As of 13 August 2026, AI can clean your customer survey data.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open the survey export in your spreadsheet and remove direct identifiers such as names, email addresses, telephone numbers and full addresses, replacing any necessary reference with a non-identifying row ID.
   2. Copy the column headings and a representative sample into a new document, then write down the allowed scale for the NPS question, valid date format and what each column means.
   3. Decide your rules for duplicate submissions, test responses, blank answers, scores outside the allowed scale and obvious spam, and add those rules to the prompt.
   4. Paste the pseudonymised CSV or table and the completed prompt into ChatGPT, Claude or Gemini, then ask it to stop and ask questions wherever a rule is ambiguous.
   5. Save the original export separately, save the cleaned table as a new file, and save the cleaning log with it.
   6. Compare the original and cleaned row counts, duplicate and missing-value counts, NPS score totals and a sample of changed rows against the cleaning log.
   7. Open every unresolved row and decide each one yourself, then record the decision in the log before using the cleaned data for NPS or customer-service analysis.

   Prompt

   Clean the customer survey data I provide below.
   
   Purpose: prepare it for feedback and NPS analysis without changing the meaning of valid responses.
   
   Data description: [describe the survey, collection period and intended use]
   Column definitions: [explain every column, including the NPS question and its allowed scale]
   Cleaning rules: [state how to handle duplicates, blank cells, invalid scores, inconsistent spelling, dates, test responses and obvious spam]
   
   Use these rules unless I explicitly replace them:
   1. Do not invent, infer or silently fill in customer answers.
   2. Do not delete a row unless it clearly matches the stated duplicate, test-response or spam rule.
   3. Preserve the original data and add a clear record of every changed, excluded or flagged row.
   4. Standardise formatting such as capitalisation, whitespace, date format and spelling only where this does not alter meaning.
   5. Keep valid NPS scores unchanged and flag scores outside the stated scale rather than guessing a correction.
   6. Treat free-text comments as customer evidence. Do not rewrite their meaning.
   7. If a rule is ambiguous, ask a question before applying it.
   8. Do not expose or repeat unnecessary personal data in your response.
   
   Return:
   A. A cleaned table in CSV or a clearly structured table.
   B. A cleaning log showing the original row identifier, action taken, reason and any new value.
   C. A list of unresolved or ambiguous rows that need my decision.
   D. Before-and-after row counts, duplicate counts, missing-value counts and invalid-value counts.
   E. A short list of checks I should run before using the data for NPS or customer-service decisions.
   
   Here is the data:
   [PASTE A PSEUDONYMISED CSV OR TABLE HERE]

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.