AI Search Optimization — get cited when the answer replaces the link

Search used to end in a click. Increasingly it ends in a paragraph: ChatGPT answers, Gemini summarises, Perplexity cites three sources, and Google writes an AI Overview above the results. Your rankings can hold steady while your sessions fall, because the answer got assembled somewhere you have no analytics on. Nothing broke. The interface moved.

AI Search Optimization, often called Generative Engine Optimization, is the work of becoming one of the sources that answer is assembled from, and of shaping what gets said about you when it is.

What AI search optimization means in our hands

We build the retrieval systems we optimise for. Redenn ships RAG pipelines, embeddings, rerankers and evaluation harnesses for clients, and runs its own AI chatbot in production. We know how a passage actually gets selected: it has to be retrievable, self contained, unambiguous about who is making the claim, and corroborated elsewhere. That is an engineering view of the problem. See /services/llm-development for the other side of the same coin.

Entity first, keywords second. A model does not rank you, it resolves you. Before anything cites you it has to know the business exists, what category it sits in, who runs it, where it operates and what it sells. We treat your entity as the deliverable: a linked schema graph on your own domain, consistent naming and description across every profile and mention, and clean signals for the knowledge panel.

Machine readable by construction. AI crawlers are less forgiving than Googlebot. Content that only appears after JavaScript runs, or hides behind an accordion, often is not there at all as far as retrieval is concerned. We serve clean, server rendered HTML with real semantic structure, and we verify it rather than assume it.

What a typical engagement includes

• An AI visibility baseline: a prompt set built from the questions your buyers genuinely ask, run across the major assistants and AI Overviews, logging who gets cited, what gets said about you, and where competitors are named instead
• An entity and knowledge graph audit covering name, address, founding, leadership, services and description across every source a model can reach
• Structured data built or repaired: Organization, Service, FAQPage, Article, Person and Product, connected with @id and sameAs so the graph resolves as one entity
• An llms.txt file and a deliberate AI crawler policy: what you invite assistants to read, what you keep out, and a machine readable map of the pages that matter
• Content rebuilt to be citable: a direct answer near the top, defined terms, dated and attributable claims, named authors, first hand experience instead of paraphrased consensus
• A source consistency pass across directories, professional profiles, review platforms and press, so corroboration works in your favour
• Monitoring: recurring prompt runs tracking answer share, citation counts and sentiment inside answers, reported on a cadence you choose

The stack and the approach

• Delivery: Next.js 16, server rendered, Lighthouse 95+ and WCAG 2.1 AA as standard. Retrievability, speed and accessibility are the same problem wearing different hats.
• Markup: hand written JSON-LD graphs rather than plugin output, validated and version controlled with the rest of the site.
• Measurement: an internal harness that runs your prompt set against model APIs on a schedule, so change over time is evidence rather than anecdote.
• Foundations: classic search discipline underneath, because assistants retrieve from the indexed web. This sits on top of SEO, never instead of it.

What this looks like in practice

Two shapes these engagements usually take.

A GEO layer on an existing site. A business that still ranks respectably but has stopped appearing inside AI answers. We baseline the prompt set, repair the entity and schema graph, publish llms.txt, rewrite the highest value pages to be quotable, tidy the off site sources, then stand up monitoring and hand it over.

Build and GEO together. A rebuild where information architecture, schema graph and content model are designed for retrieval from the first wireframe rather than retrofitted later. This pairs with /services/web-design, and ongoing monitoring can fold into a Care Plan.

Pricing

Scoped per project. The cost depends on how many locations you carry, how far your markup and off site sources have drifted, and how much content needs rebuilding. You get a fixed price after a scoping call. Contact us for a quote, or send a URL through the free audit first.

Ready to be the source, not the omission

Tell us the questions your buyers ask before they buy. We will run them through the assistants and show you exactly who gets cited today. Every brief gets a reply within four business hours.

Start an AI Search Optimization project → — or book a 30-min call