Will an ethics board accept an AI asking participants follow-up questions?

Approval comes from your board, and FormzoAI makes the application easier to write. Run the study as a one-page form and validated scales keep their exact wording; in a chat, scale items never get a follow-up. Follow-ups happen only on the open questions you choose, at a depth you set, and every follow-up asked is saved word for word with the response for your methods section.

Can it tell when a participant pasted an answer from ChatGPT?

When an open answer is generic, FormzoAI asks one follow-up about the participant's own experience, such as what happened the last time. Those details are hard to paste. Flags for responses that were too fast, pasted, straight-lined or LLM-like are coming at launch. Flagged responses are never deleted; you decide the exclusions.

How reliable is the AI coding compared with a human coder?

Treat the themes as a first pass you can audit: each opens to example answers next to the participant's own words, and the export keeps every original answer for a human coder. Per-answer codes you can override, exported as 0/1 columns for computing agreement in SPSS or R, are coming at launch.

Where is participant data stored, and can someone withdraw?

Each workspace stores data in the EU (Frankfurt) or the US. You set how long responses are kept, and you can delete a single participant's answers when they withdraw. A data processing agreement is available for your data management plan.

What does one study cost?

Joining the waitlist is free, and every plan starts with a 7-day trial. Starter is $19 a month for 500 responses. Pro is $49 a month for 3,000 responses. An Event Pass is $79 once for 60 days and 3,000 responses, which covers one study from pilot to final wave.

For online studies on Prolific, CloudResearch and MTurk

Open answers worth coding, from every participant you paid for.

Paste the link into Prolific. When an open answer is vague or generic, it asks one neutral follow-up about the participant's own experience. Then open answers are grouped into themes with counts, and you export an SPSS or R file.

Prolific study

Delaying GP visits, n = 400 · made from “A Prolific study on why UK adults put off seeing a GP: consent, a 12-item scale, two open questions”

Their chat

Think of the last time you put off seeing a doctor. Why did you wait?

People often delay care due to cost, time constraints and uncertainty about symptoms.

Follow-upWhat was the symptom in your case, and what made you wait?

A cough that lasted five weeks in March. Going in meant losing a shift's pay, so I waited until it kept me up at night.

What you get

Theme

Losing pay for time off work

96 of 406 people said something like this

“A cough that lasted five weeks in March. Going in meant losing a shift's pay, so I waited until it kept me up at night.”

Example answers, written for this page.

Where online studies go wrong.

Four tidy paragraphs in 38 seconds
Same structure, same “various factors”, no detail from anyone's life. In a 2025 study in Sociological Methods & Research (Zhang, Xu and Alvero), 34% of online participants said they used an LLM on open-ended questions.
812 open answers, one codebook
Two coders, a spreadsheet and NVivo, with the conference deadline three weeks out. Every change to the codebook sends you back to row one.
Data management plan, section 4
The ethics form wants the server location, the retention period and how a participant withdraws. You're answering it from three vendor help pages.

From one sentence to a coded dataset.

01
Describe the study in one sentence
“Why UK adults put off seeing a GP: consent first, a 12-item scale, two open questions.” You get the full instrument to edit. Run it as a one-page form and the validated scale keeps its wording exactly as published.
02
Paste the link into Prolific
PROLIFIC_PID, STUDY_ID and SESSION_ID are saved with each response, and the last screen returns people with your completion code. Only the open questions you choose get a follow-up, and only when an answer is vague.
03
Export, check the coding, analyse
An SPSS or R file with a codebook, open answers grouped into themes with quotes, and every follow-up question saved with its response for the methods section.

Open answers, grouped into themes.

Open answers are grouped into themes, each with how many participants said it and the quote that shows it best. Coding into your own codebook, exported as 0/1 columns you can check and recode by hand, is coming at launch.

Delaying GP visits · answers

0 in

Built around the way a study runs.

Prolific IDs and completion codes: Any value on the link, like PROLIFIC_PID or a CloudResearch worker ID, is saved with the response. The end screen shows your completion code or redirects straight back.
Flags on low-quality responses: Too fast, straight-lined, pasted or written like an LLM: each response is flagged, never deleted, and the flag is a column you filter on. Coming at launch.
Your own codebook: Write your categories or start from suggested ones. Every open answer gets codes you can check and override, exported as 0/1 columns. Coming at launch.
SPSS and R, with a codebook: Numeric codes, variable names, value labels and a syntax file. Each follow-up that was asked is saved with the response word for word.
Fixed scales, probed open items: Set follow-up depth from 0 to 3 per open question; scale items never get one. Run the whole study as a one-page form to keep validated wording exact.
Answers for the ethics form: Storage in Frankfurt or the US per workspace, a retention period you set, one participant's data deleted on withdrawal, and a DPA to attach.

Fair questions.

Will an ethics board accept an AI asking participants follow-up questions?: Approval comes from your board, and FormzoAI makes the application easier to write. Run the study as a one-page form and validated scales keep their exact wording; in a chat, scale items never get a follow-up. Follow-ups happen only on the open questions you choose, at a depth you set, and every follow-up asked is saved word for word with the response for your methods section.
Does it work with Prolific, CloudResearch and MTurk?: Yes. Put the participant, study and session IDs on the survey link and they are saved with each response. The last screen shows your completion code or redirects straight back to the panel. Participants answer in the browser without an account, as a short chat or a one-page form.
Can it tell when a participant pasted an answer from ChatGPT?: When an open answer is generic, FormzoAI asks one follow-up about the participant's own experience, such as what happened the last time. Those details are hard to paste. Flags for responses that were too fast, pasted, straight-lined or LLM-like are coming at launch. Flagged responses are never deleted; you decide the exclusions.
How reliable is the AI coding compared with a human coder?: Treat the themes as a first pass you can audit: each opens to example answers next to the participant's own words, and the export keeps every original answer for a human coder. Per-answer codes you can override, exported as 0/1 columns for computing agreement in SPSS or R, are coming at launch.
Where is participant data stored, and can someone withdraw?: Each workspace stores data in the EU (Frankfurt) or the US. You set how long responses are kept, and you can delete a single participant's answers when they withdraw. A data processing agreement is available for your data management plan.
What does one study cost?: Joining the waitlist is free, and every plan starts with a 7-day trial. Starter is $19 a month for 500 responses. Pro is $49 a month for 3,000 responses. An Event Pass is $79 once for 60 days and 3,000 responses, which covers one study from pilot to final wave.

Get in before your next study goes live on Prolific.

Get early access

Free to join. Spots open in batches.

Open answers worth coding, from every participant you paid for.

Prolific study

Delaying GP visits, n = 400 · made from “A Prolific study on why UK adults put off seeing a GP: consent, a 12-item scale, two open questions”

Their chat

Think of the last time you put off seeing a doctor. Why did you wait?

People often delay care due to cost, time constraints and uncertainty about symptoms.

Follow-upWhat was the symptom in your case, and what made you wait?

A cough that lasted five weeks in March. Going in meant losing a shift's pay, so I waited until it kept me up at night.

What you get

Theme

Losing pay for time off work

96 of 406 people said something like this

“A cough that lasted five weeks in March. Going in meant losing a shift's pay, so I waited until it kept me up at night.”

Example answers, written for this page.

Where online studies go wrong.

Four tidy paragraphs in 38 seconds

Same structure, same “various factors”, no detail from anyone's life. In a 2025 study in Sociological Methods & Research (Zhang, Xu and Alvero), 34% of online participants said they used an LLM on open-ended questions.

812 open answers, one codebook

Two coders, a spreadsheet and NVivo, with the conference deadline three weeks out. Every change to the codebook sends you back to row one.

Data management plan, section 4

The ethics form wants the server location, the retention period and how a participant withdraws. You're answering it from three vendor help pages.

From one sentence to a coded dataset.

Describe the study in one sentence

“Why UK adults put off seeing a GP: consent first, a 12-item scale, two open questions.” You get the full instrument to edit. Run it as a one-page form and the validated scale keeps its wording exactly as published.

Paste the link into Prolific

PROLIFIC_PID, STUDY_ID and SESSION_ID are saved with each response, and the last screen returns people with your completion code. Only the open questions you choose get a follow-up, and only when an answer is vague.

Export, check the coding, analyse

An SPSS or R file with a codebook, open answers grouped into themes with quotes, and every follow-up question saved with its response for the methods section.

Built around the way a study runs.

Prolific IDs and completion codes

Any value on the link, like PROLIFIC_PID or a CloudResearch worker ID, is saved with the response. The end screen shows your completion code or redirects straight back.

Flags on low-quality responses

Too fast, straight-lined, pasted or written like an LLM: each response is flagged, never deleted, and the flag is a column you filter on. Coming at launch.

Your own codebook

Write your categories or start from suggested ones. Every open answer gets codes you can check and override, exported as 0/1 columns. Coming at launch.

SPSS and R, with a codebook

Numeric codes, variable names, value labels and a syntax file. Each follow-up that was asked is saved with the response word for word.

Fixed scales, probed open items

Set follow-up depth from 0 to 3 per open question; scale items never get one. Run the whole study as a one-page form to keep validated wording exact.

Answers for the ethics form

Storage in Frankfurt or the US per workspace, a retention period you set, one participant's data deleted on withdrawal, and a DPA to attach.

Fair questions.

Will an ethics board accept an AI asking participants follow-up questions?: Approval comes from your board, and FormzoAI makes the application easier to write. Run the study as a one-page form and validated scales keep their exact wording; in a chat, scale items never get a follow-up. Follow-ups happen only on the open questions you choose, at a depth you set, and every follow-up asked is saved word for word with the response for your methods section.
Does it work with Prolific, CloudResearch and MTurk?: Yes. Put the participant, study and session IDs on the survey link and they are saved with each response. The last screen shows your completion code or redirects straight back to the panel. Participants answer in the browser without an account, as a short chat or a one-page form.
Can it tell when a participant pasted an answer from ChatGPT?: When an open answer is generic, FormzoAI asks one follow-up about the participant's own experience, such as what happened the last time. Those details are hard to paste. Flags for responses that were too fast, pasted, straight-lined or LLM-like are coming at launch. Flagged responses are never deleted; you decide the exclusions.
How reliable is the AI coding compared with a human coder?: Treat the themes as a first pass you can audit: each opens to example answers next to the participant's own words, and the export keeps every original answer for a human coder. Per-answer codes you can override, exported as 0/1 columns for computing agreement in SPSS or R, are coming at launch.
Where is participant data stored, and can someone withdraw?: Each workspace stores data in the EU (Frankfurt) or the US. You set how long responses are kept, and you can delete a single participant's answers when they withdraw. A data processing agreement is available for your data management plan.
What does one study cost?: Joining the waitlist is free, and every plan starts with a 7-day trial. Starter is $19 a month for 500 responses. Pro is $49 a month for 3,000 responses. An Event Pass is $79 once for 60 days and 3,000 responses, which covers one study from pilot to final wave.