GEO 101: How to Get Cited by AI Search (llms.txt, BLUF, and Content Signals)

SEO got you ranked in Google. Now there’s a second frontier: getting cited by AI.

When someone asks ChatGPT “what does the fool tarot card mean”, or searches Perplexity for “how does true solar time correction work” — where does the answer come from? Not from your meta tags. Not from your backlinks. It comes from the models finding your content, understanding it, and deciding it’s worth quoting.

This is GEO — Generative Engine Optimization. It’s early, messy, and nobody has it figured out. But the patterns we’ve tested on our own sites suggest a few things that work.

The Core Insight

Google ranks pages. AI engines synthesize answers.

When Google crawls your page, it indexes the whole thing — every paragraph, every keyword, every link. When an AI engine processes your content, it’s looking for quotable, self-contained factual statements it can extract and present as an answer.

A 2000-word article with the key fact buried in paragraph 14? Google will rank it. An AI engine might miss it entirely. The structure of your content matters more for AI than it ever did for traditional SEO.

Three Things We’re Testing

1. llms.txt — A Sitemap for AI

You know robots.txt (for crawlers) and sitemap.xml (for indexers). llms.txt is the third file — for AI models.

It’s a plain-text manifest of your best content, organized by topic, with direct URLs. Think of it as a curated reading list you hand to any AI that visits your site.

# Site Name

> One-line description of what this site is about.

## Essential Pages
- [Topic A](https://yoursite.com/page-a): Brief description
- [Topic B](https://yoursite.com/page-b): Brief description

## Deep Dives
- [Topic C](https://yoursite.com/page-c): Brief description

We run a two-layer setup:

• /llms.txt — ~50 links, the greatest hits. Any AI bot that visits gets a digestible overview.
• /llms-full.txt — ~200 links, the complete index. For deep research queries.

The simplified version links to the full version at the bottom. A bot that wants depth can follow the trail.

2. BLUF — Bottom Line Up Front

Military communication principle: lead with the conclusion. AI engines love this.

On our educational pages, the first thing after the intro is a summary block — the key fact, stated plainly, in 2-3 lines. No fluff, no “in this article we will explore…”

Example for a tarot card meaning page:

The Fool (0)
Upright: new beginnings, spontaneity, free spirit | Yes/No: Yes
Reversed: recklessness, risk-taking, naivety | Yes/No: No

That block is the single most citable unit on the page. If an AI engine extracts nothing else, it extracts this. Every card meaning page on our site has one.

3. Content Signals in robots.txt

Your robots.txt can declare how you want AI to interact with your content:

# Content Signals
ai-train: no
search: yes
ai-input: yes

What this says:

• ai-train: no — don’t use my content to train your model weights
• search: yes — but you can surface my content in search/answer results
• ai-input: yes — and you can use it as input for generating answers (RAG/grounding)

This is the nuance most sites miss. “Block all AI” is a blunt instrument. The distinction matters: you want your content cited in AI answers (that’s free traffic), but you don’t necessarily want it absorbed into a model’s training data (that’s giving away your IP for free).

Does It Work?

Honest answer: it’s too early to tell with confidence.

What we know:

• ChatGPT, Perplexity, and Gemini do crawl the web and cite sources
• Having an llms.txt makes your content easier for them to discover and parse
• BLUF blocks are objectively more extractable than buried conclusions
• Content signals give you control over the train-vs-cite distinction

What we don’t know yet:

• How much llms.txt actually affects citation frequency vs. just having good HTML
• Whether AI engines weight these signals or ignore them
• The long-term traffic impact as AI search grows

We’re running this setup across multiple sites and tracking AI referral traffic. The early signals are positive — we see AI crawlers hitting llms.txt — but it’ll take months to measure real impact.

What to Do Today

1. Write an llms.txt — 30 minutes. List your 20-50 best pages with one-line descriptions.
2. Add BLUF blocks to your key educational pages — the core fact, stated in 2-3 lines, right after the intro.
3. Add content signals to robots.txt — decide your stance on training vs. citation.

None of this costs money. None of it requires special tools. The barrier is just knowing to do it — which is why being early matters.

---

This is an active area of experimentation. We’ll publish follow-ups as data comes in. Follow along at zens.ink or star the open-source toolkit.