As of 13 August 2026, AI can only partly combine customer feedback from multiple sources.

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 relevant survey responses, NPS results, support tickets, reviews and call transcripts from each system, retaining source name, date range, rating fields and a non-identifying record ID.
   2. Remove names, email addresses, phone numbers and other unnecessary personal data, then save each export under a clear source label and note any missing fields or different date ranges.
   3. Open a chatbot and paste the prompt, followed by each source in a separate labelled section; include the rating scale and explain whether any records are repeated across sources.
   4. Ask the model to produce the combined report, then ask a second question requesting a list of every record used in each theme and every suspected duplicate.
   5. Compare the report's record counts, quotations, duplicate list and numerical calculations against the original exports, correcting any unsupported theme or calculation in the report.
   6. Send the checked report to a colleague who understands the customers and service process, and ask them to confirm the theme names, priority ranking and proposed follow-up actions before anyone changes the service.

   Prompt

   Combine the customer feedback below into one evidence-based report. Treat each source separately first, then produce a combined view.
   
   Sources:
   [PASTE EACH SOURCE UNDER A LABEL, INCLUDING SOURCE NAME, DATE RANGE, CHANNEL, AND ANY CUSTOMER OR RESPONSE ID]
   
   Tasks:
   1. Standardise obvious formatting differences without changing the wording or meaning of any comment.
   2. Identify likely duplicate records and list them separately rather than silently removing them.
   3. Group comments into themes and sub-themes. Give each theme a neutral name, a short explanation, the number of records supporting it, and representative quotations with their source and record ID.
   4. Separate praise, complaints, requests, questions and neutral observations.
   5. Identify differences between sources, including themes that appear in only one source.
   6. If numerical ratings or NPS scores are supplied, calculate only from the supplied values, show the calculation and state the denominator. Do not invent missing scores or treat text sentiment as an NPS score.
   7. Flag ambiguous comments, conflicting evidence, missing fields, possible sampling bias and any conclusion that cannot be supported by the supplied data.
   8. Rank the findings by evidence strength and likely customer-service importance, but do not claim to know financial impact or customer intent unless the data supports it.
   9. End with a table of recommended follow-up actions. For each action, include the evidence, the owner to assign, the data needed to confirm it and a proposed check before making a service change.
   
   Rules:
   - Use only the supplied data.
   - Preserve source labels and record IDs so every finding can be checked.
   - Do not expose names, email addresses, phone numbers or other unnecessary personal data; replace them with [REDACTED].
   - Do not present correlation as causation.
   - If the data is insufficient, say exactly what is missing.
   - Keep the report concise but show enough evidence for a colleague to audit it.

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