The open-ended questions are where the insight is. They are also the ones nobody has time to read.
Hand-coding free text is slow, and two coders rarely agree. So the boxes get skimmed, summarised loosely, or dropped from the report altogether — and the most honest answers in the study go unused.
How it works
Four steps from a raw export to coded, reportable data.
Upload your export
Bring the raw file from your survey platform. We detect which columns hold open-ended text.
Get a codebook
The AI proposes parent codes with sub-codes beneath them — a real coding frame, not a word cloud.
Classify every response
Each answer is tagged against the frame, with sentiment, at any sample size.
Review, explore, export
Correct tags, approve questions, then export coded data or an interactive report.
A codebook, not a word cloud
The AI reads every response and proposes a hierarchical coding frame: broad parent categories with specific sub-codes underneath. It is the frame you would have built by hand, available before you have read a single answer.
- Parent codes with nested sub-codes
- Sentiment attached to each code
- Responses that fit no sub-code fall back to the parent category
Built for recall questions
Unaided awareness questions are not thematic — they are a census. Ask which brands people can name and the platform counts every single mention rather than clustering them into themes, so nothing in the long tail is lost.
- Exhaustive count of every named brand, product, or place
- Spelling variants and abbreviations resolved to one entity
- Every count traceable to the responses behind it
The codebook stays yours
An AI-proposed frame is a starting point, not a verdict. Rename anything, split a code that is doing too much work, merge two that mean the same thing, or delete what you do not need. Affected responses are reclassified for you.
- Add, rename, split, merge, and delete codes
- Automatic reclassification after every change
- Edit the tags on an individual response by hand
A human signs it off
Work through a question in review mode, correct anything the model got wrong, and mark it approved. The coding you present is coding a person has checked — which is what makes it defensible when a client pushes back.
- Per-response tag correction
- Question-by-question approval tracking
- Every code links back to the responses inside it
Read it as text or as numbers
Coded text becomes quantitative data. See how often each code appears, how sentiment splits, and where a segment differs from the rest of the sample — with significance testing so you know which gaps are real.
- AI summary per question, alongside the counts
- Dual-basis percentages: within a code and as a share of the sample
- Crosstabs by demographic, with significance flagged
- Correlation explorer across questions
Ask the responses directly
Filter down to a segment, open the treemap to see what dominates, or simply ask a question in plain language. Answers are grounded in your responses and cited to them, so you can always click through to the verbatim.
- Filter and explore by code, sentiment, or demographic
- Sentiment leaderboard of the most positive and negative sub-codes
- Answers cited to the responses behind them
Almost entirely on environmental grounds. In this segment 'too much plastic' is the dominant packaging complaint, while older respondents raise practical issues like packs being hard to open.
Exports that fit your existing process
Take the coded data back into your own tooling, or hand a stakeholder something they can click through without a login.
- Excel export with codes attached and respondent IDs preserved, so it merges straight back into your dataset
- Standalone interactive HTML report you can send to a stakeholder
Coded Excel
One row per respondent, your original IDs intact, codes and sentiment in new columns.
Interactive report
A single self-contained file. Filters, charts, and verbatims, no account required to open it.
Fieldwork is rarely finished in one go
When the next wave of responses lands, add them to the existing analysis. They are classified against the codebook you already approved, so waves stay comparable and you never start over.
- Add new responses to a finished analysis
- Classified against your approved codebook
- Live progress while a large batch runs
Your respondents' words stay yours
Survey data is processed securely within the EU, encrypted in transit and at rest, and never used to train models. Projects are private by default, with role-based access for your team.