Survey platform for research teams

Every data point comes with its proof.

Field, phone, online and voice surveys in one place. Behind every answer sits evidence you can open: who gave it, where and when, what was actually said — and how far to trust it.

Request a walkthrough See the proof

Every screenshot on this page is a real screen, showing demo data.

AI check Click to enlarge

AI check

Mismatch

Entered by the interviewer: Very satisfied
Heard on the recording: Dissatisfied

“Not very. Honestly, I'm dissatisfied, the connection keeps dropping.”

GPS matches · ±10 m

Scroll down

01 / 03

Are you truly certain your data was actually collected?

A spreadsheet of 1,200 interviews lands in your inbox. Row 847 looks exactly like row 846. Nothing in the file says whether either interview ever took place.

02 / 03

What happens when that data turns out to be wrong or manipulated?

A decision has already been made on it, and the report has already gone out. Back-checks usually reach 5–10% of interviews, weeks later, when the team has moved on and nothing is left to examine but memory.

interviews_final_v3.xlsx1,200 rows

#InterviewerDurationSatisfaction

843Olivia Reid14:20Satisfied

844Liam Foster12:05Neutral

845Noah Bennett12:30Very satisfiedFilled in at home?

846Emma Clarke13:40Satisfied

847Emma Clarke11:52Very satisfiedTyped ≠ said?

848Olivia Reid15:02Dissatisfied

849Liam Foster3:10SatisfiedToo fast?

850Noah Bennett13:15Neutral

03 / 03

What if every data point could be proven?

Who answered, where and when, what was actually said, and how far to trust it — attached to every row, while the fieldwork is still running.

See how it works

interviews_final_v3.xlsxwith evidence

#SatisfactionGPSRecordingTrust

843Satisfied±8 m14:2096

844Neutral±11 m12:0591

845Very satisfiedOutside area12:3041

846Satisfied±6 m13:4094

847Very satisfied±9 mMismatch38

848Dissatisfied±10 m15:0295

849Satisfied±12 m3:10 · too fast52

850Neutral±8 m13:1590

How it works

Not “trust us”, but proof your research and studies can stand on.

Evidence is captured as part of ordinary fieldwork, not as extra admin afterwards. Every answer carries four kinds of proof.

Interview 847 · Emma Clarke · Very satisfied

Respondent

R-0847 · from the drawn sample

Who

Was this the person we actually sampled?

The respondent came from the drawn sample, and was interviewed by the person it was assigned to — not by whoever happened to be nearby.
Location & time

GPS ±9 m · inside the area · 13:49

Where and when

Was the interviewer really there?

A GPS point with accuracy in metres, the start time, the length of the interview, and whether it falls inside the assigned area.
Recording

11:52 · consent read aloud · 1 mismatch

What was said

Is the recorded answer the real answer?

The conversation is recorded after consent, written out, and compared question by question with what the interviewer typed in.
Trust score

38 · held for review

How far to trust it

Should I rely on this response at all?

A trust score built from 20 signals, plus a record of who reviewed it and what they decided.

Four ways to collect

Every method has its own fears. And its own proof.

A phone call has no location; an online form has no voice. Below, each method is taken one at a time: what a research buyer worries about, the feature that answers it, and the real screen where that proof lives.

01

Field survey

Face to face, with no supervisor in the room.

Advanced only

The fearThe interview never happened, or happened with the wrong person.

6 features that answer it
See the proof
02

Phone survey

No face to face — a recording of the conversation instead.

Basic and Advanced

The fearThe call was logged but never made, and consent cannot be shown.

3 features that answer it
See the proof
03

Online link survey

No interviewer at all. A respondent and a link.

Basic and Advanced

The fearA bot, a duplicate, or someone clicking through without reading.

4 features that answer it
See the proof
04

Voice Survey

Self-completed, but answered out loud.

Basic only

The fearOpen answers flattened into something the respondent never meant.

4 features that answer it
See the proof

Method 01 of 4Advanced only · 6 features that answer it

Field survey

Face to face, with no supervisor in the room.

The fearThe interview never happened, or happened with the wrong person.

FieldPhoneOnline linkVoice

Method 01 · Field survey

Face to face, with nobody watching. Here is what you can still check.

Field interviews carry the highest risk and the highest cost: the interviewer is alone, the respondent is hard to reach again, and the spreadsheet looks the same either way. GatherInsight attaches evidence to every visit, while the team is still in the field.

Fear → proof

Keep scrolling — the proofs move sideways → Swipe sideways →

01 / 06

The fear

The interview was never actually taken — the form was filled in at home.

The proof

A GPS point on every interview

When the interviewer starts an interview, the app takes a location fix and stores it with the response, together with its accuracy in metres and the time. Every visit becomes a point on the map — including the ones that ended in a refusal.

Coloured by visit outcome: completed, refused, not eligible, unreachable
Filter by interviewer, review status or AI mismatch
Click a point to open that respondent and their answers

Response Map — 40 demo interviews, coloured by how each visit endedDemo data

02 / 06

The fear

The interviewer stayed outside the area they were sent to.

The proof

GPS rules you set — and can change mid-fieldwork

Choose whether location is off, taken at the start of each interview, or tracked while the app is open. Draw the assignment area as a radius or polygon, set the accuracy you accept, and reject faked locations.

Outside the area: flag the response, or refuse it
Faked locations reported by the phone are refused
Scheduled pauses, judged against the rule in force at the time

Field Control — when GPS is taken, and what the Live Map can showDemo data

03 / 06

The fear

Problems surface only after fieldwork has closed.

The proof

A live map and alerts while the team is still out

See where each interviewer last reported, their trail, and who has gone quiet. Alerts are raised for interviewers who go offline, step outside their area, run low on battery or use a faked location.

Active, pending, offline and outside-the-area at a glance
An 8-hour trail per interviewer
Alerts and messages handled in the same place

Live Map — four interviewers, their latest interview points and trailsDemo data

04 / 06

The fear

The easiest person to reach was interviewed, not the one who was sampled.

The proof

A sample drawn first, then handed out

The sample is drawn from the frame before fieldwork, and each interviewer receives their own list. They cannot pick who to visit, and the office can see how far each one has got.

Targets and progress per interviewer
Move the remainder when the plan changes, without breaking history
Reserves released only when a unit is truly unusable

Assignments — who holds which part of the sample, and how much is doneDemo data

05 / 06

The fear

What was typed is not what the respondent said.

The proof

The recording, written out, and checked answer by answer

After the respondent agrees, the interview is recorded and transcribed. The AI compares every closed answer with what the recording says and attaches the exact quote. In the demo, the interviewer entered “Very satisfied”; the recording says “dissatisfied”.

Match, mismatch, listen again, or not clearly heard
One click plays the exact moment of the quote
Approve, send back or reject — always a person's decision

Quality Review — recording, transcript and the AI check on one interviewDemo data

06 / 06

The fear

One interviewer quietly cuts corners while the totals look fine.

The proof

Performance per interviewer, side by side

Output, average duration, visits without an interview, flagged interviews and a quality score per interviewer, in one table. An outlier stands out long before the report is written.

Average duration compared across the team
Flagged, rejected and redone interviews per person
Export to CSV for payment and follow-up

Surveyor Performance — output, duration, flags and quality per interviewerDemo data

Method 02 of 4Basic and Advanced · 3 features that answer it

Phone survey

No face to face — a recording of the conversation instead.

The fearThe call was logged but never made, and consent cannot be shown.

FieldPhoneOnline linkVoice

Method 02 · Phone survey

No face to face — so the voice and the consent become the proof.

Phone interviews remove the trip, but not the doubt. Did the call happen? Did the respondent agree? Was the answer written down as it was said? Calls run through WhatsApp Web or a call link from the browser, and every one leaves a trail.

Fear → proof

Keep scrolling — the proofs move sideways → Swipe sideways →

01 / 03

The fear

Calls were logged that never took place.

The proof

Every call runs through the Call Console

The caller starts each call from the console — through WhatsApp Web or by sending a call link — and fills in the answers while the conversation is recorded. Calls that do not connect are recorded as outcomes, separate from the data.

An outcome for every attempt: not reachable, wrong number, declined, completed
Who called, when, and for how long
A call link needs no app and costs nothing per minute

Call Console — start a call, with the office instructions beside itDemo data

02 / 03

The fear

Nobody can show the respondent agreed to be recorded.

The proof

Consent read aloud — on the recording itself

The office writes the consent statement; the caller reads it at the start, and recording begins before it is read, so the agreement is part of the recording. On a call link, the respondent also taps agree on their own screen.

Recording after consent cannot be switched off
Whoever declines is not recorded
Caller instructions and calling hours set by the office

Phone settings — channel, consent statement and caller instructionsDemo data

03 / 03

The fear

The answer written down is not the answer given.

The proof

The same answer-by-answer check as in the field

Call recordings are transcribed and checked the same way as field interviews: each closed answer against what was said, with the exact quote. The screenshot comes from a field interview; the check is identical for calls.

A mismatch is the card a reviewer opens first
Unclear answers are marked “listen again”, never guessed
Recordings play from each response, and every playback is logged

AI check — one mismatch, one to listen again, three matchesDemo data

Method 03 of 4Basic and Advanced · 4 features that answer it

Online link survey

No interviewer at all. A respondent and a link.

The fearA bot, a duplicate, or someone clicking through without reading.

FieldPhoneOnline linkVoice

Method 03 · Online link

No interviewer at all. So each response has to speak for itself.

Self-completed surveys bring a different set of fears: bots, duplicates, a link forwarded to the wrong person, and people clicking through without reading. Every response is scored, explained and — if you choose — held back until a person decides.

Fear → proof

Keep scrolling — the proofs move sideways → Swipe sideways →

01 / 04

The fear

A bot or a duplicate slipped in among real respondents.

The proof

A trust score that explains itself

Every response starts at 100 and loses points for what it shows: a hidden honeypot field filled in, answers identical to another respondent, the same device submitting twice, a finish far faster than everyone else. The reasons are written next to the score.

20 signals, each with a published weight
Normal 80–100, needs checking 46–79, doubtful 0–45
A shared device can look like a duplicate — so it is a hint, not proof

One response explained: 28, doubtful — too fast, same device, far faster than the restDemo data

02 / 04

The fear

People rushed through without reading.

The proof

Speed and patterns, measured against this survey's own respondents

The platform learns what a normal respondent does in this survey — how long they take, how many questions they answer — and flags what falls far outside it. Straightlining, always-the-same-option and zig-zag patterns are picked up too.

Comparisons start only after 30 responses, so early ones are not judged against noise
The most common signals shown at the top
A needs-a-decision queue for the responses worth a look

Response Quality — what normal looks like, the trust distribution, the queueDemo data

03 / 04

The fear

The link was forwarded, or one person filled it in twice.

The proof

Control who can answer, and how often

Limit responses per browser, cap the total, and close the link on an end date. In Structured surveys, each respondent gets a personal link or access code from the sample, so a forwarded link is visible for what it is.

One response per browser (a browser-side guard)
A total limit and an expiry date
Personal links and access codes in Structured surveys

Survey settings — submission limits and link expiryDemo data

04 / 04

The fear

Suspect answers quietly end up in the results.

The proof

Hold mode: nothing doubtful counts until a person decides

Choose what the system does with a low score: flag it and keep it, or hold it out of the analysis until someone approves or rejects it. Add a too-fast limit — the platform suggests one — and your own consistency rules.

Flag only, or hold until reviewed
A suggested too-fast limit, worked out from your own data
“If A then B” consistency rules you write yourself

Response quality settings — score threshold, hold mode, consistency rulesDemo data

Method 04 of 4Basic only · 4 features that answer it

Voice Survey

Self-completed, but answered out loud.

The fearOpen answers flattened into something the respondent never meant.

FieldPhoneOnline linkVoice

Method 04 · Voice Survey

Respondents answer out loud. Their own words are kept.

Open answers are where meaning gets lost — typed in a hurry, or summarised by someone else. In a Voice Survey, respondents speak their answers. The recording is kept, written out, and summarised with checks that keep the summary honest.

Fear → proof

Keep scrolling — the proofs move sideways → Swipe sideways →

01 / 04

The fear

Open answers get flattened into something the respondent never meant.

The proof

An AI version per answer — one click from the original words

Each spoken answer gets a short summary and its key points. One click switches back to the full transcript, so a reader always sees what the summary came from.

Summaries only for answers long enough to summarise
Shown inside each question card
Also under each answer when you read one respondent

Question cards — the AI version, and the switch to the transcriptDemo data

02 / 04

The fear

Audio that is silent, unclear or misheard goes unnoticed.

The proof

Every recording, written out and correctable

All spoken answers in one place: the recording, the transcript, its length and language. A person can correct a transcript — the correction is marked — and silent, too-short and hard-to-transcribe answers are flagged.

Filter by waiting, written down, low confidence, nothing heard
Words that come up often across all answers
Corrected text replaces the machine text — visibly

All spoken answers — recordings, transcripts, frequent wordsDemo data

03 / 04

The fear

The AI summary quietly invents a number or a name.

The proof

Summaries checked against the transcript

Numbers, quotes and names in a summary are checked back against the transcript; anything that cannot be found is marked. Personal data is masked before any text is sent. The transcript and the recording remain the reference.

Summaries are stored with the text they came from
Correct the text, and the summary is marked out of date
The audio itself is never sent to the AI

An AI version, with the note that the text and the recording remain the referenceDemo data

04 / 04

The fear

Respondents never really agreed to be recorded — or the audio is kept forever.

The proof

Consent first, and audio kept only as long as you decide

Consent is always on in a Voice Survey: respondents read exactly what will happen to their voice before anything is recorded. Set how long recordings are kept, or delete each one as soon as it has been written out.

Respondents may type instead of speaking
A microphone check before the first question
Retention from 7 to 730 days, or delete after transcription

Voice settings — language, length, typing fallback, retention and consent textDemo data

Main features

Main features, method by method

Each method has its own working screens. Below are the features used most for each one, with their real screens. The label above every image names the screen or view it shows.

Field survey

Advanced only

Live tracking

See the field team as it works.

Each interviewer's latest position, an eight-hour trail and who has gone quiet, on one map. Alerts fire for going offline, leaving the area, low battery, faked locations and targets reached.

Live Map · latest positions and trails

Live tracking Click to enlarge

Technical analysis

The whole fieldwork on one page, six ways to read it.

Sample, target, margin of error, data quality and AAPOR-style rates at the top; progress by day and by interviewer; then every question as bar, column, donut, table or trend. A view can be set for all questions at once or per card.

Views

Overview · sample, target, margin of error
AAPOR-style rates & progress
Bar view
Column view
Donut view
Table view
Trend view

Route history

Where an interviewer went, minute by minute.

Distance, active time and stops per day; a chronology from the start of tracking to the last position; and a route you can replay, with the interview points along the way.

Route History · replayable

Route history Click to enlarge

Responses & quotas

Every interviewer's target and every interview, in one table.

Quota cards show who is on track; the table below records the GPS check, review status, map location, recording and AI check of every interview.

Responses · quota monitoring

Responses & quotas Click to enlarge