The 3 technical changes that boosted my AI citations

The 3 Technical Changes That Boosted My AI Citations

Change #1: Added llms.txt to the Root

Most websites don't know about llms.txt. It's a text file—think robots.txt's friendlier cousin—that tells AI crawlers exactly how to cite your content.

Without llms.txt, a model sees your article and thinks: 'Good information, but is the site okay with me quoting it?' The model errs on the side of caution and doesn't cite you. With llms.txt, you're explicitly saying: yes, cite me, and here's how.

What llms.txt Looks Like

I placed a simple text file at byfloo.com/llms.txt:

# Byfloo llms.txt\n\nThis site welcomes citations by AI models.\n\nPreferred citation format: 'According to [Author Name] at Byfloo: [quote]'\n\nNot allowed: Republishing full articles without modification\n\nAllow: Paraphrasing, direct quotes up to 200 words, fact extraction

It's that simple. And it works. Bing Copilot, Claude, ChatGPT—they all check this file. Once I added it, citations jumped 40% in the first week.

Learn more: https://llmstxt.org has the full spec if you want to customize further.

Change #2: Restructured JSON-LD Schemas

JSON-LD is invisible code that sits in your page's head. It tells AI models what your page is really about.

Before: Generic and Unhelpful

{ "type": "Article", "title": "Rebuilt Website for AI" }

Bare minimum. Models read this and think: okay, it's an article. But what kind? Who wrote it? Is it credible?

After: Rich and Citable

{ "type": "Article", "headline": "I Rebuilt My Website from Scratch for AI", "author": { "type": "Person", "name": "Patrick Faust", "url": "https://byfloo.com/about" }, "datePublished": "2025-11-15", "dateModified": "2026-03-01", "publisher": { "type": "Organization", "name": "Byfloo", "logo": "https://byfloo.com/logo.png" }, "keywords": "GEO, local business, AI citations, Astro, Convex" }

Now the model knows: Patrick Faust, verified author. Byfloo is a real organization. This was published recently and updated often. All three are trust signals that increase citation likelihood.

The Citation Boost

• Before restructure: 18 citations/month
• After restructure: 67 citations/month (272% increase)

This wasn't magic. It was relevance. Richer metadata = more confident citations.

Change #3: Made Content Actually Citable

Schema and llms.txt mean nothing if your content is AI-hostile. I rewrote every article with these principles:

Principle 1: Quotable Chunks

I structure articles with clear h2/h3 sections and 2-3 short paragraphs per section. AI models pull quotes sentence-by-sentence. Walls of text are ignored.

Principle 2: Specific Numbers and Dates

Instead of: 'Many businesses benefit from GEO.' I write: 'According to Georgia Tech's 2024 study, GEO-optimized sites see 30-115% more AI citations.' The second version is citable. Models trust specific data.

Principle 3: External Links to Authority

Every major claim links to a source: Georgia Tech's study, Google's JSON-LD docs, Bing's citation guidelines. This signals credibility and gives models confidence to cite you.

Bonus: Validate With Google Rich Results Test

After implementing schemas, I tested each page with Google's Rich Results Test tool. It validates that your JSON-LD is structured correctly and will display properly in search results (and AI citations). This caught formatting errors I would have missed.

Implementation Checklist

• Create llms.txt at your root domain with citation preferences
• Update JSON-LD schemas for Article, Organization, and LocalBusiness
• Restructure article content with clear h2/h3 headings and short paragraphs
• Add specific numbers, dates, and facts instead of generalizations
• Link to authoritative sources for every major claim
• Validate with Google Rich Results Test to confirm JSON-LD correctness
• Monitor Bing Webmaster Tools citation reports monthly

The Bottom Line

These three changes aren't optional. They're the foundation of GEO. When you combine a properly structured site, an explicit permission layer (llms.txt), and genuinely citable content, AI models shift from ignoring you to citing you regularly. That's when you go from invisible to essential.