Get cited by the AI engines your buyers use

Per Ramp's first-party research, markdown was the only format that reliably surfaced in LLM responses. Bots ask for markdown. Browsers get HTML. The same URL can serve both. That is content negotiation, and instrumenting it per page is the single change with the widest citation coverage.

Roboto's own site is the worked example. The content is MDX on disk, so the markdown is the source rather than a format derived from a CMS export. When an LLM crawler asks for markdown at a given URL, it gets a clean document with structured headings, inline links, and no CMS wrapper noise. We wrote up the approach in our Next.js AEO/GEO guide. On a CMS-driven stack it's a transform step rather than a route handler, but the goal is the same clean markdown at the canonical URL.

Quarterly advisory cycles break at the feedback step. You get a report in month three describing what happened in month one, and by the time an edit clears approval and publish, the competitive set has shifted again. The diagnostic is stale before the fix ships.

A background agent watches citation share across platforms on a weekly cadence. When share drops on a target prompt cluster, it flags the change, identifies what moved in the competitive set, and drafts a remediation PR: a content edit, a schema patch, an llms.txt update, or a content negotiation rule. A human reviews and ships, then the agent re-monitors that cluster and measures whether share recovered. We run this on our own site and documented the build in why we built our own background agent. For an engagement we set the same loop up on your stack: your data sources, your content repo, your approval flow.

Bot classification is where most implementations break. Single-signal classification on User-Agent alone misses the bulk of AI crawler traffic. Correct classification combines UA, IP range, ASN, and bot score. The Cloudflare gotcha compounds it: ChatGPT, Perplexity, and Claude bots are categorised as "AI Assistants" in Cloudflare's taxonomy, not "AI Search", so firewall and analytics rules targeting only "AI Search" miss all three major platforms.

The rest is table stakes we verify before touching anything higher up. Schema goes in where it helps LLMs parse structure (ServiceSchema, FAQPage, BreadcrumbList, Article). The llms.txt file gets prioritisation logic rather than a flat list of 400 pages. Crawlability is genuinely non-negotiable: every list item rendered into the DOM, navigation in static HTML, correct status codes, and clean submitted sitemaps.

Volume compounds citation share only when the content clears a quality bar that LLMs trust. The editorial infrastructure is what makes volume safe to ship, so we build three gates into the publishing pipeline.

An AI-tell detector runs before merge, flagging em-dashes, negative parallelism, banned phrase families, and AI vocabulary; content that fails goes back for a humanizer pass. A fact-checker agent verifies specific claims against their cited primary sources and flags any that don't hold. A humanizer pass strips the sentence structures that read as generated. The cms-auto-updater pipeline is the worked example: it pulls evidence from primary sources on a weekly cadence, drafts updates, runs all three gates, and opens a PR for review. The result is far more content per quarter without the citation-eroding slop that comes from raw AI publish pipelines.