Generative Engine Optimisation: ranking is now data engineering

Four layers of the same machine. SEO targets the crawler: low latency, sound structure, a clean sitemap, so the engine can reach the basement. GEO targets the data: JSON-LD entity graphs and structured facts, so the model can understand what it found. AISO targets validation: citations, sentiment, brand signals and behaviour, so the model has reason to trust you enough to recommend you. WebMCP targets the feed: a direct, permanent channel from your server to the model, so it never has to scrape.

Yes, because it is fundamentally more capital-efficient. Traditional SEO runs on attrition — retainers spent on blog farming and manual link-building just to hold a rank. GEO invests in permanent data assets: once your entity graph, routing files and citations exist, they keep working, and the retainer maintains and extends them rather than re-buying the same position every month.

Just as robots.txt tells search engines how to crawl, llms.txt is a markdown document written for large language models: a clean, immediate summary of your business. llms-full.txt is the full corpus — the facts, services and answers a model needs to cite you accurately. Both live on your server, and both are maintained under the retainers.

Clarity precedes deployment. The session measures how AI currently perceives your brand against your competitors, the state of your code and your latency, and what a build would have to achieve — so that what we deploy is engineered to your data, not to a template. The fee is credited in full against the deployment that follows.

The suite asks ChatGPT, Gemini, Claude and Perplexity your target questions every week — asked without your name — records whether you are named and what is said, and charts it. You see the line, not a claim. That is Share of Model. No one can guarantee that a model will cite a business; the line shows whether it does.