discovery; - user-triggered page fetches; - model-training crawling. Then verify the same policy at the CDN and web application firewall. An allowed user agent still fails if bot protection challenges its requests or blocks its published IP range. Do not copy a generic “allow every AI bot” block without understanding it. Search visibility and training permission are different business decisions. ### 2\. Discovery: expose one stable version of every answer Every priority page needs: - a self-referencing canonical URL; - at least one contextual internal link; - inclusion in the XML sitemap; - a truthful `lastmod` value when supported; - a clean 200 response without a redirect chain; - meaningful HTML without requiring a client-side interaction. For Bing-connected surfaces, IndexNow can shorten discovery time after meaningful updates. It does not improve the content and it does not guarantee inclusion. It simply reports changed URLs faster. Optional machine-readable endpoints can help agents navigate a site. This site exposes `/llms.txt`, `/index.json`, RSS, Markdown mirrors, and RFC 8288 `Link` headers. Those are useful discovery contracts. They are not evidence of higher rankings, and Google explicitly says they are unnecessary for its AI Search features. ### 3\. Retrieval: match a real question precisely Retrieval happens before citation. If the system does not pull your page into the candidate set, rewriting one sentence will not help. Build pages around distinct intents, not minor keyword variants. A useful page should make three things obvious: 1. What question does this page answer? 2. For whom is the answer valid? 3. What evidence or experience makes this source different? This is where [semantic SEO](https://nikoalho.fi/writing/semantic-seo-for-ai/) and topical architecture matter. Related pages provide context, but duplication creates ambiguity. One canonical answer per intent is a better retrieval target than five pages competing to say the same thing. ### 4\. Extraction: write claims that survive being quoted The useful unit is not “short content.” It is a supported claim. A citable passage usually contains: - a direct answer in the first sentence; - the scope or condition that limits the answer; - a number, example, method, or source that supports it; - enough surrounding context to avoid a misleading extraction. Tables help comparisons. Lists help procedures. Definitions help entity questions. None is automatically superior. Use the structure that matches the information. Do not manufacture statistics to look citable. A transparent small sample, including the prompt set and collection date, is stronger than an impressive percentage with no method. The recommendation studies on this site follow that principle by publishing the model panel and answer counts alongside the result. ### 5\. Source and entity selection: make the author verifiable Answer systems do not only compare sentences. They compare sources. Keep the basic entity facts consistent across the site and reputable external profiles: - person and organization names; - role and area of expertise; - contact details and location where relevant; - authorship and publication dates; - product or service names; - citations to the original source rather than a chain of summaries. Structured data can reduce ambiguity when it matches visible content. It is not permission to invent authority. A `Person` node with `knowsAbout` is weaker than a body of signed work, third-party references, and original evidence that demonstrates the expertise. The off-site part is covered in [entity authority for LLM retrieval](https://nikoalho.fi/writing/entity-authority-for-llms/). The same rule applies there: corroboration beats self-description. ### 6\. Measurement: separate mentions, citations, and clicks These are different events: - **Mention:** the answer names the brand. - **Citation:** the answer links to or explicitly identifies the source. - **Referral:** a user clicks from the answer to the site. - **Conversion:** the visit produces a business outcome. Track them separately across a fixed prompt panel. Store the exact prompt, engine or surface, model when visible, date, location, brand mention, citation URL, competitors, and a copy of the surrounding response. AI answers vary between runs. One screenshot is an anecdote. A repeated panel is a time series. Use [AI citation tracking](https://nikoalho.fi/writing/tracking-ai-citations/) for the dashboard design and [AI share of voice](https://nikoalho.fi/writing/how-to-measure-ai-share-of-voice/) for competitive scoring. Pair those with GSC and analytics instead of forcing every outcome into one metric. FIG. 01 · THE LLM SEO PIPELINE ACCESS robots + WAF → DISCOVER links + sitemaps → RETRIEVE intent + relevance → EXTRACT claims + evidence → CITE source selection → MEASURE fixed prompt panel A page can fail at any stage. Fix the first broken stage before rewriting everything. ## What Not to Do ### Do not build a duplicate “AI version” of the site Maintain one canonical source of truth. A Markdown representation or JSON index can be derived from that source, but it should not become a second editorial system that drifts away from the visible page. ### Do not add schema that users cannot see Structured data must describe the page. A hidden FAQ, fake review, or generated HowTo block creates inconsistency rather than clarity. ### Do not confuse crawler access with recommendation Allowing a bot means the page can be fetched. It does not mean the page will be retrieved, cited, recommended, or trusted. ### Do not report unlinked mentions as traffic AI visibility can influence demand without a click, but that does not justify assigning revenue to every mention. Keep visibility, referral, assisted conversion, and sourced pipeline separate until the data supports a relationship. ## A 30-Day Implementation Order **Week 1: access and inventory.** Record crawler policy, WAF behavior, canonical status, sitemap coverage, indexability, and server-rendered content for every priority page. **Week 2: intent and evidence.** Map the commercial prompt set to canonical pages. Mark prompts with no suitable answer, duplicated answers, or claims that lack primary evidence. **Week 3: passage and entity repair.** Rewrite the weak answer blocks, add methodology and sources, connect related pages, and align visible entity facts with valid structured data. **Week 4: baseline measurement.** Run the fixed prompt panel, store cited URLs and competitors, and annotate the release. Do not call a one-week change a trend. The order is deliberate. Technical access without useful answers produces crawl logs, not citations. Great answers behind a blocked crawler produce nothing. Measurement before a fixed prompt set produces screenshots, not evidence. ## The Standard to Aim For A strong LLM SEO page should be useful even if AI search disappears tomorrow. It answers a real question, shows its evidence, states its limits, links to the next relevant concept, and gives the reader a reason to trust the author. AI systems benefit from those properties because people do too. That is the durable strategy: build the best source, expose it cleanly, and measure each distribution surface on its own terms. NEED THE SYSTEM, NOT ANOTHER CHECKLIST? I audit and operate AI search visibility for B2B teams. [Book a 20-min intro →](https://nikoalho.fi/book/) Questions people actually ask FAQ · 5 Q01 Is LLM SEO different from GEO? + They overlap. GEO is the broader practice of earning visibility in generative answer systems. LLM SEO is a useful operational label for the crawl, retrieval, content, entity, and measurement work that supports that outcome. Neither term describes an official Google ranking system. Q02 Do I need an llms.txt file to rank in AI search? + No. Google explicitly says llms.txt is not required for AI Overviews or AI Mode. Other platforms have not established it as a ranking signal. It can still be a useful voluntary index for agents and developers, but treat it as discovery infrastructure, not a ranking shortcut. Q03 Should I allow GPTBot? + That is a training-policy decision. OpenAI documents GPTBot for potential model-training use and OAI-SearchBot for ChatGPT search visibility. You can allow search while blocking training because the controls are independent. Q04 Does schema markup make LLMs cite a page? + No. Structured data can make entities and page facts less ambiguous, but it does not create citation eligibility by itself. Use supported schema types, keep them consistent with visible content, and do not add FAQ, review, or HowTo markup for content that users cannot see. Q05 How do I measure LLM SEO? + Run a fixed panel of commercially relevant prompts across the engines you care about. Log brand mentions, linked citations, cited URLs, competitors, date, model or surface, locale, and response context. Repeat on a fixed cadence and pair the trend with referral and conversion data. Sources & further reading 1. \[01\] [Google's guide to optimizing for generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) Google Search Central · 2026 documentation 2. \[02\] [Overview of OpenAI crawlers](https://platform.openai.com/docs/bots/overview-of-openai-crawlers) OpenAI · 2026 documentation 3. \[03\] [Search API](https://docs.perplexity.ai/docs/search/quickstart) Perplexity · 2026 documentation 4. \[04\] [Generative Engine Optimization](https://arxiv.org/abs/2311.09735) Aggarwal et al. · 2024 research 5. \[05\] [IndexNow protocol documentation](https://www.indexnow.org/documentation) IndexNow documentation ![Niko Alho](https://nikoalho.fi/assets/niko-alho-profile.webp) Niko Alho I run agentic SEO and build custom AI for B2B companies. Based in Turku. [About →](https://nikoalho.fi/about/) KEEP READING ## More on geo. - [![Editorial illustration for AI Recommendation Index: what six models recommend](https://nikoalho.fi/visuals/ai-recommendation-index.webp) GEO 2026 · 07 · 17 AI Recommendation Index: what six models recommend A transparent index of 144 AI answers across four software categories, with share of voice, first-pi… read →](https://nikoalho.fi/writing/ai-recommendation-index/) - [![Editorial illustration for Which accounting software does AI recommend?](https://nikoalho.fi/visuals/ai-accounting-software-recommendations-2026.webp) GEO 2026 · 06 · 23 Which accounting software does AI recommend? Which accounting software do AI assistants recommend? Results from 36 answers across six models, wit… read →](https://nikoalho.fi/writing/ai-accounting-software-recommendations-2026/) - [![Editorial illustration for Which CRM does AI recommend? 36 answers, 6 models](https://nikoalho.fi/visuals/ai-crm-software-recommendations-2026.webp) GEO 2026 · 06 · 23 Which CRM does AI recommend? 36 answers, 6 models Which CRM do AI assistants recommend? A 36-answer test across six models, with brand frequency, firs… read →](https://nikoalho.fi/writing/ai-crm-software-recommendations-2026/) [More writing →](https://nikoalho.fi/writing/) Booking June · 2 spots left · 20-min intro, no pitch [Book a slot →](https://nikoalho.fi/book/) ## Structured data ```json { "@context": "https://schema.org", "@type": "WebSite", "@id": "https://nikoalho.fi/#website", "url": "https://nikoalho.fi/", "name": "Niko Alho", "description": "Agentic SEO and custom AI builds for B2B companies.", "inLanguage": "en", "publisher": { "@id": "https://nikoalho.fi/#person" }, "potentialAction": { "@type": "SearchAction", "target": { "@type": "EntryPoint", "urlTemplate": "https://nikoalho.fi/search/?q={search_term_string}" }, "query-input": "required name=search_term_string" } } ``` ```json { "@context": "https://schema.org", "@type": "Person", "@id": "https://nikoalho.fi/#person", "name": "Niko Alho", "givenName": "Niko", "familyName": "Alho", "url": "https://nikoalho.fi/about/", "image": "https://nikoalho.fi/og/default.png", "jobTitle": "Agentic SEO & Custom AI Consultant", "email": "mailto:contact@nikoalho.fi", "telephone": "+358401539426", "address": { "@type": "PostalAddress", "addressLocality": "Turku", "addressCountry": "FI" }, "knowsAbout": [ "Search Engine Optimization", "Agentic SEO", "Topical Authority", "Retrieval-Augmented Generation", "Large Language Models", "Custom AI Builds", "B2B SaaS Content Strategy", "Schema.org Structured Data", "Generative Engine Optimization" ], "knowsLanguage": [ "en", "fi" ], "worksFor": { "@id": "https://nikoalho.fi/#organization" }, "sameAs": [ "https://www.linkedin.com/in/nikoalho/", "https://github.com/alhoniko" ] } ``` ```json { "@context": "https://schema.org", "@type": "ProfessionalService", "@id": "https://nikoalho.fi/#organization", "name": "Niko Alho — SEO & AI Automation", "alternateName": "Niko Alho", "description": "Agentic SEO and custom AI builds for B2B companies.", "url": "https://nikoalho.fi/", "image": "https://nikoalho.fi/og/default.png", "logo": "https://nikoalho.fi/assets/logo-mark.svg", "email": "mailto:contact@nikoalho.fi", "telephone": "+358401539426", "priceRange": "$$$", "founder": { "@id": "https://nikoalho.fi/#person" }, "employee": { "@id": "https://nikoalho.fi/#person" }, "knowsLanguage": [ "en", "fi" ], "address": { "@type": "PostalAddress", "addressLocality": "Turku", "addressCountry": "FI" }, "areaServed": [ { "@type": "City", "name": "Turku" }, { "@type": "City", "name": "Helsinki" }, { "@type": "Country", "name": "Finland" }, { "@type": "Place", "name": "European Union" }, { "@type": "Place", "name": "Worldwide (remote)" } ] } ``` ```json { "@context": "https://schema.org", "@type": "TechArticle", "@id": "https://nikoalho.fi/writing/llm-seo/#article", "headline": "LLM SEO: how to optimize for AI search and citations", "name": "LLM SEO: how to optimize for AI search and citations", "description": "A practical LLM SEO system for crawl access, source selection, citable passages, entity clarity, and measurement across ChatGPT, Perplexity, and Google.", "image": "https://nikoalho.fi/og/llm-seo.png", "url": "https://nikoalho.fi/writing/llm-seo/", "datePublished": "2026-07-17T00:00:00.000Z", "dateModified": "2026-07-18T00:00:00.000Z", "inLanguage": "en", "isAccessibleForFree": true, "wordCount": 1556, "articleSection": "GEO", "keywords": "llm seo, seo for llms, ai search optimization, llm optimization, chatgpt seo, ai citations", "author": { "@id": "https://nikoalho.fi/#person" }, "publisher": { "@id": "https://nikoalho.fi/#person" }, "mainEntityOfPage": { "@type": "WebPage", "@id": "https://nikoalho.fi/writing/llm-seo/" }, "about": { "@type": "Thing", "name": "GEO" }, "speakable": { "@type": "SpeakableSpecification", "cssSelector": [ "h1", ".tldr", ".article-body > .prose > p:first-of-type" ] } } ``` ```json { "@context": "https://schema.org", "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://nikoalho.fi/" }, { "@type": "ListItem", "position": 2, "name": "Writing", "item": "https://nikoalho.fi/writing/" }, { "@type": "ListItem", "position": 3, "name": "LLM SEO: how to optimize for AI search and citations", "item": "https://nikoalho.fi/writing/llm-seo/" } ] } ``` ```json { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "Is LLM SEO different from GEO?", "acceptedAnswer": { "@type": "Answer", "text": "They overlap. GEO is the broader practice of earning visibility in generative answer systems. LLM SEO is a useful operational label for the crawl, retrieval, content, entity, and measurement work that supports that outcome. Neither term describes an official Google ranking system." } }, { "@type": "Question", "name": "Do I need an llms.txt file to rank in AI search?", "acceptedAnswer": { "@type": "Answer", "text": "No. Google explicitly says llms.txt is not required for AI Overviews or AI Mode. Other platforms have not established it as a ranking signal. It can still be a useful voluntary index for agents and developers, but treat it as discovery infrastructure, not a ranking shortcut." } }, { "@type": "Question", "name": "Should I allow GPTBot?", "acceptedAnswer": { "@type": "Answer", "text": "That is a training-policy decision. OpenAI documents GPTBot for potential model-training use and OAI-SearchBot for ChatGPT search visibility. You can allow search while blocking training because the controls are independent." } }, { "@type": "Question", "name": "Does schema markup make LLMs cite a page?", "acceptedAnswer": { "@type": "Answer", "text": "No. Structured data can make entities and page facts less ambiguous, but it does not create citation eligibility by itself. Use supported schema types, keep them consistent with visible content, and do not add FAQ, review, or HowTo markup for content that users cannot see." } }, { "@type": "Question", "name": "How do I measure LLM SEO?", "acceptedAnswer": { "@type": "Answer", "text": "Run a fixed panel of commercially relevant prompts across the engines you care about. Log brand mentions, linked citations, cited URLs, competitors, date, model or surface, locale, and response context. Repeat on a fixed cadence and pair the trend with referral and conversion data." } } ] } ```