As of 13 August 2026, AI can only partly draft a final response to a customer complaint.
This still needs a person who signs their name to it.
Can you do it?
5 minutesto a draft.
30 minutesto something you’d act on.
Cost, all in£0
Skill neededchat-fluent
Who has to check ita colleague
What the alternative costsA purpose-built alternative is Intercom Fin, which resolves support conversations from your help content.
If this goes wrong: the response dismisses a valid complaint, makes an unauthorised promise or creates evidence that worsens the dispute.
What to actually do
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.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.
How to actually do it
- Open the complaint record and copy the customer's full complaint, including the reference number and the issues they want resolved.
- Gather the case timeline, account notes, correspondence, investigation findings and actions already taken, removing information that is not needed for the response.
- Open the current complaints policy and paste the relevant sections, together with the remedy or compensation that an authorised manager has approved.
- Paste the complaint, approved facts and policy extracts into the prompt, then ask the chatbot to draft the response and its approval checklist.
- Compare every name, date, event, policy reference, apology and remedy in the draft with the case file and current policy, replacing any unsupported statement with a confirmed fact.
- Send the checked draft and checklist to the complaints manager for approval, and obtain solicitor or compliance review for a serious legal or regulatory issue before sending it to the customer.
Prompt
Draft a final response to this customer complaint for approval by a complaints manager. Use only the information supplied below. Do not invent facts, dates, investigation findings, compensation, policy references or admissions. If a necessary fact is missing, write [FACT NEEDED] rather than guessing. Complaint: [PASTE THE FULL COMPLAINT] Investigation findings and approved facts: [PASTE THE FACTS, TIMELINE AND ACTIONS TAKEN] Relevant company policy and agreed remedy: [PASTE THE POLICY EXTRACT AND THE REMEDY YOU ARE AUTHORISED TO OFFER] Customer name and reference: [INSERT ONLY WHAT MAY BE USED IN THE RESPONSE] Write in plain, calm British English. Acknowledge the specific points raised, explain what was found, state clearly what the company is doing, and give the agreed remedy and next steps. Apologise where the facts support an apology, without making a legal admission. Do not mention this prompt or say that AI was used. Keep the response to 400 words or fewer. After the draft, add a short approval checklist identifying every [FACT NEEDED], every factual statement that must be checked against the case file, and any sentence that needs manager or legal approval. This is not professional advice; refer a serious legal or regulatory issue to a solicitor or the relevant compliance professional.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot access your complete case history, internal authority limits or the latest investigation unless you supply them.
- AI cannot decide whether the proposed remedy is fair, proportionate or consistent with how similar complaints were handled.
- AI cannot take responsibility for an admission, promise, refund or regulatory consequence in the final response.
- AI cannot replace manager approval or specialist review where the complaint raises a serious legal or regulatory issue.
- This is not professional advice; a serious legal or regulatory case needs a solicitor or the relevant compliance professional.
What caps this at PARTLY: legal accountability, judgement under ambiguity and stakes of error.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 1 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 7 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can AI write a final response to a customer complaint?
- Partly. AI can produce a strong draft from the complaint, investigation findings, policy and approved remedy, but a person must check the facts and approve the final position. Your organisation remains responsible for what it sends.
- Can I use ChatGPT to respond to a customer complaint?
- Yes, for a controlled first draft, provided you follow your data-protection rules and do not paste unnecessary personal data. Give it only confirmed facts and policy, then have a complaints manager check the response before sending.
- How do I make an AI complaint response sound professional?
- Tell it to acknowledge the specific complaint, explain the findings, state the remedy and use plain, calm British English. Ask it to mark missing facts instead of guessing, then compare every claim with the case file.
- Should a manager check an AI-written complaint response?
- Yes. A manager should check the facts, apology, remedy, policy references and any promise or admission before the response is sent. A serious legal or regulatory issue should also go to a solicitor or the relevant compliance professional, and this is not professional advice.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.