Can AI take my job? (artificial intelligence)

The role of AI in processing qualitative research (artificial intelligence)

Some professionals have wondered whether the arrival of AI in research could lead to the full automation of research. After all, LLM-based tools are advancing month by month, or even week by week. By now, they are not only capable of summarising, identifying patterns or carrying out comprehensive analysis; they can even take over moderating conversations from the researcher. This raises the question of whether there will be any research tasks left for humans to carry out in the future. (artificial intelligence) (large language model)

However, as another group of researchers gain more experience with LLM software and AI-supported research platforms, they are seeing more clearly what these tools can help with and how they can make work more efficient. They are also seeing more clearly what their limitations are, and which questions and topics cannot be meaningfully answered without human involvement. (large language model) (artificial intelligence)

Yet another group of professionals is becoming increasingly convinced that we need not worry: research will not become automated. Instead, it will become more valuable, as AI becomes increasingly able to support this work too. It is also becoming ever clearer that the work can only be complete with us, the researchers; only our expertise can deliver truly good results. This is also the position we take here at Works. (artificial intelligence)

In this article, I focus on presenting the benefits of using artificial intelligence during the analysis phase of research.

We have always sought to respond to our clients’ needs with the greatest possible care. After all, if they are satisfied, trust us and receive the right support from us in resolving their business dilemmas, they will turn to us again next time.

Our experimentation with AI is also driven by client needs. I will show which effective, machine-supported solutions appear to be delivering results at present, in January 2026. (artificial intelligence)

Client need #1: I want the results by next week, too!

Accurate transcripts, quickly

The most time-consuming step in qualitative research is producing suitable transcripts and notes from interviews. We know — and many of us find it hard to accept — that there is only one way to do this: taking detailed notes while listening back to the interviews. This often takes two or three times as long as the interview itself. After that, all the remaining steps are shorter and more enjoyable: examining patterns, identifying and categorising problems, understanding causes and effects, formulating recommendations and preparing the presentation.

We have been planning to automate the least-loved step mentioned above for a long time. We use Teams to conduct interviews, and quickly realised that although it does produce a transcript, it unfortunately requires a great deal of human correction before we can use it effectively for synthesis and analysis. Since the rise of generative AI, we have therefore been exploring tools designed to produce accurate transcripts, and last year we finally found one or more that provide suitable quality in Hungarian too. Comparing them is made possible by a method called WER — Word Error Rate — which involves counting misunderstood, omitted and incorrectly inserted words, then dividing their total by the number of correctly identified words. We achieved a rate below 15% with these tools, which is considered a particularly good result, whereas Teams produced results between 50% and 60%. (artificial intelligence)

So, with minimal human time investment, we can have an excellent transcript. We can save a great deal of time at this stage with Transkriptor or Turboscribe.

Finding quotes no longer takes hours in the final push to produce a presentation

There is another step that is fairly slow and, on a rushed project, sometimes gets left out altogether: adding suitable quotes to the presentation for each insight. You might think accurate transcripts make these easier to find again, but in reality they do not. The moderator will certainly remember which participant expressed an idea that strongly supports our point, but locating it in a 30–40-page transcript — and knowing exactly what phrase to ‘Ctrl+F’ — is more difficult. NotebookLM has sped up this step for us. When asked the right questions about the transcripts, it immediately produces the relevant quotes without hallucinating, as it only works with the uploaded texts.

Client need #2: I want even higher-quality results!

To assess how AI can deliver on this, we have introduced a new experimental way of working. Once the transcripts are ready, the research team splits into two: one group analyses them in the traditional way, while the other uses various AI tools. This is the only way to judge the quality of support AI can provide. (artificial intelligence)

Let’s look at how the AI-supported team works: (artificial intelligence)

They use LLM-based chatbots, which can provide valuable support for thinking and analysis when used well. The key is prompting and providing context. (large language model)

Effective prompting and context-setting are essential

When talking to AI chatbots, establish the context too: provide the business objective, research objective and hypotheses, then begin the conversation. After all, if your ‘conversation partner’ does not understand the parameters, it will not be able to help appropriately. (artificial intelligence)

Use effective prompts: be specific, define the research questions or the sub-questions whose answers can move you towards the results in the right direction. For example, do not ask it to summarise the conversation, nor ask which participant highlighted what when using a given feature. Instead, ask what similarities and differences emerged in each participant’s thoughts about that feature. It is also worth adding details based on your own interview experience: for example, ask it to separate thoughts about the service from those about usability, or to focus exclusively on the analytical considerations needed for the given project.

And who could be better at prompting and setting context than a qualitative researcher?

For us, prompting and creating context did not enter our lives with AI. In every qualitative methodology where flexibility is necessary for good results, and paying attention to our conversation partner may raise new questions at any time, we need to ask those questions. At the right time, with the right question. So we prompt regularly in our interviews, and this knowledge has been highly useful in conversations with AI chatbots. (artificial intelligence)

In research, the context we provide at the beginning of most conversations with participants is of particular importance. If it is not right, the participant may approach, for example, their understanding of a prototype concept from a completely different perspective. For instance, if we use dummy prices in a prototype while testing a purchasing journey, it is vital to clarify the parameters for the participant in advance, so they can focus on ergonomics rather than the prices.

Researchers consciously develop their ability to prompt and set context in their early years, before it becomes instinctive. These are the skills we need to apply when using generative AI-based tools. (artificial intelligence)

Some tools immediately suggest a new approach or next step. Do not let a good suggestion carry you away in the conversation. Just make a note of it, but first work through the questions according to your own approach. It is important that when asking questions, just as when we interview participants, we do not lead the machine either. Do not ask it to compile difficulties and problems, because then that is what it will generate, amplifying things that human analysis would not.

Using research platforms

Companies have long had access to numerous research-support platforms that could already provide good support for human analysis. However, over the past two years, almost all of them have added some kind of AI-supported feature, while new platforms promising near-miraculous results have also emerged. Naturally, we continue to experiment with these too, because comparing them lets us see where each one excels and how it can properly support our or our clients’ day-to-day work. (artificial intelligence)

Over recent months, I have been able to group the platforms we reviewed and tried as follows. What they had in common: a great landing page that makes you want to start using them, but… There is always a but; in this case, the following:

• But the platform does not properly support qualitative research, instead enabling other types of voice-of-customer processing. These include Kraftful, Collectif, Notably, Attention Insight
• But it is only available after an online demo. For example: Yabble, Dovetail.
• But it does not speak Hungarian, like Askviable, Innerview or Viable.

And after all this, a few platforms remained that we could genuinely start using, and have enjoyed working with ever since, where the project allows.

Doreveal

It not only supports human analysis with tagging, organising by tags, and the ability to highlight quotations from transcripts and videos; after video files are uploaded, it can also summarise the key statements from each interview according to the research questions, answering the relevant question in each case. With one click, I can access the related video or transcript excerpts. This provides substantial support and a quick overview during the first stages of synthesis. In other words, there is no need for a transcript: it can already generate findings from the video file itself. Hungarian remains a challenge with this software, because although it can accept Hungarian-language inputs (briefs, guides, interviews and research questions), it gives its responses in English whenever we ask for help. This does require some compromise, but otherwise it supports us like a good research partner.

Marvin

For a long time, this platform did not speak Hungarian either, but it can now support the work as a fully fledged research partner. What is particularly useful is that it not only summarises according to the research questions provided in advance, but can also answer specific questions, with references to the relevant interviews. Not only is an AI summary automatically generated for each interview, but AI notes can also be requested. This can be an enormous help if anyone outside the research team, from the business team or management, wants to gain an insight. If a team member can devote an hour to the research, this feature is more useful than listening to just one interview, as it can provide a far more complete picture of participants’ experiences. (artificial intelligence)

Client need #3: Could you understand us even better?

Based on feedback from most of our clients, we can say that clients like working with us because we understand them, their industry and their problems well, and can formulate commercially relevant recommendations to help them develop further. But this can always be done even better.

Market context

Even if our domain knowledge is strong, changes in regulation or competitors’ activities may not immediately come into view. In this area too, asking an AI chatbot for help is useful. In such cases, there is no need to ask our client to establish the current market situation; we can simply ask the machine, and only consult the client if something still remains unclear. (artificial intelligence)

Benchmark analyses

It has never been this easy. Of course, with persistent searching we could previously find the services and websites that could be reviewed as benchmarks or best practices: examples to follow or avoid. But a typical problem was that finding these solutions, especially international ones, took far more time than finding participants who used them. And it was easy enough for those participants to show us in detail the solution we ourselves saw as a good example.

Generative AI can now review pages it considers relevant much sooner and list them for us. Naturally, we experts still need to verify them in the familiar way afterwards. It can save days. (artificial intelligence)

Previous research and publications

We often encounter a new product, service or feature idea for which we do not have adequate information about potential users’ needs, and the project’s scope does not allow us — whether in time or budget — to establish through research whether the direction is genuinely right. As with benchmark analyses, we could previously search for these using Google, but finding relevant information in these searches was perhaps even harder than simply looking for a service or product. Today, however, we have access to an AI-supported solution that searches only research and publications considered sound by experts too (peer-reviewed), and provides useful results for any question, along with a short summary and access to the full document. Even for a completely new product idea, feel free to turn to it and ask whether it can find any research material on the subject. If it can, you will gain information that only a few years ago might have required research costing millions of forints. It would be a mistake to miss it! (artificial intelligence)

Change is the only constant

At the beginning of 2026, we see these as the client needs for which artificial intelligence supports us as a valuable partner. Week by week, however, we see how much more people are able to get out of AI, how significantly the tools are developing, and how they enable us to develop too. This year, we will continue working to keep pace with the development of these tools. (artificial intelligence)

We still cannot clearly see everything we will be able to use it for in the future, nor what it will not be suitable for, however hard we try. But we are following these opportunities closely, and will continue to share our knowledge and experience with you regularly.

This article was written by Bartha Gizella, Senior Experience Designer at Works.