Disclaimer: this post contains rampant anthropomorphising of AI, because it’s a useful mental model. I do not think that AI can currently think, or collaborate, or partner. Still, indulge me, it’s a good metaphor.

Disclaimer, the second: I’m going to say AI a lot. Mostly I am referring to Large Language Models, either through chat interfaces or in research tools powered by LLMs via an API. 

The flattening temptation

I’ve been messing with the ideas in this for a few weeks, but my former colleague, Lisa Koeman, wrote a great piece a few weeks ago on the “flattening” of user research, which has inspired me to actually get stuff into words. She argued that in our industry’s relentless pursuit of efficiency, we’ve adopted tools and practices that strip away the things that make our work valuable: context, nuance, and humanness. From remote testing to auto-transcription, each step towards speed has chipped away at depth and understanding.

AI has entered the chat, and we’re almost exclusively talking about it in the same way; use tool, go faster. There’s products offering auto-generated summaries, instant themes, and one-shot reports. I can see the temptation to embrace this efficiency entirely; finally it can be easy to keep pace with the demands of delivery.

The question Lisa posed was: “where is the ambition to enrich our research practice?” She offers excellent examples, in coaching up moderation skills, video diary studies, and exchanging knowledge between disciplines. I agree with all of these, but (because of my current responsibilities) I am mostly thinking about what the “right now” version looks like.

I’ve been coaching very senior researchers in AI use in the last few weeks, and I find the mental models so interesting. By default, we expect tools to do a job, so we ask AI to do a job. We don’t expect or ask for our tools to make our thinking better, so we don’t ask AI to do that. The results are often half-baked and frustrating.

If we approach it with curiosity instead, we can learn to use it as a collaborator that helps us go deeper, think sharper, and increase the value and impact of our research.

If we don’t shift our mental model we’re going to end up automating away our thinking, the bit that makes our work matter. It doesn’t have to be this way.

The need for speed, the price of shallowness

Delivery pressure is real. Research can feel slow. Research can be slow. We are very often seen as a bottleneck rather than strategically valuable. I’ve felt that tension plenty of times, and had to make all the same tradeoffs everyone else has. The risk of being seen as a ‘nice to have’ is not imaginary.

So, we can’t reject efficiency. We’ll always need to make pragmatic calls about doing just enough research, working at the pace of delivery, and picking our moments to push back against the riskiest assumptions. But we do know the price of moving too fast.

Impactful insights in research rarely come from a quick scan of the data. They often emerge from deep understanding. They come from letting the data “simmer” or “compost,” as Lisa puts it; from spending hours with transcripts and observations until your mind makes a leap and connects seemingly disparate interpretations. This is “soaking in” time. It’s when I decide what I think. It’s when we build a deep understanding of our participants’ human experience.

When you lose that depth, you lose the ability to defend your findings with confidence. You miss subtle cues, emotions, and things that point to the biggest opportunities. Your findings become shallow. This is the dilemma: we need to maintain our rigour to be valuable, but we are pressured to be faster than that rigour allows.

 The question then isn’t speed versus depth. It’s whether we’re willing to find ways to do both.

A practical framework for deeper research with AI

So how do we do it? The answer lies in shifting our mental model. I liked Kaleb Loosbrock’s dance partner analogy, although I didn’t think he went quite far enough. We need to avoid seeing AI as a vending machine for insights and research plans, and start treating it as a research partner. It can expand our thinking, but we need to lead it. This is not about offloading thought; it’s about making our own thinking better. Hopefully you’ll also save time, but that can’t be the whole point.

Here is a practical framework I’ve been experimenting with for what that can look like, applied to analysis. I’ve started there because it’s the home of the biggest trade offs right now, but these apply as principles across the research lifecycle:

  • Start with your own expertise. The most important step is to do your homework first. Before you think about involving AI in analysis, immerse yourself in the data. Read the transcripts, watch the videos, and form your own initial thoughts and hypotheses. Get your teams’ thoughts too, it’s still a team sport. You need to develop context and confidence in your material. To hand raw data to an AI without this grounding is to completely lose control over what you think. 
  • Use AI as a critical friend. Once you have your own perspective, you can bring AI in. This is where the real value lies. Treat it as a tireless, infinitely patient colleague. Ask it to challenge you. In some of my own recent experiments, I’ve found a conversational method using standard chatbot interfaces to be more powerful than professional research tools because I have more control (this seems incredibly backward)
  • Make advanced methods accessible. Many powerful qualitative analysis methods are rarely used in industry because they are too time-consuming. AI can change this, to an extent. It can act as a scaffold, making these approaches practical. For example, you can guide it step-by-step through a grounded theory analysis on a specific question, or ask it to perform a light-touch discourse analysis to reveal what language shows about users and their experiences. This doesn’t replace the need for skill, but it dramatically lowers the barrier to entry for deeper forms of inquiry.
  • Use creativity. After you’ve done the hard work of analysis and synthesis, AI can be a brilliant partner for communication. Once your insights are solid, you can use it to explore creative ways to frame your story, draft different narrative versions for different audiences and stakeholders, or generate richer artefacts that bring your findings to life.

How to keep the researcher in control

This collaborative approach has challenges. If we aren’t careful, we can easily fall into very shallow integration. The risk of churning out bland findings that aren’t actionable is significant. Likewise, the confident, articulate way that AI presents information can mask biases or complete fabrications.

Keeping the researcher in control is key. Here’s how:

  • Practice reflexivity. Qualitative research has always required us to acknowledge the biased lens we bring to our work. AI makes this non-negotiable. We need to be explicit about our biases and use AI to help us see beyond them, not to reinforce them.
  • Reject black boxes. Many bespoke AI research tools completely hide their workings. They present a cluster or a theme without showing you how they got there. We need to use (or build?) tools that start with transparency and the flexibility to guide the analytical methods the AI uses. If you can’t see the connection from the data to the insight, the insight is worthless. Chain of thought reasoning AI models can help here, but they’re not infallible.
  • You are the lead. Remember you are the senior partner in this relationship. AI is a powerful tool, but it has no real-world experience, no context beyond what you give it, and often appalling taste. You need to guide it. Your thinking and judgment are what transform its output from mere word patterns into meaningful insight.

Demanding better from our tools and ourselves

The current market for AI research tools feels distinctly unimaginative. It is built around saving time and automating tasks. It’s a low bar. It’s a vision of minimum viable effort, not maximum viable insight.

We need to be more ambitious. We should demand tools that are designed for partnership, not just process automation. Whatever we use, we need to use AI in ways that are transparent, flexible, and built to enhance a researcher’s critical thinking, not replace it.

Make a case for good things

The arrival of powerful AI presents researchers, and the people who pay us, with a choice about the kind of work we should do.

We can treat it as an automation engine, a tool for simply doing our jobs faster. This approach promises efficiency, but at the cost of our own thinking. It risks shrinking our role to that of a machine operator, eroding the critical judgment that makes our work valuable. And eventually, machines don’t need operators.

Or, we try a different relationship with this technology. We can treat it as a collaborator; a partner that, when directed with skill and curiosity, can help us achieve more. This is the opportunity to be more thoughtful and more impactful, not just more productive.

This second approach is not easy. It requires new skills, a critical mindset, and a real ambition for what our practice can be. But it is the one that leads to a richer, more human, and ultimately more valuable kind of research. If we don’t make a case for that, nobody else will.