Hiring · July 21, 2026 · 7 min read

Prompt engineering for customer support: the draft is easy, the edit is the job

In support, AI writes the draft — the skill is the edit: grounding it in policy, fixing the tone, cutting the over-promise before it reaches the customer. A practical guide with a worked example and how it's assessed.

← Part of The five pillars of a hire: what great assessments actually measure

Support was one of the first functions where AI started drafting the actual work product — the reply to the customer. That makes the skill easy to misread. It isn't 'can you get the AI to write a reply'; it's 'can you catch what the AI got wrong before a frustrated customer reads it.' This is the customer-support entry in our per-role prompt engineering series, and it's one of the clearest things the AI Sandbox surfaces.

The prompt is half the job; the edit is the other half

A strong support prompt does two things: it hands the model the relevant policy so the reply is factually grounded, and it names the tone — warm, direct, owns the mistake — so the reply doesn't read like a robot. But even a well-prompted draft needs an editor. Models are eager to please, which is exactly how an over-promise ('we'll refund that right away') slips into a reply the policy doesn't support. Catching it is AI fluency in a support seat.

A worked example

A customer is angry about a charge. The weak approach asks the AI to 'write an apology' and sends whatever comes back. The strong approach gives the model the plan, the refund policy and the tone — and explicitly forbids promising anything the policy doesn't allow. Then the agent reads the draft, softens a line that sounds dismissive, and removes a refund promise the model added on its own.

## TASK
Draft a reply to the customer message below.

## CONTEXT
- Plan: Pro (monthly). Refund policy: pro-rated, within 14 days only.
- Tone: warm, direct, no corporate filler. Own the mistake.

## RULES
- Promise nothing the policy above doesn't allow
- If their case isn't covered, say so plainly and offer the next step

• Good: supplies policy + tone, verifies accuracy, adjusts to genuine empathy, cuts an over-promise the model slipped in.
• Weak: sends a generic AI apology that's wrong on policy, tonally off, or both.

Best practices that actually move the needle

• Ground it in policy. Paste the relevant policy into the prompt so the draft starts from fact, not from the model's guess about your rules.
• Specify the tone explicitly. 'Warm, direct, owns the mistake' produces a very different reply than an unguided one.
• Forbid over-promises up front — and still read for them. Eager models invent goodwill your policy can't honor.
• Always edit before sending. The draft is a starting point; the human judgment on accuracy and empathy is the deliverable.

Common failure modes

• Send-the-draft: trusting a fluent reply that's wrong on policy.
• No tone guidance: technically-correct answers that feel cold or dismissive.
• Missing the over-promise the model added — the single most expensive support mistake.

How we assess it

A multiple-choice quiz can't tell you whether someone catches an over-promise, and banning AI tests a workflow that no longer exists. You give the candidate a realistic support scenario with the tools they'd really use and watch the edit — which is what an AI Sandbox assessment does, and how AI fluency is scored as a pillar. See what a full customer support assessment covers, why this is the honest test in AI-native hiring, or watch a role-tuned assessment get composed.

In support, the AI can write a hundred replies a day. The person you want to hire is the one who reads each one and asks, 'would this land well if I were the customer?' — and fixes it when the answer is no.