As of 13 August 2026, AI can analyse multilingual customer feedback.

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. Export the feedback into a CSV with one response per row, retaining a stable row ID, the original comment, any rating or NPS score, survey question, date, channel and non-sensitive segment fields you are allowed to use.
   2. Open a free chatbot, attach the CSV, and paste the survey question, response scale, business context and any existing theme definitions below the prompt.
   3. Ask the model to produce the language breakdown and response-level table first, keeping original comments, translations and interpretations in separate columns.
   4. Ask it to produce theme counts, complaint flags and follow-up flags only from the row-level classifications, and to include row IDs beside every quoted example.
   5. Take a sample from each language and compare the original comment with the translation and assigned theme using a fluent colleague, marking corrections in a separate review sheet.
   6. Apply the agreed corrections to the analysis, compare all totals with the number of valid rows in the CSV, then send the findings and flagged comments to the customer-service owner before making changes.

   Prompt

   Analyse the attached customer-feedback file for a UK business. Treat each row as one response and use its row ID as the reference. First identify the language of each response without changing the original text. Provide an English translation in a separate field, preserving uncertainty where a phrase is ambiguous. Then assign each response: sentiment, main theme, sub-theme, whether it contains a complaint, whether it indicates a risk of churn, and whether it needs human follow-up. Use only themes supported by the data and create a short theme list before applying it. Do not infer age, gender, ethnicity, income, location, intent or other personal characteristics unless explicitly stated. Do not present sentiment or churn risk as fact when the wording is ambiguous. For each theme, give the number of responses, the percentage of valid responses where a denominator is appropriate, a plain-English summary, and three representative quotes with row IDs and original wording. Separate translation from interpretation. Identify comments that need checking by a fluent speaker, including idioms, sarcasm, code-switching, slang, poor machine translation, threats, safeguarding concerns or legal complaints. State which fields were missing and how they affect the analysis. Do not invent comments, counts, causes or customer details. End with five practical actions, each linked to evidence in the data. Use British English and show the result as: 1) data-quality notes, 2) language breakdown, 3) response-level table, 4) themes and counts, 5) complaints and follow-up flags, 6) representative quotes, 7) limitations, and 8) recommended actions. Feedback file: [attach CSV or paste the rows]. Survey question and response scale: [paste]. Business context and current theme definitions: [paste].

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