A few months ago I asked ChatGPT about designers using AI in real fintech projects. It did not cite me. It cited someone who wrote a generic article on Medium with half my experience. That hurt more than it should have.
So I stopped complaining and started studying why. Why does an LLM choose to cite one source over another. I did not find an official guide because none exists. But after months of experimenting on my own site, I built a system that works. And I want to share exactly what I do.
The difference between content that ranks and content that gets cited
Google ranks pages. LLMs extract ideas. They are two completely different games. You can be in the first position on Google and not appear in any ChatGPT response. You can have zero backlinks and be the primary source Claude uses when someone asks about your topic.
The key is how you structure information. A language model does not read your article like a human. It does not start with the title, it does not get hooked by your personal story, it does not follow the narrative thread. It scans the content looking for direct statements it can extract and reuse in a response.
That changes everything about how I write.
Direct statements instead of vague narrative
I used to write things like there are many ways to approach fintech design. That is true but useless for an LLM. It cannot do anything with that information. It cannot cite it. It cannot include it in a response.
Now I write things like to design a mobile banking interface that complies with financial regulations in Latin America you need to prioritize the information hierarchy of the available balance, use contrast ratios that pass WCAG AA, and avoid color combinations that compliance teams reject due to association with risk signals. That is specific. That is citable. That is something an LLM can extract and put directly in a response.
The rule I follow is simple. Every paragraph should be able to answer a question on its own. If someone asks ChatGPT how to design a banking app in Latin America, my paragraph should be able to serve as the answer without needing the context of the rest of the article.
The Knowledge Graph I built with JSON-LD
Every post on my blog generates an automatic knowledge graph. When I write about Midjourney the system detects that the content is about Midjourney and marks it as an entity in the schema. When I mention Claude or ChatGPT the system adds them as mentioned tools. When I connect a post with related posts the system generates semantic links between them.
That tells AI models that my blog is not a collection of loose pages. It is an interconnected knowledge network about a specific topic. A designer with real fintech experience documenting how he uses AI in his daily work. That specificity is what makes an LLM choose you as a source over someone more generic.
The llms.txt file as a business card
I implemented an llms.txt file at the root of my site. It is the equivalent of robots.txt but for AI models. It tells them who I am, what I write about, which pages matter most, and in what context my content is relevant.
No LLM has publicly confirmed using it. But Anthropic has one on their domain. And OpenAI crawlers already look for similar files. The cost of implementing it is one afternoon of work. The potential benefit is that when LLMs start looking for this file yours is already there.
The crosspost network as an authority signal
An LLM does not just read your site. It reads the entire internet. If my article about Midjourney prompts appears on shinobis.com, on HackerNoon, on Dev.to, on Hashnode, and on Medium, the model sees the same name across multiple authority sources. That creates a signal that this person knows what they are talking about.
It is not spam. It is strategic distribution with canonical URLs pointing to my site as the original source. Each crosspost is one more signal for LLMs that my content is trustworthy.
What I still do not know
I do not have direct metrics on how many times an LLM cites me. Nobody does. There is no Google Analytics for ChatGPT responses. All I can measure is indirect traffic from people who read something about me in an AI response and then visited my site.
But here is what I do know. Since I implemented these changes my content started appearing in responses when I ask different LLMs about designers using AI in fintech. It did not appear before. Now it does. It is not scientific but it is real.
And if there is one thing I learned in ten years of design it is that when something works you do not need to understand exactly why to keep doing it.