Six layers between your business and an AI's answer.
Each layer maps to a measurable failure mode. Skip one, and the layers above it can't compensate.
Indexing & crawler access
AI assistants only cite what search engines surface. We audit crawlability, indexing status and canonical structure, and verify AI crawlers — ClaudeBot, GPTBot, PerplexityBot, Google-Extended — aren't blocked in robots.txt. We ship XML sitemaps and fix orphaned pages.
Citations, statistics, quotes
We rewrite priority pages using the three highest-impact methods from GEO-bench: sourced claims, concrete numbers, and quotable attributed statements — the changes measured at ~30–40% visibility gains.[1]
Schema & llms.txt
JSON-LD entity markup (Organization, Service, FAQPage, Article, BreadcrumbList) removes ambiguity about who you are. An llms.txt file gives language models a plain-language map of your site.
Question-first pages
Pages built around the literal questions users ask assistants, with a direct, self-contained answer in the first paragraph — the passage an AI can lift cleanly — followed by depth for the reader.
Being cited elsewhere
Assistants cross-check sources, and most citations in AI answers come from third-party pages, not brand sites. We build presence in industry publications, directories, comparison articles and reference sites — the layer with the longest lead time and the largest effect.
Answer monitoring
Monthly tracking of your share of voice across Claude, ChatGPT, Perplexity and Gemini on your target query set: cited / mentioned / absent, with which sources won instead and why.
What layers 1 & 3 look like in code.
A robots.txt that welcomes AI crawlers instead of blocking them:
User-agent: ClaudeBot
Allow: /
User-agent: GPTBot
Allow: /
User-agent: PerplexityBot
Allow: /
Sitemap: https://example.com/sitemap.xmlAnd the first lines of an llms.txt — a markdown file at your site root summarizing what you are, in language written for a model to parse: see this site's own llms.txt. This entire website practices every layer it describes; view the page source to inspect the JSON-LD.
Lines we don't cross.
No prompt injection or hidden instructions to AI models. No fabricated statistics, fake reviews or invented credentials. No cloaking (serving different content to crawlers and humans). These tactics are detectable, they poison the corroboration layer permanently, and they are the generative-era equivalent of the link farms that got sites deindexed a decade ago.