Unlike AI tools that stop at interview + transcript, Mira is a full AI researcher — plans studies, recruits globally (100M+ panel, 120 countries), runs dynamic interviews with intelligent probing, and uniquely captures what participants say AND feel via real-time facial coding, voice emotion AI, and webcam eye tracking. Extracts themes, generates insights, and produces research reports automatically. 17 patents. 70+ languages. Trusted by Unilever, Nestlé and 150+ global brands. $25M Series B.
Forum Threads
The say/feel gap in research; how big is the problem actually, and what do you do about it?
There's a well-documented phenomenon in consumer and user research that most practitioners know intuitively but rarely name directly: people don't report their experience accurately.
Not because they're dishonest. Because self-reporting is hard. In the moment of an interview, participants are performing a version of themselves. They round off hesitation. They describe their behaviour more charitably than it actually was. They say "yes, I'd probably use this" when what they felt was closer to "maybe, under the right circumstances, if the price were different." The social pressure of being in a conversation, even with an AI, shapes what gets said.
This is what we call the Say-Do Gap. The distance between what someone tells you and what they actually feel or do.
It shows up everywhere. In concept tests where participants say they love a product they'd never buy. In usability sessions where someone says "this makes sense" while visibly struggling. In brand perception studies where stated attitudes don't match purchasing behaviour.
What do transcripts miss that sends you back to the recording?
Nine years of working closely with insights teams across CPG, BFSI, and tech has given us a consistent observation: the transcript is where researchers start, but the recording is where they go when the transcript isn't enough.
They go back to hear how someone said something. To check whether the pause before an answer was genuine uncertainty or just thinking time. To see whether the person's expression changed when they described a feature they claimed to like. To catch the moment when engagement visibly dropped when they stopped leaning forward, when their eyes moved away from the screen even while their words stayed polite and positive.
That gap between the transcript and the recording is where a significant amount of research insight lives. And it's also where the most time gets consumed in manual analysis. Researchers who run 20 or 30 interviews know what it's like to sit through hours of footage looking for the three or four moments that actually matter.
This observation that the transcript captures what was said but not how or with what underlying feeling is fundamentally what drove us to build the emotion and behavioural layer into Mira. Not to replace researcher interpretation, but to give researchers a way to find the right moments faster.
I'm genuinely curious: what's the signal you find yourself wishing you had after you've finished going through a set of interviews? What consistently drives you back to the recording that the transcript alone doesn't give you? And have you found any workarounds annotation practices, tagging systems, collaborative review setups that help you capture more of what the transcript loses?
What would actually make you trust an AI moderator's findings?
We launched Mira on Product Hunt today, and the conversations in the comments have been the most interesting part of the whole day.
Researchers asked about probe neutrality whether the follow-up questions an AI asks mid-interview can lead a participant, rather than uncover them. They asked about cultural calibration, whether emotion models trained largely on Western data can accurately read a participant in Jakarta or Nairobi. They asked whether the say/feel mismatch gets surfaced as raw evidence or quietly resolved into a single confidence score.
These are serious questions. And they made me realize something: the bar for trust in AI-moderated research is fundamentally different from other AI tools.
If a writing assistant gets something wrong, you catch it before you publish. If an AI coding tool hallucinates, your tests fail. But if a research tool misreads how participants felt during a concept test, and that feeds into a product decision, the error is invisible. By the time the product ships and the market responds, the research moment is long gone.
I have run plenty of user interviews by hand, so the moderation part I get. The reads-how-people-feel part is where I would love more detail: inferring emotion from voice or wording is powerful, but it is also the kind of signal that can mislead a decision (someone nervous is not someone negative). How do you present that layer to the researcher, as a hint to probe further or as scored data? The difference feels important.
Mira
@virko_kask Virko, you've named the exact distinction that separates useful emotion data from noise — and it shaped how we built the output layer.
The short answer: it's a hint to probe further, not a verdict.
We made a deliberate decision not to present emotion as scored data; the researcher is supposed to act on it directly. A nervousness signal doesn't get labeled "negative response." It gets flagged as a moment worth returning to, a timestamp when verbal and non-verbal signals diverged or when an emotion was sustained long enough to be meaningful.
In the report, it looks like a highlighted clip with the emotional trace underneath it. The researcher sees: what was said, what the face and voice were doing at that moment, and how long it lasted. No single-number sentiment score. No resolved verdict. The interpretation is yours.
Where we do add structure, we distinguish between transient signals (a quick flash of surprise, a one-second hesitation) and sustained patterns (3–5 seconds of consistent emotional signal across modalities). Transient signals are surfaced as context. Sustained patterns are what get elevated as findings. That threshold matters for exactly the reason you named, nervousness is not negativity, but sustained disengagement during a key product moment is worth a conversation.
The emotional layer is evidence, not a conclusion. Would love to show you what this looks like on a real study, especially with your interview background. I think you'd spot things most researchers miss. Happy to set up a call → https://www.entropik.io/book-demo
@mridhu_varshini_ Thank you for the detailed answer, this is genuinely thoughtful design. The transient vs sustained threshold is the right cut, and surfacing the divergence as a clip with the trace under it, instead of a number, is exactly the difference between evidence and verdict. I am heads-down on my own launch for the next two weeks, but I would honestly enjoy seeing this on a real study after that. Good luck with the launch, Mira deserves the attention it is getting.
Mira
@virko_kask Thank you, Virko, this genuinely made my day :) The way you framed it, "evidence not verdict," is exactly the distinction we spent a long time trying to get right in the design. It's validating to hear it land that way from someone who's run as many interviews as you have.
And yes!!!! Mira deserves this attention, and honestly, so do you for asking the questions that pushed us to articulate it properly. This launch has been our killer ship moment, years of building, finally in front of the right people.
Two weeks go fast. Ping me when you're out the other side, and we'll get something on the calendar. I genuinely think you'll have interesting observations after seeing it in a real study.
Congratulations on launch! A lot of the questions are about accuracy and privacy, so I'll ask a different one: how does the emotion reading hold up across cultures and languages? People show feelings differently depending on background, and my audience is fairly reserved by nature. Does the model account for that, or is it mostly calibrated to more expressive participants?
Mira
@alieksia Anastasiia, this is one of the most important questions anyone can ask about emotion AI — and one we've had to answer in practice across 150+ brands in markets where emotional expressiveness varies significantly.
A few things we do specifically for this:
Individual baseline calibration is not a universal norm. At the start of every session, Mira establishes a neutral baseline for that participant. What matters is deviation from their baseline, not deviation from an average. A reserved participant who shifts even slightly is flagged because for them, that shift is meaningful. You're not comparing them to a participant in a different country who naturally gestures more.
Multimodal cross-referencing helps here significantly. Voice tone and pacing convey emotional signals across cultures in ways that facial expressions alone don't. Someone from a more reserved cultural background who shows minimal facial movement will often convey more signals in their voice, hesitation, changes in pacing, or a drop in confidence. Mira reads both simultaneously.
We've trained specifically on diverse participant pools. Our models have been built on research across APAC, MENA, Europe, and the Americas, not just Western expressive populations. That said, we're transparent: some cultural nuances remain an active area for improvement, and we'd rather flag lower-confidence signals than assert false precision.
The output accounts for this contextually. Signals are always shown with confidence scores and duration markers. A researcher working with a reserved audience can set their own threshold for what's worth acting on.
If your audience is particularly reserved by nature, I'd love to show you a session with that kind of participant profile, specifically, it's actually where the multimodal approach shows its biggest advantage over facial-only tools.
Worth a call? https://www.entropik.io/book-demo
@mridhu_varshini_ thank you for taking the time to explain this so fully. The baseline calibration and the multimodal reading for reserved participants answer my concern well. I'll keep Mira in mind for when I'm at that research stage.
Mira
@alieksia , sure. We can reconnect anytime when the research stage comes in. Happy to be connected :)
"Reads how people feel" is the interesting (and risky) part. When it detects hesitation mid-interview, does it adapt its questioning in the moment or just annotate for the researcher afterward? I'm running beta-user interviews right now and what I always miss is what people didn't say, I would love to know if you surface that.
Mira
@chielephant Anthony, you've named the exact thing most AI interview tools get wrong: they annotate after, when the moment is already gone.
Mira adapts in the moment. When it detects hesitation, an emotional shift, or a mismatch between what's said and how it's said, it probes during the conversation rather than as a post-hoc tag. "You paused there. Tell me more about what was going through your mind." The follow-up happens while the participant is still in the thought, not after they've moved on.
On what people didn't say: this is where the multimodal layer earns its place. Mira flags moments when vocal confidence dropped without a corresponding verbal signal, when attention shifted during a key question, and when facial affect changed but the spoken answer was flat. Those are the gaps, the things a transcript alone will never show you. They surface in the report as flagged moments with timestamps, so you can go directly to "here's where something was happening that wasn't being said."
For beta-user interviews specifically, that layer tends to be most valuable on product moments — the exact second someone's expression changes when they hit a friction point, even while they're still saying "yeah, this makes sense."
Would love to show you what this looks like on a real session, especially with beta interviews, the signal density is usually high → https://www.entropik.io/book-demo
How do you prevent Emotion AI from overinterpreting facial expressions or cultural differences in user reactions?
Mira
@robert_dimla We avoid cultural overfitting by rejecting static global priors and instead computing affect as temporal deltas against a per user baseline. Our attention layer gates visual action units against orthogonal acoustic and semantic vectors down-weighting isolated facial anomalies as noise unless validated across channels or via active verbal probing.
Two things... first, how does the "Real time facial encoding" work? Secondly, this seems like an interesting idea, but im curious regarding how you plan to deal with jobseekers, inevitably, despising software like this. Sure businesses trying to cut costs would love something like this, but on the consumer/job-seeker side how will you mitigate and ensure a candidate doesn't feel like a robot is screening them.
Mira
@lucapiekarski Great questions!
Real-time facial encoding: We track facial action units (the individual muscle movements that make up expressions) frame by frame during the session
On the job-seeker concern: Totally fair question, but MIRA isn't built for hiring or candidate screening. We're purpose-built for user research, think usability testing, customer interviews, product feedback sessions. Participants are opted-in research subjects, not candidates being evaluated for a job. The goal is understanding genuine reactions to a product or concept, not judging a person.
The idea of AI moderating interviews based on how people feel is interesting because tone can be pretty subjective. I’m curious how you balance consistency with making the conversation feel natural rather than scripted.
Mira
@amjad_shaik
Great question. To clarify how this actually works: Mira doesn't change questions based on detected emotion the conversation flow adapts only based on what the participant says (their actual answers), which keeps it natural rather than scripted.
The emotion layer runs separately, in parallel. It's predicting/measuring how someone feels while a question is being asked or answered, and that signal is surfaced to the researcher as a distinct layer of insight not fed back into the conversation logic itself.
So the moderation stays consistent and conversational because it's driven by response content, while the emotional analysis exists purely as a research lens on top, for understanding reactions after the fact. Two separate systems doing two separate jobs.
The Say-Do gap framing is sharp, and running recruit → moderate → analyze → report as one agent is the ambitious part. For a team that already has research infra, two setup questions: can I bring my own recruited participants into a Mira study, or is moderation locked to your 100M panel? And can I export the raw session data — transcripts plus the behavioral signals you capture — into our own repo, or does the analysis stay inside Mira's dashboards?
Mira
@noctis06 Good questions, both get asked by every team with existing infra, so let me be precise rather than salesy.
Bring your own participants: yes. Moderation isn't locked to our panel — you can pipe in your own recruited participants and Mira runs the same moderate → analyze pipeline on them. Two ways teams do this depending on setup:
Link-based: we generate a session link per participant, you distribute through your own recruiting/incentive flow, they land in the same moderated session experience.
API-based: if you've got a recruiting/panel system already, you can push participants in via API and we handle scheduling + session invites from there.
On export: yes, and I'll be specific about what "raw" means. CSV export gives you:
Full transcripts (turn-by-turn, timestamped)
The say-do gap annotations themselves (where the model flagged claimed vs. observed behavior divergence, with the reasoning trace)
Nothing stays trapped in the dashboard-only view. The dashboard is a convenience layer, not the source of truth — the source of truth is the same data you'd get in the export.