It’s a question I hear a lot lately, in some form or another. “Why bother with a personal site? My work is on GitHub. Anyone who wants to know what I do can just ask a model to summarize my repos.”
It’s a fair question. The answer that holds up is the opposite of what most people expect. The fact that LLMs can now read everything published online doesn’t make personal websites redundant. It makes them more important than they used to be.
The direct readership of doc pages is going down. People who would have read a tutorial three years ago now ask a model. That part is real and not a blip. What it doesn’t mean is that the writing has stopped mattering. The consumption shifted. It didn’t disappear.
Code shows what, prose shows why
LLMs are good at telling you what a piece of code does. They are not as reliable at telling you why it does that, what was considered and rejected, what’s load-bearing for a project’s design, and what’s a historical accident waiting to be cleaned up. That information lives in the writer’s head, and in the writing they did about the project. If no one writes it down, no one, human or model, can recover it from the code alone.
This used to be an internal-team problem. It is now the public version of your work that an LLM hands to a stranger. The cost of unwritten intent has gone up.
Models need anchors
The “just ask an LLM” workflow leans on the model finding useful chunks to ground its answer. Code is a noisy training signal. It optimizes for execution, not explanation, and the comments that exist are usually sparse and pragmatic. Prose written for human readers gives the model something better to retrieve and cite. A blog post explaining a design decision is what makes the LLM’s answer about a project accurate instead of plausible-sounding.
It also works in reverse. A wrong page used to mislead one developer; now it misleads everyone who asks a model about your project. Errors are amplified, not absorbed.
You own the page, you don’t own the model’s output
If you don’t write the canonical version of your work, something else will. An LLM summary built from your code and GitHub activity is one option, and it will be plausibly correct and slightly wrong about everything, the way those summaries usually are. Once the canonical version exists, that’s what humans cite, what future models train on, and what shows up when someone searches your name. The website is how you get a say.
Some readers still read pages
The other half of the half-LLM audience is still human. Hiring managers preparing for an interview. A peer deciding whether to invite someone to collaborate. A partner’s exec trying to figure out if someone is worth a meeting. None of those people are going to chat with a model about your GitHub. They are going to open a tab.
What’s worth writing now
Not every kind of writing is worth the effort it used to be. The 30-step screenshot walkthrough is dying because models compress it to five lines. What’s getting more valuable: high-precision reference content (because models cite it), conceptual writing about design intent (because models can’t reconstruct it), and interactive, runnable content like notebooks and sandboxes (because models can guide a developer through them but can’t replace doing the work).
So the page stays
Personal websites aren’t going anywhere. The cost of maintaining one is low, and the alternative is letting someone else, or no one, write the canonical version of your work. LLMs aren’t replacing the personal website. They are another reader you’re writing for, and probably the most demanding one.