Why I tried it
For some reason, I had become increasingly unhappy with the search feature on my blog, so I asked an LLM what other options were available. The recommendations were surprisingly consistent: Pagefind. It was presented as a strong solution, complete with detailed explanations, and as someone who did not know much about it, I took that advice at face value and let it drive the design and implementation.
Pagefind Pagefind is a fully static search library that aims to perform well on large sites, while using as little of your users’ bandwidth as possible, and without hosting any infrastructure.
What I ended up with
I spent quite a bit of time on it—several hours in total, spread across a few days—and reworked the setup three separate times. In the end, I only barely managed to get fuzzy search working across the full content of a static site.
To be fair, it does have real strengths.
- It can search across the entire site.
- Fuzzy matching works.
- Search is fast.
- The index is generated during the build process, which is convenient.
But the problems were big enough that I gave up on it.
The issues I ran into
1. The search was too broad to be useful
Yes, Pagefind could search everything, but that was also the problem. The matching felt so loose that entering a single keyword could bring back results from almost every static page on the site. At that point, the search stopped helping and started becoming noise.
2. It generated far too many index files
The second issue was even more frustrating: the sheer number of index files. Eventually I decided to remove the feature altogether, and deleting those files took an absurd amount of time.
Part of that delay was my own fault. I did not delete them directly; I sent them to the recycle bin first. Still, what caught me off guard was the scale. My blog only has a few hundred pages, yet Pagefind generated more than 10,000 index-related files.
Even worse, when I tried to empty the recycle bin, it got stuck. In the end, I had an LLM write a small Python script, checked what it was doing, and used terminal commands to remove the files instead. Thankfully, that finally worked.
What this taught me
This whole experience was a good reminder of what happens when you trust LLM recommendations too much.
The real mistake was not just using Pagefind—it was accepting the recommendation without also asking for its downsides and tradeoffs. If I had looked at both the benefits and the weaknesses before committing, I might have avoided wasting that time.
There is also a broader lesson here: using a tool or framework you do not really understand is risky in itself. When choosing technology, it is not enough to ask why something is good. You also need to ask where it falls short, and whether those shortcomings matter for your actual use case.