Working with the assistant
The assistant works through the steps on this page with you, and you can ask it about your study at any point. When a change you ask for would reach something built from an earlier step, such as your Theory of Change or the interview outline, it shows the change on a card in the chat with what accepting it does, and nothing changes until you accept. Once an outline exists, a regeneration of the whole outline is proposed the same way. A preference you state for the whole study, such as “keep questions short” or “never ask about income”, becomes a standing instruction. Standing instructions appear beside the chat under The assistant keeps to:, every later step and regeneration follows them, and you can remove one with its ×. You can attach documents (PDF, Word, Markdown, text or CSV files) or add a website at any point in the conversation, and the assistant reads them into your study. It cannot publish your study, invite respondents or change the answer options of a library question, and it tells you so when you ask.Step 1: Describe your project
You tell voi about your project in plain language: who you serve, what you do, and what you are trying to change. This works as a conversation, not a form. If you start the study from the home page, your first description and any files you attach there become its first message in Plan. You can add your website and attach documents or an existing survey, at the start or later in the conversation. voi pulls relevant details and questions from them, instead of starting from a blank prompt. Questions you bring keep their wording, and you choose which of them the interview uses. If a document reads as a complete survey, the assistant offers to run it as your interview as it stands, or to build its questions into an interview it composes around them. As you talk, voi builds up a summary of your organisation and your project, called the Context. voi shows it back to you, so you can review and correct it before moving on. If something important is missing, such as exactly who you serve, or anything sensitive about the topic, voi asks a clarifying question rather than guessing. The Context includes your study focus, under What do you want to find out?: an optional purpose for the study and any further topics the interview should cover beyond your Theory of Change, such as which of your services respondents use most. Each topic shows where it came from, such as AI suggested or Custom. Once the Context looks right, approve it with Approve Context on its card in the chat. If something it needs is still missing, the card says what to add. The Context stays editable after you approve it, and the assistant tells you before a change reaches a later step. See Working with the assistant.Step 2: Build a Theory of Change
Once you approve the Context, the assistant asks whether you want a Theory of Change, with Build a Theory of change and Continue without one. A Theory of Change is the logic connecting what you do to why it matters. It is the reason every study starts here, rather than with a generic checklist of indicators. This step is optional. You can skip it and go straight to your interview guide, though voi will have less to work from when it drafts your questions as a result.What it’s made of
A Theory of Change in voi is built activity by activity. For each activity, voi tracks:- Input: what goes into the activity, such as resources, funding, or effort.
- Output: what the activity produces directly.
- Short-term outcome: the first change that output leads to.
- Long-term outcome: what that short-term outcome builds toward, for that same activity.
How the draft gets built
- With enough context to work from, voi drafts the full chain right away, working backwards from your long term impact through to your activities. Where voi finds a weak or unsupported link in this logic, voi asks you a couple of short, targeted questions about it before presenting the draft. voi never asks generic filler questions, and never asks questions about how you will measure anything. Measurement comes later, once the Theory of Change itself is settled.
- With less to go on, it walks you through building the same chain step by step in a guided conversation instead of drafting one upfront.
Assumptions
Alongside the chain itself, voi surfaces the assumptions your logic depends on. These are the things that have to hold true for a given link to work, which you haven’t necessarily stated outright. Each assumption is tied to the specific link it could undermine, rather than listed generically at the end, so you can see exactly where your Theory of Change is most exposed if that assumption turns out to be wrong.Reviewing and approving it
You can edit the draft directly. You can rewrite an activity, an outcome, or its rationale. Alternatively, you can describe the change you want in chat, and voi updates the draft for you. Both methods work at any point while you are still reviewing the draft. Every question in your guide traces back to this chain, so it is worth checking that the chain actually reflects your project’s logic before you move on. Editing an outcome or its rationale here is faster than fixing individual questions one by one afterwards. When the chain looks right, approve it with Approve Theory of change on its card in the chat, and voi uses it to sharpen the interview guide. It stays editable afterwards: a change to your outcomes marks the interview outline as out of date until you regenerate it, and a change to wording alone, such as a rationale, leaves the outline as it is. You can also continue without a Theory of Change, though voi has less to work from as a result.Step 3: Review the generated interview guide
Before it writes the first guide, the assistant asks which commonly chosen indicators every interview should include: Net Impact Score, which it recommends, Most Significant Change and Net Promoter Score. If you choose Most Significant Change, the assistant also asks which period the Most Significant Change question should look back over, such as the last six months. voi then recommends indicators and matching questions for each outcome in your Theory of Change and each topic in your study focus, drawn from its indicator library, and writes an AI-generated question to fill any gap the library doesn’t cover. See How indicators work, and for the principles behind the questions and their order, How voi writes your interview. What you get is a full interview outline, organised into a fixed sequence of sections:
Each section is collapsible and shows how many questions it holds. A button at the top of each section adds a new question to it directly.
The interview outline and the indicator library
The guide sits behind two views: the interview outline, which is the sequence of questions respondents actually see, and an indicators view, which lists the indicators your questions feed. You can switch between them at any time while you’re shaping the guide. To bring in something the generated guide doesn’t already include, use the library browser. It searches across both indicators and questions at once, marks anything you’ve already added so you don’t duplicate it, and lets you add an item straight into whichever section makes sense. To replace a suggested indicator, ask the assistant for a different one for that outcome. It lists the alternatives with how well each fits and the one it recommends, and once you accept, it swaps the indicator’s questions in place and keeps any you locked. You can also remove an item and add a different one from the library, or use Regenerate with AI on an AI-written question, described below. Removing one of the commonly chosen indicators in the indicators view turns it off for the study, after you confirm.Editing a question
Every question card has a menu of actions.
You can also drag any question by its handle to reorder it within a section, or move it into a different one. To delete several questions at once, tick them and delete them together in one confirmed step.
You can make the same kinds of change through the chat. Ask the assistant to reword or remove questions, several in one message, to change the answer options of an AI-written question, to add questions on a topic, or to regenerate only the questions of the outcomes you name. When it regenerates, questions you locked and questions you wrote or brought yourself stay as they are.
A question that appears only after a certain answer to an earlier question shows that condition under it. You can change the condition for this study, or reset it to the rule the question has in the library. See Skip logic.
voi flags problems inline as you edit: a warning appears under a question if, for example, it asks about two things at once, its answer options overlap or leave a gap, its skip logic conflicts with another question, or a translation would run over its length limit. The same notes appear in the Add question dialog and in the question sidebar while you write, and only a few problems, such as a closed question with fewer than two options, stop you from saving. See Automatic checks. If a warning traces back to how a library item itself is defined, the message tells you to fix it in the library rather than on the question card. A separate banner (which you can turn on or off in the guide’s settings) flags bigger issues across the whole guide, such as a Theory of Change outcome that’s no longer covered, and offers to recompose the guide to fix it.
Setting a target completion time
Without a target, voi covers every outcome first and keeps the interview as short as that allows. The line above the outline always shows the estimated length, the number of questions and the follow-ups, and clicking it opens a breakdown. To set a target, hover over that line and click + (Set a target length), or turn on Enable time budget on the Questions tab of the guide’s settings. You can choose from 3 to 30 minutes, and the target starts at the guide’s current estimated length. voi treats a target as a cap: when it composes or regenerates the guide, it takes out the questions that matter least to your outcomes first, and removes the last question of an outcome only when nothing else fits. Questions you locked, questions you wrote or brought yourself, the questions about who is answering and the questions of the commonly chosen indicators you picked always stay. If an outcome does not fit within the target, the assistant names it and offers ways to make room, such as a shorter indicator, fewer follow-ups or a few more minutes, and you decide. While a target is set, a Target chip above the outline changes it a minute at a time. The length line turns red when the guide runs over the target, and Recompose beside it trims the guide to fit whenever a recompose would change it. To drop the target, hover over the chip and click its × (Remove target).Other settings before you share
Beyond the questions themselves, a study has its own settings, separate from your workspace-wide settings. They sit on three tabs, General, Questions and Voice & language, and each tab saves on its own:
You pick the commonly chosen indicators (Net Impact Score, Most Significant Change and Net Promoter Score) in the chat before the first guide, and you can change them later by asking the assistant. Removing one in the indicators view turns it off for this study.
Your organisation’s privacy policy link applies to every study in the workspace; there’s no separate, study-specific privacy policy. See Workspace settings.
The welcome and closing consent text is editable per study in the interview outline. See Share for what you can and can’t change about it.
Before you move to Share
A few things worth getting right before you send a study out:- Balance your question mix. Open-ended questions, especially answered by voice or video, take longer than tapping a scale. Decide deliberately whether you want depth (more open-ended questions, a longer completion time) or a larger sample in less time (more closed questions).
- Use the indicator library, but don’t force-fit it. If a suggested indicator doesn’t quite match what you’re trying to measure, edit or replace it rather than leaving a mismatched question in place.
- Plan ahead for disaggregation. If you know you’ll want to compare results by location, age, or another group once you’re in Analyse, make sure your study actually asks for that information now; it’s easier than wishing you had it once collection is underway.