Using AI · Growth Library
Let AI Draft Routine Replies, Keep the Human on the Conversation
Use suggestions to reduce repetitive wording without delegating the relationship
Try this
Provide only information you are allowed to share, plus the customer question and approved policy or knowledge source. Ask for a concise draft. Check that it answers the real issue, does not invent policy or status, and gives the correct next action before sending.
Check-in: Did the suggestion save wording effort while leaving the human more attentive to the customer's actual problem?
Evidence: Reviewed · source · Brynjolfsson, Li & Raymond (2025) — Generative AI at Work
Review note: Reviewed against Brynjolfsson, Li & Raymond (QJE, 2025). Large-scale field evidence supports productivity gains for a specific human-in-the-loop customer-support assistant. The protocol preserves human diagnosis, policy checking, tone and accountability.
A support reply has two jobs: communicate clearly and move the real case forward. AI can help with the first. It only helps with the second when the underlying facts and policy are right.
1. Ground the draft
Give the model the verified case facts and the policy or knowledge material it is allowed to use. Avoid asking it to infer account state or policy from thin context.
2. Draft the response
Ask for a concise answer that states what is known, what happens next, and any action required from the customer. Keep routine wording routine.
3. Check the diagnosis before the prose
A beautiful response to the wrong problem is still the wrong response. Confirm that the draft addresses the actual issue and that any troubleshooting step is appropriate.
4. Edit as the accountable human
Adjust tone, remove invented certainty, and verify commitments before sending. Escalate cases that exceed the known policy or evidence.
Questions to consider
Does this mean AI should answer customers autonomously?
No. The cited field study examined an assistant that suggested responses to human agents. The agents remained responsible for the conversation and could edit or ignore suggestions.
What should always be checked?
Case facts, account or order status, policy, promised actions, deadlines, pricing, safety-sensitive information, and anything that could create a commitment or escalation.
Sources
Evidence state: practical. Sources are provided so the underlying material can be inspected directly.
Maintenance record
Review history
Current status: Reviewed. An editorial or evidence review is recorded and no later event changes that conclusion.
2026-09-15 · reviewed · Brali editorial agent Source: review-registry-using-ai-starters.json
Translate strong workplace evidence into a bounded human-in-the-loop reply workflow without extending the causal claim beyond the studied customer-support setting.
Browse the review ledger · Challenge or update this hack
Implementation observations can trigger a review, but they cannot change an evidence conclusion by themselves. Commercial relationships do not control review status or outcomes.
Continue from here
Choose the next useful path.
Article versions
Brali keeps substantive article history visible. Later reviews may refresh wording, sources, examples, or boundaries without silently replacing the record.
- Version 2026-09-15: first recorded publication in the migrated Brali corpus.
- Evidence review 2026-09-15: reviewed by Brali editorial agent; evidence state is
reviewed.
Canonical source record · Evidence state: reviewed.