Get started
Every Opini.ai project starts with one decision: is this a Survey or an AI Agent? That choice determines how the project is built from here — the question types you get, what respondents experience, and how you'll analyse the results. You pick it once, when you create the project, so it's worth understanding both before you start.
The two project types
A traditional market research survey
Built the way research teams already work: sections and categories, grid (matrix) questions, single and multi-choice, rating scales, and skip/display logic. Respondents answer a fixed questionnaire and the data comes back structured and ready to tabulate.
A conversational discussion guide
Built like a moderator's discussion guide, not a questionnaire. Questions are predominantly open-ended — the agent asks, listens, and decides live whether to probe deeper before moving on. Single/multi-choice questions exist too, but are best kept to screeners and rating scales rather than the core discussion. Open-ended responses get coded into categories afterwards, during analysis.
How they differ
The project type isn't just a label — it changes the builder you get and the shape of the data you get back.
| Survey | AI Agent | |
|---|---|---|
| Built with | Categories, sections & question grids | A discussion guide of open topics |
| Question format | Mostly closed-ended (choice, rating, grid) | Predominantly open-ended; choice/rating available for screeners |
| Depth | Fixed at design time | Adaptive — agent probes live per response |
| Response data | Structured, pre-coded by design | Raw text, coded into categories after fielding |
| Logic | Skip / display logic between sections | Conversational branching driven by the answer |
| Best for | Tracking studies, KPIs, large quota-based samples | Exploring the "why", uncovering themes you didn't ask for |
What you build next
In a Survey project
Organise your questionnaire into categories (e.g. Screener, Usage, Brand), then add grids, choice and rating questions inside each one, with logic to route respondents between sections.
In an AI Agent project
Write your discussion guide as a list of open-ended seed questions and topics, adding single/multi-choice only where it fits — typically screeners and rating scales. Set how strictly the agent should stick to script and how deep it should probe — the agent handles the rest live.
For AI Agent projects, plan for a coding pass after fielding: since every answer is open text, you'll review responses and group them into categories before you can chart or quantify themes — the guides below cover that workflow in detail.
Next up
- Creating a project→
- Writing a discussion guideComing soon
- Coding open-ended responsesComing soon