Short answer
A conversational AI survey replaces the form with a short written conversation. Each person answers in their own words, and the AI asks a follow-up question based on what they just said. You end up with reasons, examples and disagreements, where a form would give you a score.
Use one when you already know something is wrong, or simply unclear, and you need to know why. Keep the classic survey for what it does well: measuring the same thing across many people, over time.
Have a question in mind already? Ask your own question and see the topics a conversation would cover.
What a conversational AI survey is
A classic survey asks everyone the same closed questions. That makes the answers easy to count and hard to interpret. "My workload is manageable: 3 out of 5" tells you there is a problem and nothing about where it comes from.
A conversational survey starts from one open question and adapts. If someone says the week gets difficult "when approvals pile up", the next question asks about the last time that happened, who was involved and what it cost. Three or four exchanges later you have a concrete situation, told by the person who lived it.
The AI plays two roles:
- during the conversation, it asks follow-up questions that stay on the topics you approved;
- afterwards, it groups what people said into findings, and counts how many people described each one.
The first role is what separates it from a form with a comment box. A comment box collects first answers. A conversation gets to the second and third answer, which is usually where the useful detail sits. How AI follow-up questions work goes further into that.
Survey, conversational AI survey, or a round of calls?
| Classic survey | Conversational AI survey | Calls you run yourself | |
|---|---|---|---|
| Best for | Measuring how much, across many people | Understanding why, across a small group | Sensitive or complex subjects, a handful of people |
| Questions | Fixed, mostly closed | One open question, then follow-ups that adapt | Fully flexible |
| Your time | Low | Low | High: scheduling, talking, note-taking |
| What you read | Scores and charts | Findings, counts and quotes | Your notes |
| Typical size | Hundreds or more | Ten to a few dozen | Five to ten |
| Weak spot | Does not explain the score | Does not give a statistic for a whole population | Hard to repeat, and people may hold back in front of you |
None of the three replaces the others. A score tells you where to look. A conversation tells you what is happening there. A call with a person you trust goes further on a delicate point.
When to use one instead of a survey
A conversational AI survey is a good first choice when:
- you do not yet know the possible answers. A closed question only works when you can list the options in advance. If you cannot, any list you write is a guess.
- the group is small. With 12 clients or 20 team members, percentages mean little. What each person describes means a lot.
- you already have a score and it has not helped. Another survey will return another score.
- people are tired of forms. One question answered in their own words feels different from forty rating scales.
- you need to act soon. Ten conversations can be read in an afternoon.
You do not need to run a survey first. Many questions never needed a score to begin with: "Why do new customers not come back after their first order?" starts as a question about reasons.
When a classic survey is still the right tool
Stay with a survey when you need:
- a figure you can track quarter after quarter;
- a comparison between sites, teams or segments with enough people in each;
- a result that represents a large population;
- a simple yes or no from many people.
A conversational study of ten people does not tell you that "70% of customers" think something. It tells you that seven of the ten people you asked described the same problem, and shows you what they said. That is strong enough to decide what to test next, and it should be reported as what it is. How many conversations are enough covers how to size the group.
What the questions look like
The quality of a conversational survey depends on the opening question. Good ones ask about something that happened.
| Instead of | Ask |
|---|---|
| How satisfied are you with delivery? | Tell me about your last delivery. What happened on the day? |
| Do you feel supported on your course? | Think of the last time you got stuck. What did you do next? |
| Would you recommend us? | What nearly made you cancel, and what kept you? |
| Is your workload manageable? | What makes a normal week harder than it needs to be? |
Each one invites a story, and a story contains the detail you can act on. How to write a research question shows how to turn a vague worry into a question like these.
A fictional example: one question, six kinds of organisation
The same method travels well. These are invented examples of what the score said and what the conversations added.
| Who was asked | The survey said | The conversations added |
|---|---|---|
| A team of ten | Workload manageable: 3 out of 5 | Work is ready but waits for one manager's approval (7 of 10) |
| Seasonal staff | 41% would work here again | Rotas are published less than three days ahead (8 of 10) |
| New customers | Satisfaction with the first order: 4.4 out of 5 | No delivery time fits a working day (6 of 10) |
| Clients of a consultancy | Overall satisfaction: 4.1 out of 5 | The report is read once; the weekly calls change decisions (7 of 9) |
| First-year students | "I feel supported": 3.1 out of 5 | They did not know who to ask when they fell behind (7 of 10) |
| Gym members | Likely to recommend: 7 out of 10 | Their class moved and nobody told them (6 of 10) |
In every row the score was accurate. It simply did not say what to change.
How to run one in a week
- Write the question in one sentence, the way you would ask a colleague.
- Check the topics the conversation will cover, and remove anything you do not want asked.
- Choose ten people who have lived the situation recently. Mix those who are happy and those who are not.
- Tell them what will happen: how long it takes, who reads the answers, and what you will do with them.
- Read the findings with the quotes next to them. Note where people disagree.
- Decide one or two changes to test, and tell participants what you did.
Where Lontra fits
Lontra is a conversational AI survey tool built for this job. You write your question, review the topics it proposes, and invite people with a private link that works on a phone and needs no account. Lontra holds each conversation, follows up on what the person says, and shows you findings with how many people described each one, where they disagree, and the quotes behind them. It is free to try with 10 conversations. See how it works or start with your own question.
Two limits are worth knowing. Lontra does not give you a statistic for a large population, so keep your survey for tracking. And participants are told before they start that you will read their answers, so think about who is asking and how the findings will be used. Who sees what is set out on the trust page.
Start with one question
Pick the score that has puzzled you longest, or the question you have never managed to fit into a form. Write it as a sentence. Ten conversations later you will know whether it deserves a bigger study or a simple fix.
Frequently asked questions
- What is a conversational AI survey?
- A survey held as a short written conversation. Each person answers an open question in their own words, and the AI asks follow-up questions based on what they said. You read findings with counts and quotes.
- Does a conversational AI survey replace a normal survey?
- For questions about why, yes: you can use it without running a survey first. For measuring the same thing across many people over time, a classic survey is still the right tool.
- How many people do you need for a conversational survey?
- Ten to a few dozen people who have lived the situation recently is usually enough to see what repeats. It does not produce a statistic for a large population.