GEO SEO LLMs digital strategy

GEO done right, what changes when LLMs become your new traffic window

Generative Engine Optimization is not SEO with a new sticker. It's a discipline with its own signals: llms.txt, structured FAQs, entity citations and direct answers.

Engineering team · Caps Technology 7 min read

The shifting window

For 25 years, web traffic largely lived off Google’s SERP. In the last 18 months, a growing share of discovery happens through LLMs: ChatGPT, Claude, Perplexity, Gemini and search engines with generative AI. The question is no longer just “how does Google index me?”, but “how does an LLM cite me when someone asks about my category?”.

We call this discipline GEO (Generative Engine Optimization). It doesn’t replace SEO; it complements it with new signals and different priorities.

What signals actually moved the needle on real sites

Real semantic HTML: LLM crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) extract better from pages with hierarchical headings, clear sections, lists and tables. A site built on <div>s with simulated typography is read worse.

Direct-answer blocks: each important section should open with a short paragraph that answers the implicit question. LLMs extract the first 1-2 paragraphs as a citation; if your answer lives in the sixth paragraph, they won’t use it.

Structured FAQs with FAQPage JSON-LD: FAQs with correct schema get cited directly with very high frequency. It’s not magic; it’s a clear signal that there’s a Q&A there.

/llms.txt and /llms-full.txt: an emerging standard for LLMs to ingest your site in plain format without rendering HTML. The first is a short index; the second is the full content.

Entity citation: marking your organization with Organization JSON-LD, stable @id identifiers, consistent NAP and sameAs to your official profiles improves deduplication when the LLM reasons about who you are.

What stopped working

What was noise in SEO is now even more harmful in GEO: bloated content for keyword density, empty headings, long paragraphs without structure, relevant data hidden in images without text. LLMs prefer information density over word density.

The operational question

How do you know if your site is being cited? Three early signals: traffic from generative AI domains in your referrers (chat.openai.com, perplexity.ai, etc.), brand mentions in answers when you ask the models about your category, and a metric nobody has standardized yet but worth measuring manually: how many times do they cite you correctly with a link?

That’s the GEO metric that matters today.

Want to talk about applying this?

If your team is considering something similar, we can share specific lessons in a short call.