Code open-ended survey responses against one codebook
Coding free-text answers by hand means building a codebook, then reading every answer and tagging it. This tool does both steps in that order: it proposes a codebook from a sample of your answers, or takes yours, and then codes every answer with up to three codes in a new column. You get the coded file and a codebook table with counts and example quotes. The first answers are free; up to 20,000 answers are $5.
- Codebook first: at most 14 codes plus "Other", each with a one-sentence description
- Built from up to 150 answers spread across the file, or your own codes
- Every answer coded with one to three codes from that list
- A codebook CSV with counts, share of answers and two example quotes per code
- Up to 20,000 answers for $5, paid once
Hand coding, a chatbot and this tool
Each has a place. The difference is how many answers get coded, and whether the same codebook is used for all of them.
| Your situation | A chatbot | This tool |
|---|---|---|
| Exploring 50 answers for ideas | Good at suggesting themes | Also works; free for small files |
| Coding 2,000 answers for a report | Summarizes; rarely codes each answer | Codes every answer, 40 at a time |
| Keeping last year's codebook | Tends to add new codes | Uses only your codes, plus "Other" |
| Showing evidence for each code | Quotes a few answers it remembers | Two example quotes per code, taken from coded rows |
How the codebook is built
The tool takes up to 150 answers evenly spread across the file, not just the first ones, and asks the model for at most 14 codes that do not overlap, each with a description of which answers belong to it. "Other" is added for answers that fit none. The codebook appears on screen with the first coded answers before you pay, so you can replace it with your own list if it is too broad.
Giving the survey question helps: "What could we do better?" and "What did you like most?" call for different codes even when the words in the answers overlap.
What you can report from it
The codebook CSV gives, for each code, the number of answers, its share of all answers and two example quotes. Because an answer can carry up to three codes, the shares can add up to more than 100%; say so in the report. The coded file keeps your respondent IDs and other columns, so you can cross-tabulate codes against age group or region in your own tool.
How the tool works, what it costs and how your files are handled is on the Code Survey Responses page; prices for every tool are on the pricing page.
Last updated
Questions and answers
Can ChatGPT code open-ended survey responses?
It can suggest themes and summarize the answers it reads. For a full survey it does not code each answer against a fixed codebook, so its counts are estimates. This tool builds or takes a codebook and codes every answer against it.
Can I give my own codes?
Yes, one per line, optionally with a description after an equals sign: "Price = mentions cost or value".
Can an answer have more than one code?
Yes, one to three, most important first, separated by semicolons.
Does it measure agreement between coders?
No. It is one consistent coding pass; review a sample yourself and describe the AI-assisted step in your method.
How many answers fit in one payment?
Up to 20,000 answers to one question for $5. Unlimited covers up to 100,000 answers.
Is it free?
Files of up to 100 answers are free in full; for bigger files the codebook and the first answers are free on screen.