Research
Evaluators didn’t take this job to hunt a quote.
For evaluation shops. It reads interviews on a machine you own and shows the sentence. You write the codebook. Nothing goes to ChatGPT.
The interviews
A late theme means walking the interviews again.
Someone still hunts a quote they already read, then keys it into NVivo.
The finding still has to cite the line.
The donor still checks. A quote that isn’t in the interview does not ship.
How it works
Ask the question that takes hours.
It reads the interview and the codebook the way your people already do.
Ask where they said it. You see both files.
Interview 14 and the code. You still write the codebook.
Where did they mention the bus?
Interview 14, line 6
interview 14 · wave 2
Q: And after the route cut?
“We lost the bus in March, so the kids just stopped coming.”
The quote is on this page.
codebook v3
Code. Access / transportation.
Barriers to attendance after a service cut.
You still write this.
Same pass for a quote that isn’t in the transcript.
The IRB
The AI runs on this machine.
Interviews stay on it. You can point at the box.
Someone will ask where the AI runs. You point at this.
Nothing goes to the cloud.
If interview 14 is in ChatGPT, that is an IRB problem.
Don’t rip out NVivo. Point it at the interview share.
It does not train anyone else’s model. How the machine works →
Questions
Who still writes the codebook.
Where does the AI run?
On a machine you own. On-premises. We can ship one, or we put the AI on a machine you already have. You can point at it. It does not run at OpenAI.
Does anything go to ChatGPT?
No. Interviews stay on the machine you own. An IRB can ask where they went. You can show them the box.
If interview 14 is in ChatGPT, is that an IRB problem?
Yes. The interviews were given under an agreement. ChatGPT is not that agreement. Nothing goes there.
Does it write the codebook?
No. You write the codebook. It shows the sentence it used. A quote that isn’t in the interview does not ship.
Do we rip out NVivo?
No. Keep NVivo. Point it at the interview share. It sits next to the tools you already code from.
Does it train on our interviews?
No. It does not train anyone else’s model. Nothing leaves to improve a public model. If you stop, it comes off.
Book a time
Ask where it would sit.
A first call. Bring a codebook that’s sitting.