Why We Removed Auto-Tagging

Auto-tagging arrived in Quillfold 2.2 and left in 2.6. Removing a feature we built ourselves is never pleasant, and this one was popular on the day it launched. Here is why it went anyway.

What it did

When you saved a note, Quillfold read it and suggested up to three tags. If you ignored the suggestions after a few seconds, it applied them for you. The idea was that a notebook should organise itself, so that nobody had to remember to tag anything.

It worked as designed. That's the uncomfortable part. The suggestions were reasonable, the matching was quick, and almost nobody reported a bug.

What we saw in the numbers

Six months after launch, Linnea pulled the figures. Notebooks with auto-tagging switched on had, on average, more than four times as many distinct tags as notebooks without it. That sounds like a feature doing its job. Then she looked at how many of those tags were ever used to find something.

Roughly one in twelve. The rest sat in the sidebar, untouched, making the tags that people did care about harder to see. A notebook with two hundred tags isn't organised. It's a list.

We also looked at renames. People who turned auto-tagging off typically renamed or merged a handful of tags in their first week, which is a sign of someone shaping a vocabulary. People who left it on almost never did. The tool had decided the vocabulary for them, and they'd stopped thinking about it.

What people told us

The numbers pointed one way, but we didn't want to decide from a spreadsheet alone, so Ravi emailed forty users who'd kept it enabled for a year.

Eleven replied. Most said some version of the same thing: the suggestions were nice at first, then the tag list became something they scrolled past. Two said they loved it and would be sad to lose it. We took that seriously. They deserve a real answer, which is further down.

The decision

We weighed three options.

1. Keep it as it was and add a cleanup screen. Rejected, because it hands people a chore created by our own feature.

2. Make it opt-in and hide it. Rejected, because a feature that most people never see is a feature we still maintain and still describe in the docs.

3. Remove it and replace it with something smaller.

We chose the third. Removal meant an announcement, a changelog entry, and about three weeks of notes in the release candidates, so nobody found out by surprise. It also meant leaving existing tags exactly where they were. Nothing you'd already accepted was touched.

What replaced it

Quillfold now suggests a tag when you start typing one, from the tags you already have. It never creates anything. If the tag you type doesn't exist yet, you see a plain "New tag" row and make that choice yourself.

It's less clever, and that's deliberate. A suggestion that can only point at things you've already decided is hard to turn into clutter.

If you relied on the old behaviour, 2.6 includes a one-time export of the tags auto-tagging had applied, with the note titles beside them, as a plain text file. You can keep it, search it, or ignore it. We'd rather give you your data than argue that the feature was wrong for you in particular.

We think removing it was right. We also think it was a real feature that real people liked, and both of those are true at once.