Do We Still Need Taxonomies Now That We Have AI-Powered Search?

Taxonomies are used to keep research data organized and have played a key role in making research repositories searchable. But they do have their downsides. For example, they can require a lot of manual effort to create and maintain. This is an issue which AI-powered search, with its ability to process natural language and find data easily, seems to have resolved.

So now that artificial intelligence and modern AI systems are transforming how we search for and organize research, the big question is, do we still need taxonomies in user research?

For starters, it can feel intimidating to get started on creating a taxonomy since deciding on the categories and tags to use may not be clear-cut. As a result, it’s pretty easy to fall into the trap of trying to create a perfect taxonomy, which frankly isn’t possible.

As an organization evolves - perhaps developing different product lines, teams, and priorities - so does the taxonomy. Something that can be considered both a blessing and a curse.

On the one hand, there’s no need to try and create a perfect taxonomy. But on the other hand, a good taxonomy can steadily become a poor one if it isn’t updated and maintained according to changing needs. This makes taxonomies a tool that requires regular maintenance and oversight.

Adding tags and fields to your data is how you make it searchable. This is a task that has traditionally been done manually and can be rather time-consuming.

Additionally, as you can only fully make use of the data that has already been tagged, tagging requires a good amount of foresight and may require you to tag not only what you need for your current project, but also for future projects without knowing if they’ll ever actually be used.

To effectively use a taxonomy, it’s necessary to be familiar with the tags and fields and how to use them. This means that when introducing a taxonomy to your team, you need to teach them how and when to apply the various tags in the right way so that your taxonomy stays organized and effective.

It’s also likely that teaching your team about the taxonomy will be a task that needs to be repeated as your taxonomy develops and new members join the team.

AI-powered search, or semantic search, is changing the way that we interact with qualitative data by making it easier to locate the information that we need using search engine technology. This means we no longer need to manually tag data up-front in order to use it for future topics and projects. And the likelihood of “losing” data if it isn’t properly tagged beforehand is also much lower.

Here’s an example: Your product team wants to improve a feature on your app. Before AI, you would have needed to preemptively apply tags to relevant evidence in order to resurface and use it later. Otherwise, you’d need to go back through past research and put a lot of work into finding the data you need and tagging it to make use of it.

Using a keyword search would also likely result in critical evidence and valuable insights slipping through the cracks because synonyms and variations in language would be overlooked.

But now with the help of AI-powered search, you can run text-based searches, or find interesting and relevant examples based on a few concrete examples using features like Find similar Highlights.

AI search doesn’t just benefit the people who do research. Its ease of use makes it a great tool for helping stakeholders engage more with research!

It used to be the case that stakeholders needed to learn at least the basics of how a taxonomy worked in order to find what they needed. But now with the help of AI-powered search, all they need to do is type out what they’re looking for, and AI can direct them there.

Now having gone over the benefits of using AI to facilitate research and engagement, let’s revisit the big question.

If AI makes it so easy to interact with qualitative data, do we still need taxonomies at all?

In response, I’d pose a follow-up question that considers a different perspective: Can we fully rely on AI tools to help us find what we need in qualitative data?

I believe the answer is still no.

Let’s say you’re a product designer working on a new feature and want to keep track of a specific topic. In this case, it would still be better to use a tag to make sure that all the information you want is accounted for and easily accessible to you.

Here’s another scenario. You get some feedback from a customer call and realize that you have a key piece of evidence that you’d like to expound on. So you ask AI to find more related evidence and it comes up with 7 more. But only 3 are actually relevant. You keep those 3 and later recall another 1-2 points of evidence that AI didn’t come up with.

In this scenario as well, using a simple tag could be really helpful. By tagging the evidence that you know is important, especially the points that AI failed to recognize, you ensure that you’ll be able to find it again in the future. Whereas without a tag, it might be difficult to resurface again using AI-powered search since it didn’t recognize some of the evidence the first time around.

When used in this way, tags function as 'human-verified searches', signaling that a piece of evidence has undergone at least one round of human quality control.

Now, to be clear, having a human check AI’s work isn’t a guarantee for quality, but it is an additional layer of scrutiny that is much better than trusting the AI-generated content straightaway.

Tags also address another downside of AI that you’ve probably experienced yourself. When running the same search through AI multiple times, it can produce different results, especially as the data in your repository changes.

But by tagging the evidence that you know you’ll want to revisit, you ensure that it will be easily accessible when you need it again.

These are just a few of the ways tags and taxonomies continue to add value to the research process, demonstrating that the introduction of AI-powered search hasn’t made them obsolete. Rather, AI’s reshaped how they can best be used.

Now, instead of the former rigid way of applying predefined tag sets to all raw data, we seem to be moving toward a more flexible, organic approach where tags emerge as needed and can be selectively applied.