A Generative Engine Optimization (GEO) playbook covering llms.txt, citation-grade content, and the entity-graph signals LLM crawlers actually use. This guide breaks down the playbook into ordered steps with the tools, metrics, and common pitfalls at each stage — built for operators who'd rather execute than read theory. Built for SEO and content leads optimising for LLM citations.
A Generative Engine Optimization (GEO) playbook covering llms.txt, citation-grade content, and the entity-graph signals LLM crawlers actually use.
GEO ≠ SEO. LLM crawlers don't use rankings; they use citation density, entity strength, and structured data.
llms.txt + llms-full.txt is the new robots.txt for LLM ingestion.
Cited sources earn 4–7x more LLM-driven referrals than uncited.
Built for SEO and content leads optimising for LLM citations. Updated 2026.
Includes step-level execution detail + common mistakes + metrics + tools + adjacent question cross-links.
Anchored to the Frameleads Growth System™ — the open methodology that's documented end-to-end at /frameleads-growth-system.
This page is part of the Frameleads operator library. It's intentionally long — operators report that the short version sells, but the long version actually executes. Skim the key points if you're scanning; read top-to-bottom if you're committing.
Below: the direct answer, the operational detail, the common mistakes that show up in our audits, the metrics to track, the recommended stack, and adjacent reading.
The playbook matters because in 2026 operators have access to more execution surfaces than at any point in the last decade — yet most engagements still fail not from lack of options but from operating without a documented framework. This page is the framework, written down.
Each step builds on the previous; out-of-order execution leaves gaps that the later steps can't fill. Where steps overlap in calendar time, that's called out per-step.
Per the llmstxt.org convention: /llms.txt is a Markdown index of your high-value pages. /llms-full.txt is the concatenated full content. These guide LLM crawlers to canonical sources without scraping HTML.
Use schema.org sameAs to link your brand entity to Wikipedia (if eligible), Crunchbase, LinkedIn, and Wikidata. LLMs disambiguate entities through this graph.
LLMs cite specific claims with named sources. Format: '<claim> — <source>'. Avoid passive aggregation; use named expert quotes, named methodologies, and named benchmarks.
FAQPage schema + 8–12 FAQ items per pillar topic. Perplexity and Claude lift FAQ answers verbatim 40%+ of the time when schema is clean.
GA4 referral source filter for 'chat.openai.com', 'perplexity.ai', 'claude.ai', 'gemini.google.com', 'copilot.microsoft.com'. Tag in BigQuery as 'llm_referral' for cohort analysis.
OpenAI (GPTBot), Anthropic (ClaudeBot), and Google-Extended respect robots.txt. The llms.txt convention is voluntary but increasingly honored by Perplexity and Anthropic. Block training, allow inference if you want citations.
LLM index refresh cycles run 2–8 weeks depending on the model. Expect first citations from Perplexity in 4–6 weeks, ChatGPT in 6–10 weeks, Claude in 8–12 weeks after publishing.
The named-step durations are listed inline; total elapsed time depends on how many steps run in parallel. A typical sequential execution takes 20-30 weeks; parallel execution compresses that by 30-50%.
In-house works when you have the seniority + bandwidth on the named-step disciplines. Most teams that try in-house solo end up doing 60-70% of the work and missing the cross-step optimisation. An agency or fractional senior compresses time-to-result by 30-50% on average.
Budget breaks into three lines: agency fee (if applicable), media spend, and tools. The combined minimum to make data-driven decisions in 2026 is ₹1L/month for paid-heavy playbooks. Below that, manual optimisation in-house is more honest than an agency retainer.
Quarterly. Each quarter, review the leading indicator (movement) and the lagging indicator (outcome). If both are positive: scale. If leading is positive but lagging isn't: wait one more quarter. If leading is negative: change the playbook, not just the spend.
The framework transfers; the specifics (CPCs, channels, compliance, language overlays) need adapting. The named steps are universal; the within-step tactics adapt to the local market.
Cited primary and analyst sources. Independent of Frameleads' own data.
European data protection regulation.
US influencer / endorsement disclosure rules.
The operator framework that informs this guide.
Full operator library — glossary, calculators, guides, comparisons.
Frameleads runs this playbook for live clients. Free 30-min audit — three concrete moves for your specific stage, even if you don't engage us.