By Apercu team ·
How to Code Open-Ended Survey Responses (with a Free Codebook Template)
How to code open-ended survey responses: build a codebook, apply codes consistently, group codes into nets and check reliability. Free Excel template.
Quick answer: coding open-ended survey responses means tagging each answer with one or more short labels (codes) from an agreed list (the codebook), then counting how many respondents received each code. Build the codebook from a sample of answers, define every code with include and exclude rules, code all responses, group related codes into nets, and check that two coders would agree.
Free downloads:
- Open-ended survey coding template (Excel): a codebook sheet, a coding grid and a summary sheet with the counting formulas already set up
- Codebook template (CSV): import it into Excel, Google Sheets or your survey tool
This guide goes deeper on the coding step of our full guide to analyzing open-ended survey responses.
Key terms
| Term | Meaning |
|---|---|
| Code | A short label for one idea, such as "Delivery speed" |
| Codebook (or code frame) | The full list of codes with definitions, rules and examples |
| Net | A group of related codes reported together, such as "Delivery" for "Delivery speed" and "Shipping cost" |
| Multi-coding | Giving one response more than one code |
| Base | The number of respondents who gave a valid answer, used for percentages |
| Inter-coder reliability | How often two coders apply the same codes to the same responses |
Step 1: Decide what the codes are for
Write down the decision the survey supports before you code anything. "Which three problems should we fix first?" leads to codes about problems. "Why do customers choose us?" leads to codes about reasons. Codes that do not support the decision are noise.
Step 2: Draft the codebook from a sample
Read a random sample of 50 to 100 responses and note every distinct idea. Then decide your approach:
- Inductive coding builds codes from what people actually wrote. Use it for exploratory questions.
- Deductive coding starts from a predefined list, such as last year's codebook. Use it when you need to compare survey waves.
Most projects combine both. See inductive vs deductive coding for a deeper comparison.
Step 3: Write a definition, rules and an example for every code
A code name alone is not enough. Two people will read "Support" differently. Each code needs:
| Code | Definition | Include when | Exclude when | Example |
|---|---|---|---|---|
| Delivery speed | Orders arriving later than expected | Slow, late or delayed delivery | Only delivery cost is mentioned | "Took almost three weeks to arrive" |
| Shipping cost | Delivery fees seen as too high | Shipping price, fees, minimum order | Only speed is mentioned | "Shipping costs more than the item" |
| Customer support | Slow, unhelpful or hard-to-reach support | Replies, agents, refund handling | General praise with no support mention | "Emailed twice, no answer" |
The template includes this codebook with seven example codes you can replace.
Step 4: Set your coding rules
Agree on these before coding starts:
- Multi-coding: allow several codes per response. "Late delivery and the box was crushed" gets both Delivery speed and Product quality.
- Non-answers: code "n/a", "nothing" and blanks separately so they do not inflate other codes, and remove them from the base if you report only on valid answers.
- Other: use it for relevant ideas mentioned by fewer than about 2 percent of respondents. If Other passes 10 percent, add codes.
- Sentiment within a code: decide whether "Price" covers both "too expensive" and "great value". If it does, the code tells you the topic but not the direction. Separate codes such as "Price too high" and "Good value" are usually clearer.
Step 5: Code every response
Work through the responses in order and apply every code that fits. In the Excel template, each code is a column, and you type 1 under each code that applies. The summary sheet counts mentions and calculates the percent of respondents automatically.
Practical tips:
- Code in focused blocks of 30 to 45 minutes. Consistency drops when you are tired.
- Keep a "new code ideas" list instead of changing the codebook mid-block.
- When you add a code, go back and recheck earlier responses for it.
Step 6: Group codes into nets
Nets make long codebooks readable. A report with 25 codes is hard to act on. A report with 5 nets, each broken down into its codes, tells a clear story.
| Net | Codes |
|---|---|
| Delivery | Delivery speed, Shipping cost, Tracking |
| Website experience | Checkout problems, Search, Account login |
| Product | Product quality, Sizing, Range |
| Service | Customer support, Returns and refunds |
A respondent counts once in a net even if they mention two of its codes. Otherwise the net percentage is inflated.
Step 7: Check reliability
If more than one person codes, have each coder independently code the same 30 to 50 responses and compare:
- Percent agreement: the share of code decisions where both coders agree. Above 80 percent is a common working target for applied research.
- Cohen's kappa: adjusts agreement for chance. Academic projects often report it.
Where coders disagree, the definition is usually the problem, not the coder. Tighten the include and exclude rules and test again.
Step 8: Count and report
Report the percent of respondents for each code and net, always with the base: "Delivery was mentioned by 34 percent of respondents (n = 412)." Percent of respondents can add up to more than 100 percent because of multi-coding, and that is expected.
For how to turn the numbers into slides, see how to present open-ended survey results.
How long does manual coding take?
Expect roughly 4 to 8 hours for 300 to 500 responses, including building the codebook. Large surveys with thousands of responses can take days, which is where automated tools help.
Faster alternative: let AI draft the code frame
An AI thematic analysis tool such as Apercu groups similar responses and names each group, which gives you a draft code frame with counts in minutes. Each response goes into the single theme it fits best, so use the output as a first pass if your project needs multi-coding. You can try it without an account using the free thematic analysis tool.
FAQ
What is the difference between a code and a theme?
A code labels one specific idea in a response, such as "Delivery speed". A theme is a broader pattern that groups several codes, such as "Customers do not trust our delivery promises". In survey research, the reporting groups of codes are often called nets.
How many codes should a codebook have?
Usually 8 to 20 codes for one question. Fewer makes the results too broad to act on, and more makes coding slow and inconsistent. Group codes into nets if you need more detail.
Should one response get more than one code?
Yes, in most survey projects. Many answers mention several topics, and forcing one code hides information. Report percent of respondents per code, which can add up to more than 100 percent.
How do I code open-ended responses in Excel?
Make one column per code, type 1 under each code that applies to a response, and count with SUM and COUNTIFS. Our Excel template has this set up, and the Excel guide walks through it.
What is a good inter-coder agreement score?
For applied business research, percent agreement above about 80 percent is a common working target. Academic studies often report Cohen's kappa and follow their field's thresholds.