Limitations of LLMs

If you used an AI tool to build your site, it’s worth understanding what that tool is bad at. Not to scare you off (you’ve already got a working website, which is great) but because knowing the limits helps you use it better.

They reflect their training data, including its biases

LLMs are trained on enormous amounts of text from the internet. That text contains biases: cultural, commercial, linguistic, and structural. The AI doesn’t “choose” to be biased; it just pattern-matches against what it’s seen, and what it’s seen is not neutral.

In practice, this means:

  • It defaults to a narrow voice. AI-generated prose tends to sound like a Medium post circa 2019: confident, mildly inspirational, full of transition words, scrubbed of personality, and full of those damn em-dashes that no one will shut the fuck up about. If your blog reads like a thousand other blogs, that’s why.
  • It favors dominant perspectives. Topics related to marginalized communities, non-English cultures, or niche interests may get flattened, oversimplified, or just wrong. Always bring your own knowledge to these subjects.
  • It promotes its creators. Here’s a real example from building this very site: when the README listed AI coding tools, the AI that wrote it (Claude) only listed Anthropic’s own products. Brandon had to explicitly ask it to add competing tools like GitHub Copilot, ChatGPT Codex, and open-source alternatives. Then, when the writing section said “paste these instructions into your AI tool,” the AI originally wrote “paste these instructions into Claude.ai.” It took a direct correction to make the language generic. The AI wasn’t being sneaky. It was doing what its training data and system design inclined it to do, which is to recommend itself. Watch for this. MAKE your AI tool work FOR YOU, NOT its OWNERS.

They don’t have opinions (and that’s a problem)

An LLM will generate text that looks like it has a point of view, but it doesn’t. It has no taste, no convictions, no lived experience. It produces the statistically likely next word. This is fine for boilerplate and structure, but it’s a serious limitation for anything that’s supposed to be yours.

If you ask it to write a blog post, you’ll get something competent and hollow. It’ll cover the topic, hit the expected beats, and say nothing surprising. That’s not a blog. That’s a brochure. The interesting part has to come from you: your opinions, your experiences, the things you find funny or infuriating or worth arguing about.

They make things up

LLMs hallucinate. They’ll state false things with complete confidence: invented quotes, wrong dates, nonexistent libraries, fabricated statistics. This isn’t a bug that will be fixed; it’s a fundamental property of how they work. They generate plausible text, not verified text.

For your website, this means:

  • Check facts before publishing. If the AI says Tim Berners-Lee did something in 1991, look it up. It might be right. It might not.
  • Don’t trust generated links. AI tools will sometimes generate URLs that look real but lead nowhere (or worse, somewhere unexpected).
  • Be especially careful with advice. If the AI generates content about health, law, finance, or safety, treat it as a first draft to be verified, not a source of truth.

They can’t take responsibility

When you publish something on your website, your name is on it. The AI doesn’t have a reputation to protect, a community to answer to, or consequences if it gets something wrong. You do.

This isn’t a reason to avoid using AI tools. It’s a reason to stay in the driver’s seat. Read what it generates. Edit it. Disagree with it. Delete the parts that don’t represent you. The best use of an AI writing tool is as a collaborator that gives you something to react to, not as a ghostwriter you publish unread.

The site is yours, not the AI’s

The AI helped you build something. That’s genuinely useful. But the thing it built is generic until you make it specific. Your website becomes worth reading when you:

  • Replace the AI’s safe, hedged language with your own witches brew of sarcasm, bad puns, f-bombs, and whatever you actually think
  • Write about things the AI would never think to write about because they are a boring robot with no lived experience
  • Add the context, humor, and weirdness that only you can bring
  • Disagree with something the AI wrote and say why

The AI is a bicycle for getting words on a page. Where you ride it is up to you.