I don't write code. I'm a UX/UI designer with 10 years in fintech. Everything on my blog is built with Claude as my coding partner. I describe what I want, Claude writes the PHP, and I verify it works.
Six months ago I found a study that changed how I think about content. Researchers ran 602 prompts across ChatGPT, Gemini and Perplexity, tracking 21,000 citations to find out why AI models cite certain content and ignore everything else.
The findings were simple. The implementation was simpler. And the results surprised me.
What the study found
Five content strategies ranked by how much they influence whether an AI model cites you:
Numerical data increases citation influence by 61.55%. Clear definitions at the start of sections increase it by 57.33%. Structured comparisons increase it by 55.28%. Step-by-step instructions increase it by 41.20%.
And Q&A format, the one every SEO guide tells you to use, decreases it by 5.74%.
That last one is the one that stuck with me. I had "What is llms.txt?" as a heading on one of my posts. I changed it to "llms.txt is a plain text file at the root of your domain that tells AI models what your site is about." Same information. Completely different structure. One is a question a model has to skip. The other is a statement it can cite directly.
What I actually changed
I'm not going to pretend I did anything sophisticated. I can't. I don't know enough about code to do anything sophisticated.
What I did was go through my existing posts and apply three rules.
First rule: every section starts with a definition or a fact. Not a question. Not "In this section we'll explore..." Just the information, first sentence, no preamble.
Second rule: numbers are always specific. Not "many companies lost traffic" but "an 8 billion dollar company lost 49% of organic visibility in 6 weeks." Not "my blog scored well" but "my blog scored 50 out of 100 on the Cloudflare Agent Readiness test."
Third rule: no Q&A format anywhere. Every question heading became a statement heading. Every "Have you ever wondered..." became a direct sentence.
That's it. Three rules. No new tools. No plugins. No framework migration. I opened each post in my admin panel and edited the text.
The part that surprised me
One ChatGPT citation is worth 4.6 times more than a Google click.
I have a domain rating of 3.2. I run vanilla PHP on shared hosting with cPanel. I'm competing against sites with teams of developers and marketing budgets I can't imagine. On Google, I don't stand a chance for competitive keywords.
But ChatGPT doesn't care about domain authority. It reads the content and decides if a specific statement answers the query. My blog has started getting cited not because it's authoritative in the traditional sense, but because the content is structured in a way that's easy for a model to extract and reference.
Five months in: 48 indexed pages up from 26. First ChatGPT citations without any backlink campaigns. An acceptmarkdown.com score of 88 out of 100.
The thing nobody tells you about AI optimization
It's not technical. That's the thing.
Every article about GEO (Generative Engine Optimization) makes it sound like you need to implement complex systems. JSON-LD schemas. Content negotiation protocols. Agent discovery files.
I did implement all of those things. But the single biggest impact came from editing how I write. Definitions first. Numbers specific. No Q&A format. That's free. Anyone can do it tonight.
The technical infrastructure matters for the long game. But if you're starting from zero and wondering where to begin, start with your words. The study data says that's where most of the citation influence lives.
My stack, for context
I mention this because I think it matters. You don't need a fancy setup.
Vanilla PHP 8.2. MariaDB. Apache. Shared hosting on cPanel. I pay less than $10 a month. Claude writes all the code. I'm the one who decides what to build and why, but I don't write a single line myself.
If a designer with no coding background can get AI models to cite a blog on shared hosting, the barrier to entry is not technical. It's knowing what to optimize for.
The full analysis with the study data and everything I implemented is here:
→ https://shinobis.com/en/602-prompts-21000-citations-how-ai-models-choose-what-to-cite
If you've noticed AI models citing your content (or not citing it), I'd love to hear what you think is driving it. Are you seeing the same patterns?
