Hugo provides a way to easily theme your website with a premade theme. Here is a list of hundreds of themes to choose from. With so many choices, it’s easy to run into analysis paralysis, the state of being indecisive because there are so many choices to pick from.

It’s the year 2026, so naturally, AI can be useful, but I don’t want to oversell it.

Know what you want first

You could always go with the default, but sometimes (often?) the default does not align with your interests. The default theme on the Hugo quick start guide is the Ananke theme, but after looking at it fr about 10 seconds I immediately disliked it for a number of reasons:

1. As the default theme, it’s generic. For a personal blog, I want at least some degree of personality.
2. The list of blogs only had a few blog posts visible on the viewport, with their first paragraph visible, followed by a read more button. This feels odd to me.
3. The header for each article is overly large. Not a big deal, but it’s just not my thing.

This taught me I needed to know what I wanted.

What do I value in a blog theme?

1. It needs to be lightweight, both on the frontend and on the backend. People might be browsing it on weak hardware like a Qualcomm Snapdragon 425, so if it works smooth on that it’ll be even smoother on stronger hardware.
2. It needs to be small, ideally under 250KB gzipped data transferred for the theme. (I call this my GZ250 principle after my first motorcycle.)
3. No stupid anti-user stuff like pop-ups, trackers, or ads (not that Hugo themes would contain the latter two).
4. It should be easy to access the archive of blog posts, ideally in 2 clicks or less.

So many mainstream blogging platforms egregiously violate the first 3 principles, and many violate the last one as well. Substack, Medium, etc all struggle on low-end hardware, download multiple megabytes to load simple text articles, and have multiple pop-ups interrupting you to sign in to a newsletter, download an app, or something like that.

The part where AI comes in

It turns out that you can prompt an AI with specific requirements for a theme, and it will narrow down your choices elegantly. I prompted Claude with these requirements, and it suggested Hugo Bear blog, followed by Hugo Bear Cub and PaperMod. I did a little digging into each, settling on Hugo Bear Cub for minimalism but with a couple modern niceties like accessibility and security.

AI is so good at doing this that it is easy to become reliant on it for decision-making, especially if you don’t know the subject matter very well. If you don’t do your research on the AI’s picks, you could fall into a trap. Let me explain.

The trap of AI suggestions

All digital services controlled by for profit corporations go through a cycle popularly known as “Enshittification”. Basically, the service will at first act good to its users to expand the userbase, often at a loss, then abuse its users users to make things better for their business customers, later abusing the business customers to claw back all the value for themselves.

Many AI companies are in the first stage right now to gain users. They are not operating profitably, losing tens of billions of dollars per year, bolstered by the promise of future growth. Then, once the userbase becomes captive, they will switch on the ads. Once the ads start coming, they don’t stop coming (unless you pay the subscription, and even then they will find ways to weasel in “featured products”, looking at you Spotify). Some are transitioning into the second stage, like OpenAI, who have already started putting ads in ChatGPT. For now, they supposedly don’t influence the output, but the screws will always tighten over time.

There will be a day in the not-too-distant future where popular AI models will absolutely start insidiously changing the output of their AI models to align with some advertiser’s interests. Sometimes, it might be easy to avoid the ads, but other times, the output might be just ads. Imagine you are asking AI for the best washing machine that is under a certain size and price, and has some feature. The output might just be ads for crappy washing machines, and you will never know! It might get very hard to know what the quality washing machines are, since both the SEO-swamped traditional internet search and the AI search are just layers of ads.

This example highlights the enormous potential of LLMs becoming a competent digital salesman. I bet that soon, most online shopping platforms will heavily push this kind of AI salesman onto prospective buyers, with many outputing insidious sponsored suggestions. This is to say, always do your research when asking AI for what option to pick, espeically when it costs money (but even if it doesn’t!) Anyways, wasn’t this post was supposed to be about a blog post theme?