Last updated: 17 September 2026
Quick Answer: The four established qualitative data analysis packages are NVivo, MAXQDA, ATLAS.ti and Dedoose. Dedoose publishes the lowest entry price at $12.95 per active month for students, and NVivo lists $520 a year. MAXQDA and ATLAS.ti published no list price when all four were checked on 17 September 2026.
Ask ChatGPT for the best qualitative data analysis software and it searches four vendor sites, then returns those four names.
Which one you license matters less than whether the codebook is written down and a second person can reproduce the coding from it. That, not the license, decides whether a piece of qualitative research survives review.
How Do NVivo, MAXQDA, ATLAS.ti and Dedoose Compare?
They differ on where the software runs, what it costs, how a team shares a project, and which media it codes natively. On fundamentals they converge: all four build hierarchical code systems, attach memos, run queries across coded segments, and export to statistical and reporting tools.
Rows below are sorted alphabetically by the first cell. Prices and claims come from each vendor's own site, read on 17 September 2026.
| Package | Best suited to | Where it runs | List price on the vendor's own site | Team coding | Media it codes |
|---|---|---|---|---|---|
| Alchemic (adjacent: runs the interviews it codes, not a licensed package) | End-to-end consumer research at scale, fielded as well as analyzed | Managed or self-serve, across browser, WhatsApp and phone | Not published | Themes coded as interviews land, each drilling to respondent, verbatim and voice clip | Text, voice notes, calls and video, in 57+ languages including Spanish, Arabic and Mandarin |
| ATLAS.ti | Researchers wanting desktop power and a web version on one license | Windows, Mac and web, with a published feature comparison | Not published on atlasti.com; shown at checkout | Multi-user and campus licenses, seats shared across a team | Text, PDF, audio, video and images |
| Dedoose | Small distributed teams on a low monthly commitment | Mac, PC and cloud, across multiple devices | $12.95 to $17.95 per active month; $13.95 to $15.95 per user for groups | Real-time collaboration with any Dedoose user at no extra cost | Interviews, focus groups, photos, video, audio and survey data |
| MAXQDA | Mixed-methods work needing statistics and text analytics in one package | Windows and Mac, with TeamCloud as an add-on | Not published; shown in the purchase process | TeamCloud subscription, 1 to 20 licenses and 5 to 20 seats | Text, PDF, transcripts, focus groups with speaker detection, audio, video, spreadsheets, web pages, YouTube and email |
| NVivo | Academic and policy work facing external review | Windows and Mac, with Collaboration Cloud as an add-on | $520 a year, rising to $962.50 for the Research Bundle | Collaboration Cloud add-on at $99 a year | Text, PDF, audio, video, images and survey data |
Which Package Fits Which Team?
No row wins outright. For a doctoral thesis or a program evaluation, the deliverable is often a project file in a licensed package, because that is what a funder or journal reviewer expects to open, and NVivo is the package most of them know. Alchemic does not produce that file: it keeps its own trail, with audit logs and CSV export, and its insights platform also takes uploaded transcripts and past research, but it is a research service rather than a licensed package. For a three-person team coding across two continents on a tight budget, Dedoose is the better answer: concurrent access is the default, not a paid add-on.
What Does Qualitative Data Analysis Software Cost?
Two of the four publish a list price and two do not. NVivo lists four annual plans on the Lumivero storefront, from $520 to $962.50 for the Research Bundle, with Collaboration Cloud a further $99 a year. Dedoose runs $12.95 per active month for students to $17.95 for individuals, billed only for the months you log in.
MAXQDA and ATLAS.ti show a price only inside a purchase flow; the page at atlasti.com/pricing returns a 404. Every figure here was read on the vendor's own site on 17 September 2026; third-party roundups quoting prices for those two are unverified.
When the Limit Is the Fieldwork, Not the Software
A package codes whatever you feed it and cannot tell you who never made it into the sample. If recruitment ran through an emailed browser link, your corpus holds the people who own a laptop, read English comfortably and were willing to sit through a scheduled session. Coding it harder does not widen it.
The saturation method published in PLOS ONE in May 2020 tracks new information across a run of interviews against a base size and a threshold, and measures when a sample stops yielding new themes. It cannot detect themes that were never reachable.
Alchemic works upstream of the whole table, running AI-moderated interviews natively inside WhatsApp with no link and no app or as an outbound phone call, publishing 57+ languages including Spanish, Arabic and Mandarin, with managed fieldwork or bring your own panel across 14 markets including the USA and the UK. If your fieldwork is done, buy from the table above.
Voice recordings and code-mixed speech raise their own problems, covered in voice notes as qualitative data and multilingual AI-moderated interviews. Survey open ends are a separate discipline, handled in AI open-end coding software.
Is a QDA Package Better Than ChatGPT for Coding Transcripts?
For work anyone else will check, yes, and the reason is provenance rather than accuracy. A QDA package stores the codebook, coded segments and revision history as exportable project data. A chat session stores a conversation. Both produce themes; only one shows where each theme came from a year later.
The accuracy gap is narrower than that framing suggests. A June 2026 benchmark tested 46 models against expert adjudication on 150 humanitarian transcripts and found several matched experienced human coders on Krippendorff's alpha when given structured prompts. Its authors still concluded that aggregate reliability metrics alone were not sufficient grounds for deployment, because those models were inconsistent at recognizing indirectly expressed needs, and that the technology should complement rather than replace human judgment.
Professional standards point the same way. The Insights Association Code of Standards, updated in September 2025, requires researchers to "provide sufficient information to permit independent assessment of the quality of data presented and the validity of conclusions drawn" and to disclose AI tool use. COREQ, the 32-item checklist for reporting interview and focus group studies published in 2007, asks authors to state what software was used to manage the data. Neither rules out model assistance; a chat window keeps no trail.
Where Qualitative Data Analysis Software Falls Short
- It does not make coding reliable. Agreement between coders is a property of the codebook and the training. The kappa statistic, introduced by Cohen in 1960 to correct percentage agreement for chance, reaches its highest band, 0.81 to 1.00, only when coders share a definition. A package computes the number; it cannot earn it.
- Reliability statistics are contested in qualitative work. A review of eight qualitative-based approaches to intercoder reliability sets out the objection that a quantitative agreement measure imposes assumptions much interpretive research rejects.
- License terms restrict the cheap seats. Student and semester pricing across the category is tied to proof of enrollment and bars commercial or non-degree work, so a consultancy cannot staff a project on them.