Qualitative research produces a lot of material. A two-week video diary with thirty participants can easily mean hundreds of videos, thousands of photos and many hours of talk. Reviewing it all used to take most of the project time. AI changes that, but only if it is used in a way that keeps the researcher in charge and the evidence in view.
Where AI helps most
Transcripts
Transcription is the clearest win. Every video and voice note can be turned into text with times, so you can read, search and quote instead of watching everything twice. Click a line and the video plays from that moment.
Translation
Multi-country studies used to wait for translators. Now a German or Polish video can arrive with a transcript in the language spoken and a translation into the team's main language. The original stays available for checking nuance.
Reading photos and screens
Photos and screen recordings carry text: labels, prices, menus, error messages. Reading that text makes it searchable. A short description of what a photo shows ("a fridge door with three yoghurt brands") helps you find it later.
Search by meaning
Keyword search finds "expensive". Search by meaning also finds "not worth the money" and "I'd wait for an offer". For a researcher looking for every mention of price, that is the difference between a few results and all of them.
First-draft themes
AI can read every answer and transcript in a study and suggest the main themes, with how people feel about each one and the quotes that support it. That is a useful first draft, especially at the start of analysis when you do not yet know the shape of the data.
Where AI can mislead
- Made-up or bent quotes. A language model may tidy, merge or invent a quote that sounds right. In research, a quote that was not said is worse than no quote.
- Lost context. "I love it" said with a sigh is not love. Text alone misses tone, faces and the room.
- The loud majority. Themes backed by many short answers can crowd out an important point made by two people in depth.
- False completeness. A neat list of themes feels finished. It may have left out answers that did not fit, or material that was too large to read in one go.
Five rules for using AI in qualitative analysis
1. Every quote must be checkable
Each quote should link back to the exact entry or moment in the video. Better still, the tool should check each quote word for word against what the participant said, and drop any it cannot find.
2. The researcher decides
Treat AI themes as suggestions. Accept, edit or reject each one, and keep a record of what you changed. The client pays for your judgement, not the model's.
3. Watch the clips behind the theme
Before you put a theme in the report, watch a few of the moments it is based on. That is where you catch the sigh behind "I love it".
4. Know what was left out
If the material is too large for one pass, the tool should say how much it left out, so you can analyse in parts, such as task by task.
5. Know where the data goes
Participants agreed to share their answers with you, not with any AI service in the world. Know which provider reads the data, and prefer a setup where your agency can use its own AI account under its own contract.
Cost and speed
AI use costs money, and costs add up across many studies. Some simple choices help: do not run the same analysis again when nothing has changed, reuse translations of identical texts, and read the text on photos and screens with a local tool, using AI only for the short description. Faster and cheaper analysis means more time for the part only a researcher can do.
How Zolensa approaches it
On Zolensa, every video and voice note is transcribed and translated into the study's main language, and photos and screen recordings are read so they can be searched. The Analysis tab suggests themes with sentiment, and every quote is checked word for word against the source, with a link to the moment. Your team accepts, edits or rejects each theme and turns quotes into highlights. Agencies can use their own AI provider, so the data and the costs stay under their own contract. See the features for details.
