Framework Analysis With AI: A 2026 Guide
Qualitati Research Team · 2026-07-02 · 12 min read
Last updated: July 2, 2026
Short answer
Framework analysis (the Framework Method) is a structured qualitative approach that charts coded data into a matrix — cases in rows, themes in columns — so researchers can compare within and across participants. To run it with AI in 2026, build an analytical framework from early transcripts, apply it deductively with an LLM, chart summaries into the matrix, and keep a human researcher interpreting the cells. AI speeds indexing and charting; it does not replace the interpretation stage.
Key takeaways
- Framework analysis was developed by Jane Ritchie and Liz Spencer at Britain's National Centre for Social Research for applied policy work, and is now standard in health, implementation, and evaluation research (Gale et al., BMC Medical Research Methodology, 2013).
- Its signature move is the framework matrix: a spreadsheet of rows (cases) by columns (codes/themes), with summarized data in each cell — making cross-case comparison visual and auditable.
- The method runs in five stages: familiarization, framework development, indexing, charting, and mapping/interpretation.
- AI is strongest at the indexing and charting stages — applying an agreed framework and summarizing data into cells — and weakest at framework development and final interpretation, which stay human-led.
- A 2026 comparison in the International Journal of Qualitative Methods found LLM-assisted analysis can approximate researcher-interpreted results but still needs substantial human oversight for validity (Misra et al., IJQM, 2026).
- Use the Framework Analysis Readiness Checklist below before you trust an AI-charted matrix for reporting.
What is framework analysis?
Framework analysis is a qualitative data analysis method that organizes coded material into a structured matrix so patterns can be compared systematically across cases and themes. Unlike reflexive thematic analysis, which resists rigid procedure, the Framework Method is deliberately stepwise and highly visible — every analytic decision is recorded in the framework and the matrix, which makes it well suited to team research and to studies that must be auditable.
The approach originated in the late 1980s for UK applied social-policy research and has since spread across health services research, implementation science, education, and program evaluation because it handles both a priori questions (from an interview guide or policy brief) and emergent themes from the data (Stalmeijer et al., 2023).
Framework analysis vs thematic analysis
- Thematic analysis produces themes but does not prescribe a matrix; comparison across participants is left to the analyst.
- Framework analysis forces every case and theme into a shared grid, so gaps, outliers, and cross-case contrasts are visible at a glance.
- Framework analysis is often preferred when you have defined questions, multiple coders, and a need for transparency — for example in health and evaluation work.
The five stages of the Framework Method (with AI)
1. Familiarization
Read a subset of transcripts closely and note early impressions. AI can produce per-transcript summaries and candidate topics to orient you, but do not skip your own reading — familiarization is where you form the questions the matrix will later answer. Treat AI summaries as a reading aid, not a substitute for immersion.
2. Developing the analytical framework
Agree a set of codes and group them into categories — the framework. This is a human-led, judgment-heavy stage. Seed deductive codes from your research questions and interview guide, then let a few inductive codes surface from early transcripts. AI can propose candidate groupings, but the research team owns the framework because it encodes what the study is actually about.
3. Indexing
Apply the framework systematically to every transcript, labeling segments with codes. This is deductive coding — the stage where LLMs are most reliable, provided the framework has clear definitions. Published work shows deductive coding with a clean codebook can approach human agreement, while interpretive categories remain harder (JMIR AI, 2025). Always spot-check AI indexing against the definitions.
4. Charting
Summarize the indexed data into the matrix: one row per case, one column per theme, a concise summary (with source references) in each cell. Charting is where AI saves the most time — condensing coded passages into faithful cell summaries — but the summaries must stay anchored to real quotes, not paraphrase away the meaning. Keep a link from each cell back to the underlying transcript lines.
5. Mapping and interpretation
Read across and down the matrix to find patterns, contrasts, and explanations. This final stage is human territory. AI can surface candidate patterns ("cases in column B cluster on X"), but deciding what those patterns mean — and what the study concludes — is the researcher's responsibility. Over-trusting AI here is the most common way framework analysis goes wrong.
The framework matrix, explained
The matrix is the deliverable that distinguishes this method. A simplified example:
| Case | Barriers to adoption | Perceived benefits | Support needed |
|---|
| P01 | Time cost; low trust in output | Faster first-pass coding | Training; audit trail |
| P02 | Data privacy concerns | Consistency across coders | On-device option |
| P03 | Learning curve | Handles multilingual data | Templates; examples |
Read down a column to compare all cases on one theme; read across a row to understand one case holistically. Each cell should carry a pointer back to its source text so any claim can be traced — the same audit discipline funders and journals increasingly expect.
Where AI helps — and where it doesn't
| Stage | AI suitability | Human role |
|---|
| Familiarization | Medium — summaries orient you | Read the data yourself |
| Framework development | Low — judgment-heavy | Own the framework |
| Indexing | High — deductive coding | Spot-check against definitions |
| Charting | High — summarize into cells | Verify fidelity to quotes |
| Mapping/interpretation | Low — meaning-making | Draw the conclusions |
Where Qualitati fits
Qualitati is an AI user research platform for product, UX, and customer insights teams. For framework analysis, its QDA Workspace supports AI-assisted inductive and deductive coding, codebook generation, and theme visualization — mapping directly onto the framework-development, indexing, and charting stages. For larger corpora, ThemeLens runs a map-reduce thematic pipeline across up to 100 transcripts at once, mapping codes to research questions and synthesizing themes with participant-anchored quotes. Both keep a human researcher as the interpreter of the matrix, which is exactly where the Framework Method places analytical authority. Qualitati positions itself as an AI-native alternative to NVivo, ATLAS.ti, and MAXQDA, with transparent per-credit pricing.
Limitations and methodological cautions
- Structure can flatten nuance. Forcing rich talk into cells risks losing context; keep quotes and cell-to-source links so meaning is recoverable.
- AI charting can drift. Summaries may subtly reframe what a participant said. Verify a sample of cells against transcripts every project.
- Interpretive codes are hard for LLMs. As of July 2026, agreement is strong on concrete deductive codes and weaker on subtle, interpretive ones (Misra et al., 2026).
- Framework quality caps everything downstream. A vague framework produces an unreliable matrix regardless of the model. Invest in definitions first.
- Human-review note: validity, reflexivity, and final interpretation claims should be confirmed by a qualified researcher, not delegated to AI.
Framework Analysis Readiness Checklist
Score each item 0 (absent), 1 (partial), or 2 (solid). Aim for 14+ / 18 before you report from an AI-charted matrix.
| Criterion | What "solid" looks like |
|---|
| Familiarization done by a human | Team read a transcript subset, not just AI summaries |
| Framework mapped to questions | Every code traces to an RQ or the interview guide |
| Clear code definitions | Inclusion and exclusion rules per code |
| Indexing spot-checked | Human verified a sample of AI-applied codes |
| Cells anchored to source | Each matrix cell links back to transcript lines |
| Charting fidelity verified | Sample cells checked against original quotes |
| Cross-case comparison usable | Columns are comparable; few empty cells unexplained |
| Interpretation is human-led | Conclusions drawn by researchers, AI as prompt only |
| Audit trail kept | Framework version and changes logged with reasons |
Who this is for — and when not to use it
Who this is for: product, UX, health, evaluation, and policy researchers who have defined questions, multiple coders, or a need for transparent, comparable cross-case analysis.
When not to use this approach: if your goal is deep interpretive theory-building from a handful of cases, reflexive thematic analysis may fit better; if you have unstructured exploratory data with no stable questions yet, start with open inductive coding before imposing a framework.
FAQ
Is framework analysis the same as thematic analysis?
No. Both identify themes, but framework analysis adds a defined five-stage procedure and a case-by-theme matrix for systematic comparison. Thematic analysis, especially the reflexive form, deliberately avoids that rigid structure.
Can AI do framework analysis on its own?
Not reliably. AI performs well at indexing (deductive coding) and charting, but framework development and final interpretation require human judgment. Treat AI as an accelerator within a human-led process.
What is a framework matrix?
A framework matrix is a spreadsheet with cases in rows and themes/codes in columns; each cell holds a summary of that case's data for that theme, with a pointer back to the source transcript.
Which tools support framework analysis?
Traditional QDA software such as NVivo, ATLAS.ti, and MAXQDA support framework matrices. AI-native platforms like Qualitati's QDA Workspace and ThemeLens add AI-assisted coding, codebook generation, and theme synthesis while keeping the researcher in control.
How many transcripts do I need?
Framework analysis works with small and large samples. The matrix stays legible with a handful of cases and scales to larger studies when charting is AI-assisted, though very large corpora may benefit from a map-reduce thematic pipeline first.
Bottom line
Framework analysis gives qualitative work a transparent, comparable structure, and AI makes its most tedious stages — indexing and charting — far faster in 2026. The method's own logic keeps humans where they belong: developing the framework and interpreting the matrix. Used that way, AI-assisted framework analysis is rigorous, auditable, and fast.
Start free with 30 credits — no credit card required. Create an account, run an AI-assisted coding project in QDA Workspace, or compare transparent pricing with NVivo, ATLAS.ti, and MAXQDA.