Published 14 October 2026. We update this guide as the market and our product change.
Search for "AI survey platform" and you'll get three different kinds of product under the same name.
Classic survey platforms with AI added. A mature tool for building and sending questionnaires, now with a button that drafts questions for you or summarises the open ends after fieldwork. The survey itself works the way it always has: fixed questions and fixed routing, with a text box at the end.
AI interview platforms. These came from qual. An interviewer holds a conversation with each respondent, follows up and asks why, and gives you transcripts and themes. They're often excellent at depth, but less so at the things a quantitative study depends on: quotas, weighting, a clean data file.
Full research platforms. These try to do both: a structured quantitative survey where the open-ended questions are conducted as short conversations, with the probing joined to the closed data, and analysis and reporting built in.
All three can be the right buy. The trouble is that a demo of any of them looks impressive for twenty minutes. The questions below are the ones that tell them apart, and they come from running studies, not from feature lists. We build one of these platforms, so read our answers knowing that. The questions are useful whichever way you go.
1. Does it probe the open ends, or just write the survey?
"AI-powered" can mean the software wrote your questionnaire, or that it summarised the answers after fieldwork, or that something is listening to each respondent while they answer. Those are very different products.
Ask to see a live open-ended question as a respondent would. Type a lazy answer, like "it's fine" or "good value". Does anything happen? If the next screen is just the next question, the AI isn't in the interview. A probing interviewer should ask about the vague answer, take a cue from a specific one, and stop once it has what it needs rather than asking three follow-ups by habit.
While you're there, ask what respondents can see and how they can answer. Redge shows images and video as stimulus, and respondents can answer on video as well as by typing.
2. Who writes the probe instruction, question by question?
This matters more than anything else on the list, and most demos skip it.
A question about why someone churned needs a different follow-up from a question about a brand image. If the platform probes every open end with the same generic prompt ("Can you tell me more?"), you get longer answers but not better ones. Ask whether you can write the brief for each question: what to probe, what to leave alone, when to stop. Then ask whether that instruction is exported with the data. When a client asks "what did you ask them, exactly?", you should be able to show the follow-up and the instruction that produced it.
In Redge, every open-ended question has its own probe brief, and probes are exported with the instruction that generated them.
3. Can you see the routing before you field?
Most survey errors are routing errors: a filter pointing at the wrong code, or a question nobody can reach. Ask how the platform shows you the questionnaire's logic before a single respondent sees it.
The best answer is that you can bring your questionnaire in as it already exists, usually a Word document, and see the routing as a map you can check, not as rules buried question by question. Redge imports a Word questionnaire and compiles it into a routing graph you can review before launch.
4. Can it hold an interlocked quota?
"Quotas" on a sales page can mean a cap on one variable. Real samples need interlocked cells: age by gender by state, say, each with its own target. Ask to see one set up, and ask what happens when a cell fills: is the respondent stopped at the right point, and is the quota counted on completes or on starts?
AI interview tools are often weakest here, because they came from a world of twenty interviews, not two thousand.
5. Does it work with your panel?
Unless you have your own list, your respondents come from a panel. That means the platform has to pass respondent and study IDs back and forth, send respondents back to the panel when they complete, screen out or hit a full quota, and reconcile the numbers at the end.
Ask which panel integrations exist today and which are "on the roadmap". Ask who places the sample order and who holds the panel relationship. Redge handles panel callbacks, respondent and study IDs, and quota-full redirects. Sample orders with our panel partner are prepared inside Redge, and live ordering end to end is rolling out now.
6. Do you get a transcript, or a dataset?
This is the question that separates qual tools from research platforms.
A transcript is lovely to read and hard to analyse at scale. A dataset is something you can cut by age, weight to the population, test for significance and join to the closed questions from the same respondent. Ask what the export looks like. Is there an SPSS file with labels? Are the open-ended answers coded into themes that line up with each respondent's closed answers? Does the platform report effective base sizes after weighting, or just the raw count?
Redge exports labelled SPSS (.sav) files, applies weighting with effective bases, runs significance tests, and builds an editable deck rather than a locked PDF. Every open-ended answer is coded.
7. Where does the data sit, in writing?
"Australian hosted" sometimes means the website is. Ask where the respondent data, the transcripts and the backups live, and get the answer in the contract or the security pack, not the sales call. Then ask the question that's new for AI tools: are responses used to train a model, yours or anyone else's?
Redge is hosted in Microsoft Azure's Australia East region, and respondent data isn't used to train models. Be just as direct about what's missing. We don't hold SOC 2 or ISO 27001 certification yet. If your procurement requires either one today, you need to know that before the demo, not after.
8. What does the quality checking actually catch?
Every platform says it removes bad respondents. Ask which checks it runs, whether you can see which respondents were flagged and why, and whether you can overrule a flag. Speeders and straight-liners are the minimum. Open-ended answers make the problem bigger and also make it easier to catch: copied text, answers that don't match the question, and the same answer from "different" people all show up in text.
Redge runs 17 quality checks and shows each flag against the respondent, so the decision to remove someone stays yours.
9. Can it track?
A one-off study and a tracker are different jobs. For tracking, ask whether a new wave is a copy of the last one (which drifts) or the same study fielded again (which doesn't), and whether results compare across waves without anyone rebuilding the tables.
Redge runs tracking studies in waves on a single study definition. Trended dashboards across waves are the next feature to ship.
10. What does the second study cost?
The first study is priced to win you. The tenth is the real price.
Ask how the price moves with volume. Is it per project, per complete, per seat or a subscription? What happens if you run twice as many studies next year? Are the open-ended questions priced like closed ones, or as a premium? And who pays for the panel: is sample marked up inside the platform price, or passed through so you can see it?
There's no right model. There's the model that matches how often you'll use it. If you run three studies a year, a per-project price is probably cheaper. If you run thirty, a subscription is probably cheaper. We wrote more about what research costs here in How much does market research cost in Australia?
11. Who can see what?
Research teams share. Agencies share with clients, and in-house teams share with stakeholders. Ask how access works: can you give a client access to one study without access to the rest, and can people sign in with the accounts they already use?
Redge supports Google and Microsoft sign-in and access limited to a single study. If your team works in Claude, there's also a hosted Redge connector, with a read-only option, so a study can be questioned from there.
12. What happens when you want a researcher alongside you?
This is the question buyers raise last and most often wish they'd asked first.
A platform can make you faster. It can't tell you that your sample frame is wrong for the decision, that the concept test needs a control, or that the board will ask a question your questionnaire can't answer. Some studies need a senior researcher, at least for a few hours: scoping, questionnaire review, a second read of the findings before they go upstairs.
There are three ways platform vendors handle this, and you should know which one you're buying:
- No one. The software is the product, and support means support for the software. That's fine if your team is already senior.
- An agency upsell. "Need help? Our research team can run it for you." Useful, but now you're buying a project, at project prices, and the platform becomes their tool, not yours.
- Expert hours on call. You keep running the study on your platform, and you buy senior researcher time in blocks for the moments that need it. The researcher works inside your study, not alongside it in a separate process.
Redge does the third, and we call it the bench. You buy expert hours in blocks, and the bench is managed by our Head of Research Design. Use them to scope a study, review a questionnaire before it fields, write the probe briefs with you, or sense-check findings. When the hours are used, the study is still yours, in your account, and you can take it from there.
Ask any vendor: if I want an experienced researcher for four hours next Tuesday, what happens, and what does it cost? The answer tells you whether you're buying a platform or an agency with a platform attached.
Which kind should you buy?
Honestly, it depends on the job.
Buy a classic survey platform if most of your work is closed questions, you already have the expertise in-house, and you need breadth of features and integrations more than depth of answers. They're mature for good reason.
Buy an AI interview platform if the job is qual at scale: exploring a problem before you know the right questions, with dozens or hundreds of conversations and no quotas, weighting or significance testing needed. They do depth well.
Buy a full research platform if your studies combine closed questions with open-ended ones and you need both to stand up to scrutiny together: quotas, panel, weighting, coded open ends, a deck at the end. That's what Redge is built for.
And here's when Redge is the wrong choice:
- You need to field in languages other than English. Redge is English only today.
- Your procurement requires SOC 2 or ISO 27001 now. We don't have either yet.
- You want someone else to do the whole thing. The bench is there for expert hours, not to run your research programme. If you want an agency, hire one.
If none of those rule us out, the quickest way to judge is on a study of your own. Book a 30-minute walkthrough and bring a questionnaire, or see what's in the platform now. If you already know you want a researcher in the room, scope it with the bench.
We update this guide when the market or our product changes, and we note the change at the top. If something here is wrong about us, tell us and we'll correct it publicly.