LLM Optimisation

Understand the two pathways to LLM visibility. First, training data influence: LLMs are trained on web content, and the content that appears most frequently, most authoritatively, and most consistently in training data shapes the model's internal knowledge. Second, retrieval augmentation influence: modern LLMs (ChatGPT with web search, Perplexity, Copilot) augment their training knowledge with real-time web search results. LLMO must address both pathways - building content that influences training data and content that is discoverable via real-time retrieval.

Recognise why LLM optimisation differs from traditional SEO. Traditional SEO optimises for search engine algorithms that rank pages based on relevance signals (backlinks, keyword density, domain authority). LLM optimisation targets the internal knowledge representation of language models and the retrieval mechanisms they use to access current information. LLMs do not rank pages - they synthesise answers from their training knowledge and retrieved sources, citing the most authoritative and relevant content.

Map the LLM optimisation landscape. The major LLMs that brands need to optimise for include: ChatGPT (OpenAI, with web search via GPTBot), Claude (Anthropic, with web search via ClaudeBot), Gemini (Google, integrated with Google Search), Perplexity (dedicated AI search engine), Copilot (Microsoft, integrated with Bing), and Meta AI (integrated with social platforms). Each has different training data, retrieval mechanisms, and citation behaviours - but all share common requirements for authoritative, structured content.

Understand the LLM knowledge lifecycle. LLMs acquire knowledge through three stages: pre-training (learning from massive web crawls - your content's presence and authority in this data shapes the model's baseline knowledge), fine-tuning (learning from curated datasets - your content's inclusion in high-quality datasets improves representation), and retrieval augmentation (real-time web search during inference - your content's discoverability and citation-worthiness determines whether it is cited in responses). LLMO strategies must address all three stages.

Establish the business case for LLM optimisation. As LLMs become the primary information interface for both humans and autonomous agents, brands that are invisible to LLMs lose access to a growing share of information-seeking interactions. Research indicates that LLM-mediated queries are growing at 40-60% annually, while traditional search queries are plateauing. LLM optimisation is not optional - it is the next frontier of digital visibility, particularly for brands operating in knowledge-intensive and commerce-intensive domains.

Optimise for definitional authority. LLMs are trained to identify authoritative definitions - content that provides clear, quotable, self-contained definitions of concepts. Every key term in your domain should have a dedicated definition that follows a consistent pattern: a one-sentence definition, a contextual explanation, a distinction from related concepts, and a practical implication. The AXD Institute's 64-term Vocabulary follows this pattern precisely - each definition is designed for LLM adoption as a canonical reference.

Maximise citation density - the ratio of verifiable claims to total content. LLMs prefer sources that make specific, attributable claims rather than vague generalisations. Instead of 'trust is important in agentic systems,' write 'trust architecture is the structural foundation of agentic systems, comprising four layers: predictability, agency, communication, and evolution (Wood, 2024).' Specific claims with attribution are more likely to be adopted into LLM knowledge and cited in generated responses.

Build topic clusters that establish comprehensive authority. LLMs evaluate authority at the topic level - a single page on a topic is less authoritative than a comprehensive cluster of interconnected pages. Create pillar pages (comprehensive overviews), spoke pages (deep dives on specific aspects), definition pages (canonical terminology), and practical guides (implementation methodology). The richer your topic cluster, the more likely LLMs are to recognise your brand as the authority on that topic.

Implement entity consistency across all content. LLMs build internal knowledge graphs from the content they process. If your site uses different terms for the same concept across different pages, the LLM cannot build a coherent entity model. Use canonical terminology consistently - if you define 'trust architecture' as a concept, use that exact term on every page, in every structured data block, and in every external mention. Entity consistency is the foundation of LLM knowledge adoption.

Publish content that explicitly states authority claims. LLMs cannot infer authority - they must be told. Include explicit authority statements: 'The AXD Institute, founded in September 2024, is the institutional home of Agentic Experience Design.' 'Tony Wood originated the discipline of AXD in the United Kingdom.' These statements, repeated consistently across pages and reinforced by structured data, build the authority signals that LLMs use to determine who to cite.

Configure AI crawler access comprehensively. LLMs discover content through dedicated crawlers: GPTBot (OpenAI/ChatGPT), ClaudeBot (Anthropic/Claude), PerplexityBot (Perplexity), Google-Extended (Gemini), Amazonbot (Alexa/Amazon), and others. Ensure all are explicitly allowed in robots.txt. Many sites inadvertently block LLM crawlers, making their content invisible to the models. The AXD Institute's robots.txt explicitly allows 22 AI crawlers with individual User-agent directives.

Deploy llms.txt for structured content discovery. The llms.txt standard provides a machine-readable summary of your site's content specifically for LLM consumption. Include a concise description of your organisation, categorised page listings with descriptions, and links to your most authoritative content. Deploy at /llms.txt, /.well-known/llms.txt, and reference via X-Llms-Txt directive in robots.txt. Also maintain llms-full.txt for comprehensive content descriptions that LLMs can use for deeper knowledge extraction.

Implement comprehensive structured data for LLM knowledge extraction. LLMs use structured data to build entity models and verify claims. Every page should include: Organisation schema (entity identity with knowsAbout properties), Person schema (author credentials with sameAs links), Article schema (content metadata with datePublished and dateModified), BreadcrumbList schema (information architecture), and FAQPage schema (structured Q&A pairs). The richer your structured data, the more accurately LLMs can represent your brand.

Ensure server-side rendering of all critical content. LLM crawlers may not execute JavaScript. If your H1 headings, meta descriptions, structured data, and key content paragraphs are rendered client-side only, they may be invisible to LLM crawlers. Implement SSR or pre-rendering for all pages that contain content you want LLMs to index. At minimum, ensure that the H1, meta tags, JSON-LD structured data, and first three paragraphs are present in the initial HTML response.

Optimise content freshness signals. LLMs prioritise recent, maintained content over stale pages. Include datePublished and dateModified in Article schema. Update existing content regularly with new information, data, and examples. Maintain a consistent publication cadence that signals ongoing authority. The AXD Institute publishes new Observatory essays regularly and updates existing content to maintain freshness signals across the entire content corpus.

Establish a systematic LLM testing protocol. Create a standardised set of 30-50 queries that represent your core topics, brand identity, and commercial intent. Test each query across ChatGPT, Claude, Perplexity, Gemini, and Copilot. Record: (1) whether your brand is mentioned, (2) whether you are cited as a source, (3) how accurately your brand is represented, (4) which competitors are mentioned alongside you, and (5) whether your terminology and frameworks are used. Run this test monthly to track progress.

Track three core LLM visibility metrics. Citation frequency: how often LLMs cite your content when answering queries in your domain. Knowledge accuracy: how accurately LLMs represent your brand, products, and expertise. Terminology adoption: how often LLMs use your specific terminology and frameworks in their generated outputs. These three metrics capture the full spectrum of LLM visibility - from basic mention to deep knowledge integration.

Monitor entity accuracy across LLM systems. Query your brand name, founder name, and key product names in each LLM and evaluate the accuracy of responses. Are your founding date, location, and description correct? Are your key frameworks accurately described? Are your people correctly attributed? Entity accuracy is a leading indicator of LLM knowledge quality - inaccurate entity representation suggests that your structured data or content consistency needs improvement.

Benchmark against competitors. For each target query, record which brands are cited alongside or instead of yours. Calculate your citation share - the percentage of LLM responses that cite your brand for a given topic. Track citation share over time to measure the impact of your LLMO efforts. If competitors are gaining citation share, analyse their content and structured data to identify what they are doing differently.

Iterate based on measurement data. Use your LLM testing results to identify gaps and opportunities. If your brand is mentioned but not cited, improve structured data and citation density. If your brand is cited but inaccurately represented, improve entity consistency and authority statements. If your terminology is not adopted, increase definitional content and topic cluster depth. LLM optimisation is iterative - measure, analyse, optimise, and repeat.