AI-Moderated Interviews vs Voice in Your Existing Survey

The question researchers are actually asking in 2026

Search for voice survey tooling today and most of what comes back belongs to a category that barely existed two years ago: platforms where an AI conducts the interview. Listen Labs, Strella, Outset, Conveo, Glaut, Perspective AI, TheySaid and Koji all sit in that space. The pitch is consistent and genuinely appealing — you write a discussion guide instead of a questionnaire, an AI moderator asks the questions and follows up on what it hears, and you get back transcripts with an analysis layer on top.

That is a real advance, and for some studies it is the right tool. It is also not what Voice Capture is, and pretending otherwise would waste your time and ours. So this is the comparison written the way we would want it written if we were on the buying side: what each approach actually is, what changes operationally, what each one costs you in flexibility, and — stated plainly — when the AI-moderated platform is the right answer and Voice Capture is not.

One correction before anything else, because the entire decision turns on it: Voice Capture does not moderate interviews. It puts a record button next to an open-ended question in the survey you already run, transcribes the answer, and writes the text into your data file. An add-on can ask one follow-up question on that answer, generated from what the respondent actually said — but nothing decides the shape of the session except the questionnaire you wrote. If adaptive probing across a whole interview is what you need, there is an honest answer further down this page, and it is not "our widget does that too".

What each approach actually is

Approach A — an AI-moderated interview platform

The study lives on the vendor's platform. You author a guide there, the platform runs the conversation with each respondent, and the questioning adapts: when someone says something interesting, vague or contradictory, the AI asks about it. What comes back is a set of conversation transcripts plus whatever analysis the vendor layers on them.

The defining property of this category — the one that holds for every product in it, and the one your decision should hinge on — is that the interview runs on their platform. Whatever else differs between them, that does not. Everything else follows from it: the instrument is authored there, respondents are routed there, the response record is created there, and your data leaves through their export.

We have not fielded studies on each of these platforms, so we are not going to score them against each other on features, quality or price, and you should be suspicious of any vendor who does. Read their own documentation, and ask them the operational questions in the sections below.

Approach B — voice inside the survey you already run

Nothing moves. Your questionnaire stays in Alchemer or QuestionPro, with your routing, your quotas, your screener and your panel wiring exactly as they are. You paste a snippet, and one open-ended question gains a microphone button beside its text box. A respondent speaks instead of typing, the audio is transcribed, and the text lands in the response record joined to the session — so it merges with the rest of the data file automatically.

The question is still the question you wrote, and nobody asks a second one. What changes is how much you get back from it. Across seven client studies we measured, typed answers averaged 10 words and spoken answers to the same questions averaged 23 — roughly twice the material, from an unchanged questionnaire (how we measured this). The medians tell the same story: 8 words typed, 16 spoken.

This is what Voice Capture is: an add-on to the survey platform you already pay for, not a platform of its own.

What changes operationally

Fieldwork and the instrument

Approach A means authoring the study twice: once as it exists, once as a conversation guide on the new platform. That is not a copy-paste job, because a good questionnaire and a good discussion guide are not the same artefact. Approach B changes one element on one page and leaves the instrument otherwise untouched.

This matters most for trackers. Changing how a question is asked changes the answer — moving a tracked open end from a typed box to a spoken one already does that, which is the point, so introduce it at a wave boundary and note it in the deliverable. Moving the entire instrument to a conversational platform changes far more than one question, and there is no honest way to splice that into an existing time series.

Panel, quotas and incidence

Your quota machinery lives in your survey platform. On Approach B it never comes up: quotas fill exactly as they did before, because the same instrument is doing the same routing.

On Approach A you have to decide where the screener lives, and this is where budgets get surprised. If only a minority of the people you contact qualify, you are paying to screen several for every one you interview. In your existing survey that cost is already in your fieldwork line and terminating a respondent is free. On a separate interview platform you either rebuild the screener there, or keep it in your survey and redirect qualified respondents out to the interview and back.

The redirect is the part people underestimate. You are matching a respondent ID across two systems, and every respondent who leaves and does not come back is a complete you paid for and cannot count. Ask the vendor exactly how the hand-off and the return redirect work before you price the study.

The data file

On Approach A the response record is created on their platform. You get an export, and if you have a quant instrument as well you merge the two yourself, on whatever ID you managed to carry across the hand-off.

On Approach B there is only one file. The transcript is written into the response record of the survey that produced it, joined by session ID, and it appears in the same CSV or Excel export as every closed question — so it can be crosstabbed against your banner points with no merge step. For a tracker that gets tabulated the same way every wave, that is not a convenience, it is whether the wave is comparable at all.

Timeline

Rebuilding an instrument on a new platform is days rather than hours: authoring, then a full QA cycle on routing and quotas you have just re-implemented, then a soft launch you actually watch. Adding voice to an open end is minutes per question once the account exists — one script tag and one block in Alchemer (setup), a single Pre JavaScript Logic block in QuestionPro (setup), and a test recording on a phone to confirm it arrives joined to the session.

Cost shape

We publish ours and we will not invent anyone else's. Voice Capture is bought as one-time credit packs, where one credit is one transcribed response, and credits never expire:

• Free Trial — 250 credits, no card. Enough to pilot a real open end in a real wave.
• Essential — $99 for 1,000 credits. Adds Excel export alongside CSV and removes widget branding.
• Professional — $299 for 4,000 credits. Adds API export and custom domains.
• Custom — negotiated, for continuous high-volume fieldwork.

The structural consequence: a tracker running two waves a year does not pay for the ten months in between. Current details on the pricing page.

For the AI-moderated platforms, ask each vendor directly — we do not publish other companies' pricing, and neither should anyone comparing them to us. Four questions worth asking whoever you shortlist, because the answers change the shape of the cost rather than just the number:

• Is it priced per completed interview, per study, per seat, or as an annual commitment?
• What does a month with no fieldwork cost?
• What happens commercially to a respondent who starts and drops out, or who screens out?
• Does the analysis layer come with it, or is it a separate line?

What you give up, either way

Choosing the AI-moderated platform

• Your instrument as it exists, and with it the comparability of anything you have already fielded.
• The routing, quota and panel machinery you have already debugged, which has to be re-implemented or bridged.
• A single response record, unless you build and maintain the merge yourself.
• Team familiarity — someone has to learn to write good AI probing instructions, which is a genuine skill and not the same one as writing a questionnaire.

Choosing voice in your existing survey

• Adaptive depth. One question, one spoken answer. If the respondent says something fascinating, nothing asks them to elaborate. This is the real trade, and you should not let anyone talk you out of feeling it.
• A survey platform. Voice Capture hosts no questionnaires, routes no respondents, manages no quotas and sources no sample. It needs a survey to live inside.
• Coverage of every platform. Alchemer and QuestionPro are live today; anything that accepts custom HTML and JavaScript can use the generic embed. A locked-down instance that forbids custom scripts cannot run any snippet-based tool, ours included.
• Voice on closed questions. It replaces typing on the open ends. It does not turn a grid into a conversation.

The decision table

When the AI-moderated platform is the right answer

These are the cases where Voice Capture is the wrong tool, and we would rather you knew now than three weeks in:

• Exploratory depth interviews. You do not yet know what the questions are. The value of the exercise is precisely that something unexpected gets followed. A fixed open end cannot do that, no matter how long the answer runs.
• Small-n qualitative where probing is the method. Twenty conversations that each go somewhere different beat two hundred parallel answers to the same prompt. If you would otherwise have booked a moderator, you are comparing against a moderator, not against a survey.
• The study is not already running on a survey platform. If there is no questionnaire, no routing and no quota plan yet, there is nothing for Voice Capture to embed into. You need a platform first, and a conversational one may well be that platform.
• Concept and message testing where the reasoning matters more than the count. When you need to understand why a claim lands, and to push back in the moment, the moderation is the deliverable.

In all four, buy the AI-moderated platform. Our widget will not get you there.

When voice in your existing survey is the right answer

• You already run the study somewhere. Alchemer, QuestionPro or anything that accepts custom JavaScript — the instrument exists, the panel is wired, the quotas are set, and rebuilding all of it to improve three open ends is a bad trade.
• It is a tracker. Comparability across waves is a hard requirement, and you can afford to change one question at a wave boundary but not the whole instrument.
• The open ends are the weak point of a strong quant study. The closed questions are fine; the verbatims come back in three words and get one slide. That is the exact gap voice closes.
• You need the transcript in the same file as everything else — crosstabbed against your banner points, not merged in afterwards.
• You came off Phonic and only ever used it for the microphone. See the migration guide, which walks through exactly that case.

"I want AI probing, but I do not want to move the study"

This is a legitimate position, and it deserves a precise answer rather than a blurred category.

If what you actually want is the probing — an AI that reads the answer and asks the next question — without moving the study, that is exactly what the follow-ups add-on does: AI follow-up questions inside the survey you already field, on the same account and the same credit balance as voice. It is one follow-up per targeted answer, not a moderated conversation, and that distinction is the whole point of this page.

What it is not is moderation, and anyone who tells you the microphone widget also conducts the interview is describing something that does not exist.

How to decide in an afternoon

1. Write down whether the study already exists as an instrument somewhere. If it does not, you are shopping for a platform, and Approach A is a real candidate.
2. Ask what the open ends are for. If they need to be followed up on to be worth anything, the probing is the method — buy moderation. If they are strong questions returning thin answers, the problem is the keyboard, not the questionnaire.
3. Check comparability. A live tracker with history is an argument against moving platforms that usually settles the question on its own.
4. Price the hand-off, not just the tool: where the screener lives, the redirect, the ID matching, and what a screened-out respondent costs.
5. Pilot the cheap option first. One open end, one wave, a free account — the result is real data from your own respondents rather than a vendor's case study.

Frequently asked questions

Does Voice Capture moderate the interview?

No. An add-on can ask one AI follow-up on the answer to an open end you choose, inside your existing survey — see AI follow-up questions. What it does not do is moderate: the follow-up hangs off a question you wrote, and the questionnaire still decides everything else.

Can I use both approaches in one research programme?

Yes, and plenty of teams should. Run the exploratory phase as AI-moderated interviews to find out what the questions are, then field the quantified wave on your own survey platform with voice on the open ends. They answer different questions and are not substitutes.

Will an AI-moderated platform replace my survey platform?

Ask each vendor directly what it does about screening, quotas, panel integration and data export, because that is what you would be replacing. Some of it is a fair swap and some of it is not, and the answer is specific to each product.

Does voice recording work on phones?

Yes — it uses the browser's standard microphone permission, with no app or plugin, and works on iOS Safari and Android Chrome. If a browser cannot record, the widget hides itself and the text box carries on working, so a respondent never hits a dead end.

What happens to the audio?

It is transcribed and discarded; only the text is stored, in EU infrastructure. If you are moving from a tool that retained recordings, tell your privacy team — this is a change in your favour.

Try the cheap option on one open end

The Free Trial includes 250 credits and no credit card, which is enough to run voice on a real open end in a real wave and see what comes back before committing budget to a platform migration.

Start Free — 250 Credits, No Card →

Also read: Phonic alternative: voice inside the survey you already run · How to migrate from Phonic to Voice Capture