An AI-moderated interview is a research interview where an AI asks the questions and the follow-ups, and a real person answers. In ScatterSpoke you write the discussion guide as a survey spec (goal, topics, how deep to probe, what’s off limits), send a link, and Spoke interviews each participant in their browser. You read every transcript and a synthesis with the quotes and the number of people behind each theme. People use it for AI user interviews and other AI qualitative research where the questions are already known. It’s not a replacement for a skilled researcher on exploratory or sensitive work.
The pain is committing to dates, not finding places. The organizer does the chasing alone.
Examples of how Spoke follows up on a vague first answer in a research interview. The probe comes from what that participant said, so two people can get different follow-ups.
The pain is committing to dates, not finding places. The organizer does the chasing alone.
Unclear add-ons push self-serve buyers into a sales conversation they want to avoid.
They didn’t leave for a better product; they left for one that was already where they work.
The most-read critique, Carl Pearson’s “methodological error amplified”, grants that AI interviews reach many people fast, then argues they “quant the qual”: they explore and count in one pass, treat every answer as equal, and report numbers on categories nobody checked. He’s right about the core problem, and our synthesis does count in the same pass. Here’s how to use it without overclaiming, and where a human is still better. Nielsen Norman Group’s study of AI interviewers reaches a similar conclusion: use them to complement human moderation, not replace it.
You set the goal, the topics with a reason for each, their priority, and a probe depth per topic (light, standard or deep). Spoke follows up on what each participant said within those limits. Priorities guide the interview; they don’t guarantee every participant covers every topic.
Treat an agent survey round as the exploring wave. Once the themes hold steady across rounds, count them properly with a classic survey built from those categories. That’s the explore-then-count sequence the critique recommends.
You review the goal, topics, opening prompt and off-limits topics before anything is sent. The follow-ups are written live, so you can’t pre-approve each one: read the first transcripts for leading probes before sending widely, the way you’d pilot a guide. Guardrails tell the moderator which subjects and behavior to avoid.
The synthesis shows each theme with the supporting quotes and the number of participants behind it, and where people disagree. Treat counts as “how many people raised this,” not as a measured share of a population. Every mention counts the same; the synthesis doesn’t know that one person’s offhand remark matters less than another’s core problem. You do. Check the quotes against the transcripts before you report a theme.
Early exploratory work where you don’t know the questions yet, sensitive or emotional topics, building rapport over a long session, watching someone use a product, and reading body language. Use a human there.
A defined guide you need to run with many people, participants spread across time zones, quick follow-up rounds after a first study, and people who’d never find time for a scheduled call.
A strong first draft of the synthesis plus every raw transcript. You still decide what the findings mean.
Many tools sold as “AI focus groups” generate answers from simulated personas. That can help you brainstorm, but no customer said it. ScatterSpoke is the other kind: real people answer, and the AI only moderates. Planning the questions? See focus group questions for a group and qualitative interview questions for one-on-ones.
| Question | Answer |
|---|---|
| Do participants stay anonymous? | It’s your choice per study. Anonymous responses are available on paid plans. Don’t promise anonymity unless the setting is on. |
| Do I get raw transcripts? | Yes. Every response is readable as a transcript, with the follow-ups Spoke asked that person. |
| Can I export? | Yes, the survey data exports as CSV. Raw answers stay readable even when the synthesis is locked by plan. |
| Do participants need an account? | No seat and no access to your workspace. They answer through a link; some settings ask them to verify an email address. |
| Languages? | We don’t publish language support yet. Pilot with a few participants in your language first. |
| What do participants know? | Tell them an AI runs the interview and what happens to their answers. NN/g found people hold back when they don’t know. |
| Data retention? | Workspace admins can set a retention policy from 30 to 3,650 days. With no policy set, nothing is deleted automatically. |
| Recruiting panel? | No. You invite your own participants. |
Product facts from docs.scatterspoke.com, checked September 27, 2026.
Your first ten agent surveys are free, and they don’t expire. Paid plans include 20 or 150 agent surveys a month; see pricing for current prices. Classic surveys are free and unlimited on every plan.
Want the question craft behind the follow-ups? Read the probing questions guide, or see how the same approach works for product feedback and customer exit surveys.
A research interview where an AI asks the questions and the follow-ups and a real person answers, usually in a browser. It lets you run the same guide with many participants at once.
For a defined guide with many participants, they can be consistent and fast. For exploratory, sensitive or rapport-heavy sessions, a skilled human moderator is still better.
In ScatterSpoke it follows up on what each participant said, within the goal, topics, priorities, probe depth and off-limits topics you set in the survey spec.
For early discovery where you don’t know the questions yet, sensitive topics, long rapport-heavy sessions, and anything where you need to watch someone use a product.
No. It runs interviews and drafts a synthesis; deciding what to study, judging what findings mean and doing exploratory work still need a researcher.
Many tools sold as AI focus groups generate answers from simulated personas. AI-moderated interviews ask real people; only the moderator is AI.
In ScatterSpoke your first ten agent surveys are free, and each is one survey sent to your participants. Paid plans include a monthly allowance; see the pricing page.
Real people answer, Spoke moderates, and you read every transcript. Your first ten agent surveys are free.
