how to use AI to analyze survey responses (without a data team)

by

how to use AI to analyze survey responses (without a data team)

open-ended survey responses are the most valuable and the most annoying data to analyze.

the value: people explain their reasoning, use their own language, and reveal things that rating scales miss.

the annoying part: 200 text responses cannot be pivot-tabled. manually reading and coding them takes hours. and “coding” — grouping responses into themes and counting frequency — requires making judgment calls that are hard to make consistently across 200 entries.

AI tools handle this now. this guide shows the workflow.

the survey analysis problem: too much text, not enough time

before AI, the standard approach was:
1. export responses to Excel
2. manually read through each
3. create a coding scheme (a list of themes)
4. tag each response with one or more themes
5. count the frequency of each theme
6. write a summary

for a 30-response sample, this takes 2-3 hours. for 200 responses, a full day.

with AI, steps 2-5 happen in under 10 minutes. you still review the output (step 1 in the AI workflow) and write the summary (still your job), but the mechanical part is automated.

step 1: export your survey data for AI analysis

from your survey tool, export responses as CSV or Excel.

the columns you need:
– a respondent identifier (row number, response ID, or timestamp — does not need to be a name)
– each open-ended question in its own column

for the AI analysis, you will work one question at a time.

prepare the text for analysis

copy the responses for one open-ended question. if you have 200 responses, paste all 200 into a text block — one response per line or clearly delimited.

if the responses are very long (multi-paragraph), trim to the first 2-3 sentences per response. AI analysis works best on short, direct input.

step 2: prompting Claude or ChatGPT to categorize responses

open Claude or ChatGPT. paste the responses into the chat with one of these prompts:

for emergent themes (you do not know the categories in advance):

I have 200 open-ended survey responses to the question: 
"What is the biggest challenge you face with [topic]?"

Here are the responses:
[paste responses here]

Please:
1. Identify the 5-7 most common themes across all responses
2. Name each theme clearly
3. Count how many responses fit each theme (a response can fit more than one)
4. Show 2-3 example quotes for each theme
5. Note anything surprising or that does not fit a theme

for predefined categories (you already have hypotheses):

I have 150 open-ended survey responses to the question:
"Why did you choose our product over alternatives?"

Please categorize each response into one of these categories:
- Price
- Ease of use
- Features
- Recommendation from someone
- Other (please specify)

Show me:
1. Count of responses per category
2. Percentage breakdown
3. 2-3 example quotes per category
4. Any responses that were hard to classify

for sentiment analysis:

I have 100 customer feedback responses. 
Please:
1. Classify each as Positive, Negative, or Mixed
2. Show the count and percentage for each sentiment
3. Identify the most common positive themes
4. Identify the most common negative themes or complaints
5. Flag any responses that suggest high urgency issues

Responses:
[paste here]

step 3: turning AI categories into quantified insights

the AI output gives you theme labels and example quotes. now you need counts — how many responses fit each theme.

option A: AI counts for you (reliable for smaller samples)

the prompts above ask the AI to count. for samples under 100, AI counting is generally accurate. spot-check by searching for a few responses manually.

option B: use the AI categories to tag in Excel

for larger samples or higher-stakes analysis, use the AI to define the categories but do the counting yourself:

  1. add a “theme” column to your spreadsheet
  2. go through each response and apply the AI-defined theme labels
  3. build a pivot table on the theme column for accurate counts

this is faster than starting from scratch because the hard work (defining what the themes are) is done.

option C: ask AI to tag each response individually

for samples under 50, you can paste the full list and ask the AI to tag each one:

For each of the following responses, assign a theme from this list:
[Theme A, Theme B, Theme C, Other]

1. "response text here"
2. "response text here"
...

the AI returns a numbered list with theme assignments. copy the themes into your spreadsheet column.

tools that automate the whole workflow

for teams doing regular qualitative research, three tools automate the entire process:

Dovetail: research repository and analysis platform. upload transcripts and survey exports. Dovetail automatically identifies themes, creates highlights, and builds a searchable repository across all your research.

best for: product teams, UX researchers, customer success teams that regularly analyze qualitative data.

pricing: starts at ~$30/user/month.

Qualtrics iQ: if your organization already uses Qualtrics, the iQ analytics layer applies text analytics, sentiment analysis, and topic modeling automatically.

pricing: enterprise only.

ChatGPT with custom GPTs: build a custom GPT trained on your coding scheme for consistent tagging across surveys. suitable for teams that run the same survey repeatedly and want consistent analysis.

accuracy and verification

AI theme analysis is good but not perfect. known failure modes:

over-broad categories: AI sometimes creates a theme so broad it catches 60% of responses. if one theme covers more than 40% of responses, ask the AI to break it down further.

missed nuance: sarcasm, double negatives, and irony trip up language models. read through the AI output and check any responses you suspect were misclassified.

hallucinated counts: for large samples (500+ responses), AI counting can drift. for high-stakes analysis, use the AI for theme definitions and count manually or in Excel.

verification step: after the AI analysis, read 20-30 responses yourself. do the themes match what you are seeing? if not, refine the theme names or add new categories.

the fast end-to-end workflow

  1. export survey responses as CSV
  2. copy open-ended responses for one question
  3. paste into Claude or ChatGPT with the theme-identification prompt above
  4. review the themes: do they make sense? are any too broad?
  5. ask for counts if not already provided
  6. spot-check 10% of responses manually
  7. write your summary based on the verified output

total time for 200 responses: 20-30 minutes versus 4-6 hours manually.

for the tools to collect the survey data in the first place: best survey tools for market research 2026.

for the broader market research workflow: how to do market research without a budget.