LLM SEO 2026 — Citation Tactics for ChatGPT, Claude, Gemini, Perplexity

LLM SEO (Large Language Model SEO) is the discipline of optimizing pages for citation by language models — ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Perplexity, Copilot (Microsoft), Mistral. Emerged 2023-2024 with mass adoption of LLM as search tools (~25% of searches in 2026 happen in LLM instead of traditional SERP). LLM SEO is SUBSET of AI SEO, focused specifically on chat-based engines (LLM) — distinct from optimization for Google AI Overviews. 3 types of citation in LLM: (1) DIRECT QUOTE — LLM cites sentence with link to source (Perplexity always, ChatGPT/Claude in Web Search mode). (2) ATTRIBUTED MENTION — LLM mentions brand/source without quote. (3) PARAPHRASED INTEGRATION — LLM includes your facts without source attribution (most common + hardest to measure). LLM SEO fights for (1) and (2).

llms.txt is a new file (proposed standard September 2024 by Jeremy Howard, fast.ai) — markdown file at root of domain (gmweb.pl/llms.txt) that provides LLM crawlers a focused overview of website: list of most important pages, brief description of each, links, key facts. Works like sitemap.xml but more semantically — in markdown format LLM easily parses. STRUCTURE llms.txt: # Site name + brief description (1-2 sentences). ## Section (e.g. "Main services") -- [Link](url) one-sentence description. STATUS 2026: unofficial standard but adopted by Anthropic (Claude.ai), Cursor.com, mintlify.com. Google/OpenAI not officially declaring but their crawlers likely don't ignore. MOST IMPORTANT: llms.txt does NOT REPLACE sitemap.xml or robots.txt, but COMPLEMENTS. Create both. EXAMPLE GMWEB: gmweb.pl/llms.txt — LLM sees our top 30 pages, each with 1-sentence description.

Active LLM crawlers 2026: (1) GPTBot (OpenAI for ChatGPT) — User-agent: GPTBot, docs: openai.com/gptbot. (2) OAI-SearchBot (OpenAI for ChatGPT Search new 2024) — User-agent: OAI-SearchBot. (3) ClaudeBot (Anthropic for Claude) — User-agent: ClaudeBot. (4) Claude-Web (Anthropic legacy) — User-agent: Claude-Web. (5) Google-Extended (Google for Gemini/Bard) — User-agent: Google-Extended. (6) PerplexityBot (Perplexity AI) — User-agent: PerplexityBot. (7) cohere-ai (Cohere) — User-agent: cohere-ai. (8) FacebookBot (Meta for LLaMA training) — User-agent: FacebookBot. (9) Bytespider (TikTok for content discovery LLM) — User-agent: Bytespider. (10) Bingbot (Microsoft Copilot) — uses Bingbot mainly. CONFIGURATION: if you want to allow LLM citation (recommended for content sites): User-agent: * Allow: /. If forbid: User-agent: GPTBot Disallow: /. GMWEB STANDARD: we allow all LLM crawlers — it's free brand awareness.

TESTING METHODOLOGY (takes 30-60 min): (1) MAKE LIST OF 15 QUERIES — primary keywords + informational queries from niche. (2) TEST IN EVERY LLM — ChatGPT (Free + Plus + Search), Claude (Sonnet 4.6), Gemini (Pro), Perplexity (Pro), Copilot (Microsoft), Mistral Le Chat. (3) FOR EVERY QUERY RECORD — does your domain appear in sources / citations / mentions, how long excerpt cited, position in sources. (4) COMPARE WITH COMPETITION — which domains most often cited? They are your real LLM SEO competitors. (5) DASHBOARD IN GOOGLE SHEET — rows = queries, columns = LLMs, values = position. AUTOMATING TOOLS: Profound (most popular 2024+), Otterly.ai, BrandRank.ai — $100-500/month, automatically test 100+ queries weekly. For SMB: ChatGPT API script for $50/month suffices.

8-point per-page optimization framework: (1) ANSWER-FIRST — first section 40-80 words = concise answer to primary query. (2) FACT-DENSITY — inject numbers, dates, names, brands in first 2 paragraphs. (3) UNIQUE PERSPECTIVE — don't copy-paste from Wikipedia. Give your interpretation, case study, own benchmarks. (4) CHUNKING H2/H3 = QUESTIONS — use headers as questions. (5) STRUCTURED DATA — Schema.org Article / FAQPage / HowTo / Person / Organization / Product. (6) BRAND CONSISTENCY — always same brand name in title/H1/copy/footer. (7) OUTGOING CITATIONS — link authoritative sources (Wikipedia, gov, edu). (8) RECENCY SIGNALS — datePublished + dateModified Schema + "Updated: date" in copy.

RAG (Retrieval-Augmented Generation) is architecture where LLM, instead of answering only from its training data (which is "stale" — cut at training date), AT QUERY TIME pulls fresh content from web (or private knowledge base) and includes it in generated answer. CHATGPT with Web Search uses RAG. Perplexity is RAG-first product. Claude.ai with Web Search uses RAG. Gemini Pro Real-time uses RAG. HOW RAG AFFECTS LLM SEO: (1) FRESHNESS MATTERS — in RAG mode LLM cites fresh content. dateModified Schema + new dates in copy = higher citation chance. (2) FRESH CRAWL — pages must be crawled by search engine used by LLM. (3) PASSAGE INDEXING — RAG cuts short passages (50-200 words) from whole page. Short chunked sections = better retrievability. (4) QUERY-CONTENT MATCHING — RAG uses embedding similarity. Write in user language, not corporate jargon.

PARTIALLY. GA4 still weak at tracking AI referrers in 2026. PROBLEMS: (1) ChatGPT, Claude, Gemini RARELY pass referer URL — often direct or utm-less. (2) Perplexity has referer "perplexity.ai" — trackable. (3) Bing Copilot passes "bing.com" — impossible to distinguish from regular search. SOLUTIONS 2026: (1) PLAUSIBLE ANALYTICS — introduced "AI referrers" custom property in 2024 (chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com). (2) CLOUDFLARE ANALYTICS — shows LLM bots crawl + AI referrer traffic. (3) URL TAGGING — signal LLMs to use your UTMs. (4) BRANDED SEARCH MONITORING — track increase in branded queries in Google Search Console after appearing in LLM. (5) CONVERSION FORMS — field "where did you hear about us" with "ChatGPT/Claude/Gemini/Perplexity" option = direct measurement.

PROTOCOL "FIRST 30 DAYS LLM SEO" for SMB: WEEK 1 — AUDIT BASELINE: (a) test 15 queries in 4 LLM, save mentions vs competition. (b) audit robots.txt — are GPTBot/ClaudeBot/Google-Extended allowed? (c) audit Schema.org. WEEK 2 — TECHNICAL FOUNDATION: (a) create llms.txt (5-15 pages with 1-sentence description). (b) check SSR/SSG rendering. (c) add missing Schema. WEEK 3 — CONTENT REWRITE: (a) pick top 5 pages. (b) rewrite first 100 words = answer-first format. (c) add 6-10 FAQ Q&A with Schema. WEEK 4 — MEASURE: (a) re-test 15 queries. (b) set monthly cadence. RESULT AFTER 3 MONTHS: typically 20-50% queries start citing you in 1-2 LLMs (vs 0% baseline). AFTER 12 MONTHS: 50%+ branded mentions in LLM for niche queries.