•
11 mins
AI is now genuinely useful for a first pass through qualitative data, and studies comparing it with human analysts find it can surface themes people miss. It also produces mostly descriptive themes, can quote words that are not in the transcript and cannot account for its own reasoning, which matters when you have to defend your method. NVivo still does what formal analysis needs, including a systematic codebook, queries across cases, memos, team coding and an audit trail, and Lumivero has now built AI into NVivo itself. For most researchers the best answer is both: AI to get oriented quickly, and NVivo, or a similar tool, for coding you can stand behind. Qualitative analysis used to start with weeks of reading and rereading. Now an AI tool can read a set of interviews and suggest themes in the time it takes to make lunch, and tools built for researchers, including Qualtranscribe's Smart Insights, go further with structured prompts for frameworks such as Braun and Clarke's thematic analysis. It is fair to ask whether dedicated analysis software, known in methods courses as CAQDAS (computer-assisted qualitative data analysis software), still earns its place, and the short answer is that AI and NVivo now do different jobs, so the job in front of you decides what you need.
What AI Coding Tools Do Well Now
AI is fast at the part of analysis that used to take the longest, which is getting familiar with the data. It can summarize each interview, suggest initial topics, pull out quotes that illustrate a pattern and answer questions such as "where do participants talk about cost?" across a transcript.
Research on this is still young, but it is encouraging for first-pass work. When Hamilton and colleagues (2023) compared themes from human coders and ChatGPT on interviews from a guaranteed income pilot, each found themes the other missed, and the authors concluded that the two approaches complemented each other. In a study published in Qualitative Health Research, Wachinger and colleagues (2024) found that ChatGPT's themes overlapped considerably with those of an experienced researcher, and that it produced a codebook and illustrative quotes that looked credible on first reading, which for exploration and quick internal summaries saves a great deal of time.
Where AI Coding Still Falls Short
The same study by Wachinger and colleagues also documented the problems. ChatGPT's themes were mostly descriptive rather than interpretive, some of the quotes it presented did not exist in the transcript, it edited quotes without flagging the changes, it forced the data into a framework that did not fit without noting the mismatch, and it could not reflect on its own biases.
Each of those matters for research that has to be defended. A committee or reviewer will ask how you developed your codes, why one theme outweighs another and where each quote came from. "The AI suggested it" is not a method you can write up, and a quote that does not exist in your data is a serious problem if it reaches a thesis or a client report.
Privacy is the other limit. A general-purpose chatbot is rarely covered by your IRB approval or your client's confidentiality agreement, and some AI platforms state in their terms that they use uploaded content to train their models. Before pasting participant data into any tool, check what its terms say and whether your protocol allows it. The guides to using AI transcription in IRB-approved research and the ethics of AI in academic transcription cover what to check.
What NVivo Still Does That a Chat Window Cannot
NVivo is built for the systematic part of analysis, where you apply a codebook consistently across every case, compare how codes behave across groups and keep a record of each decision. You build and refine a codebook and apply it across dozens of transcripts, with every coded passage traceable to its source. Queries show how often a code appears among different groups of participants or which codes tend to occur together, memos record your reasoning as it develops and later become the backbone of a methods chapter, and team projects let several researchers code the same data and compare their work. It also holds very different kinds of material in one project, from transcripts and PDFs to audio, video, survey data and literature, and for mixed methods studies codes can be linked to participant attributes and survey responses.
These features have been around for years, and they still matter because they make an analysis transparent and repeatable, which AI on its own struggles to do. Many departments and supervisors also still teach and expect CAQDAS as part of a qualitative method, so for a dissertation, using a recognized package is often the easiest way to show your committee how the analysis was done.
NVivo Now Has Its Own AI
The gap between the two has narrowed, because Lumivero, which makes NVivo, has built AI into the product. NVivo AI Cloud is a new browser-based version with an assistant called AURA that can summarize sources with citations back to the original text, suggest initial topics or work from your own framework, and run a "Devil's Advocate" mode that looks for disconfirming evidence and negative cases.
Lumivero says every AI action needs the researcher's approval before it becomes part of a project, that an exportable audit trail records what happened, and that research data is not used to train AI models. NVivo 15, the desktop version, remains a separate, license-based product with its own AI Assistant, which can summarize text, documents and coded content, suggest child codes for an existing code and autocode sentiment. Lumivero says it has an enterprise agreement with OpenAI that prohibits using data uploaded from NVivo for training, with uploaded data deleted once each request is complete, and administrators can switch the AI features off. According to Lumivero, AI Cloud is currently available to individual subscribers, with team and organization access expected in early 2027, and pricing depends on the license type, so check the current options on Lumivero's site. That leaves a more practical decision about where AI belongs in your process and which tool keeps the record of your decisions.
ATLAS.ti and MAXQDA Have Added AI Too
NVivo is not the only CAQDAS package to change. ATLAS.ti's Intentional AI Coding, currently in beta, asks you to describe your research intentions and concepts of interest, then suggests research questions and codes that you review and refine before applying. It uses OpenAI's technology, and an AI Privacy Mode can switch the OpenAI integration off entirely, with administrators of multi-user licenses able to enforce that setting.
MAXQDA's AI Assist is an add-on that offers summaries, code and subcode suggestions, AI coding and chat, with a free version limited to single documents and a premium version that codes across documents and can run AI models locally on your own computer or server. MAXQDA says data is processed with zero retention and stored encrypted in the EU.
Across all three, the AI suggests and the researcher decides, with the software keeping a record of what was coded and why, which is the part a general chatbot does not provide.
A Workflow That Uses Both
Stage | What happens | Tool |
|---|---|---|
1. Transcripts | Turn recordings into accurate, speaker-labelled transcripts | Human transcription for data you will code formally, or Instant Draft for a fast AI draft |
2. First read | Summaries, initial themes and key quotes to get oriented | Smart Insights or NVivo's own AI |
3. Formal coding | Build and apply a codebook, write memos, run queries | NVivo or another analysis package |
4. Write-up | Report themes with verified quotes and a documented method | Your analysis software, with quotes checked against the recording |
Smart Insights fits stage 2. It offers 17 research prompts, including Braun and Clarke thematic analysis, along with AI chat and color-coded notes, and its findings link to timestamps in the transcript, so you can click any quote and hear the moment it came from. That link back to the source is the practical answer to the invented-quote problem, because every quote can be checked in seconds before it goes anywhere.
Qualtranscribe transcripts export ready for NVivo, ATLAS.ti and MAXQDA, and the NVivo import guide shows how to move from transcript to coding without reformatting. For the coding itself, the guides to thematic analysis and analyzing interview transcripts walk through the process step by step.
Who Still Needs NVivo, and Who Might Not
You are... | Do you still need NVivo? |
|---|---|
A PhD student or master's student writing a thesis | Usually yes. Committees expect a documented, systematic coding process, and NVivo or a similar package is the easiest way to show one. |
A funded academic team | Yes. Team coding, audit trails and consistent codebooks across researchers are hard to manage without it. |
A mixed methods researcher | Yes, for linking qualitative codes with participant attributes and survey data. |
A market research agency producing a topline | Not always. AI summaries and verified quotes can carry a quick-turnaround topline, with formal coding saved for larger studies. |
A solo consultant running a few interviews | Possibly not. AI-assisted analysis on accurate transcripts may be enough, provided every quote is checked against the recording. |
If your analysis only has to inform a decision, AI plus careful checking may be all you need. If it has to survive a viva, a peer reviewer or an audit, keep a tool that documents how you got there.
The Transcript Still Decides How Good Any Analysis Can Be
AI coding and NVivo coding share one weakness, which is that both can only work with what is in the transcript. Misattributed speakers, misspelled names and words the transcription got wrong flow straight into your codes and quotes. In a test of seven AI transcription tools on a real focus group, two tools gave no usable speaker labels and every tool got at least one participant's name wrong.
For data you plan to code formally or quote in a thesis, report or publication, accurate speaker-labelled transcripts are the foundation. Qualtranscribe's human transcription starts at $1.20 a minute for English, with focus groups at $1.70 a minute, and Instant Draft is free for 75 minutes a month for researchers who want to try AI-assisted analysis first.
Frequently Asked Questions
Can AI replace NVivo for qualitative analysis? Not entirely. AI is useful for summaries, initial themes and finding quotes, but studies have found it produces mostly descriptive themes and can present quotes that are not in the data. NVivo remains better for systematic coding, team projects and a documented audit trail.
What is CAQDAS? CAQDAS stands for computer-assisted qualitative data analysis software, the category that includes NVivo, ATLAS.ti and MAXQDA. These packages help researchers organize, code and query qualitative data while keeping a record of the analysis.
Does NVivo have AI features? Yes. NVivo AI Cloud includes an AI assistant called AURA that summarizes sources, suggests topics and looks for disconfirming evidence, with researcher approval required for each AI action. NVivo 15 for desktop has an AI Assistant that summarizes content, suggests child codes and autocodes sentiment.
Is it acceptable to use AI for coding in a dissertation? It depends on your institution, your committee and your IRB approval. Many researchers use AI for early exploration and then code formally in software such as NVivo, describing both steps in their methods chapter. Check your university's guidance before using AI with participant data.
How do I stop AI from inventing quotes? Check every quote against the transcript and, ideally, the recording before using it. Tools that link each finding to its timestamp, such as Smart Insights, make that check quick.
Can I import Qualtranscribe transcripts into NVivo? Yes. Transcripts are delivered ready for NVivo, ATLAS.ti and MAXQDA, with speaker labels and timestamps, and NVivo export is included in Instant Draft from the Pro plan.
Turn your recordings into analysis-ready transcripts.
Human Transcription
Clean verbatim and full verbatim transcripts, delivered by specialist transcriptionists
AI Transcription
Instant Draft powered by AI, with Smart Insights for analysis-ready output
Translation Services
Accurate translation across 99+ languages for multilingual research workflows
Keep reading
Related articles
The Ultimate Guide to Academic Transcription Services: Everything You Need to Know
You have just finished conducting 15 in-depth interviews for your dissertation. The conversations were rich, the participants were candid, and now you are staring at 22 hours of recordings wondering how any of it becomes analyzable data. This is where most researchers hit a wall, because transcribing it yourself can easily take four hours or more for every hour of recording, which means weeks of typing before you touch a single line of analysis. A professional service is the obvious way out, but the market is crowded, compliance requirements are real, and the wrong choice can leave you with anything from a sloppy transcript to a problem with your IRB.
Read article
AI Transcription Tools for Market Research Interviews and Focus Groups (2026): 7 Tools Tested
Market researchers record more IDIs, TDIs and focus groups than ever, and most of that fieldwork now passes through AI transcription before anyone codes a word. Feature pages all promise high accuracy, but a real group session, with six voices, crosstalk and names the software has never heard, is a different test. This guide puts seven popular tools through the same focus group recording and scores what matters to researchers: who said what, whether names survive, and which mistakes would end up as the wrong verbatim in a client report.
Read article
In-Depth Interviews vs. Focus Groups: Which Method Fits Your Research Question?
Brand teams often book focus groups because the last project used them, and academic researchers often default to interviews because their supervisor did. Those habits work most of the time, but they also produce studies where the method quietly works against the question, like a concept test run as twelve separate interviews where nobody reacts to anybody, or a study of a private health decision held in a room of strangers.
Read article
© 2026 Qualtranscribe LLC. Services Provided Globally