Generic AI prompts for customer success look sharp in a shared doc and fall apart the first time you paste them into a real account. The model invents ARR, invents a discount, or invents a “risk” that never appeared in a call note. The prompt was not the problem. The missing account context was.
This guide gives you 15 AI prompts for customer success managers, grouped by lifecycle: kickoff, health check, save plan, renewal brief, QBR outline, and churn review. Each prompt says what data to paste and what to forbid. At the end, you will see why one-off prompts break without account context, and how teams turn the best ones into reusable shared instructions.
If you already bring live customer context into Claude, ChatGPT, or Cursor, see bringing customer context into Claude, ChatGPT and Cursor with MCP.
Rules That Apply to Every Prompt
Before the fifteen prompts, lock four rules. They matter more than clever wording.
- Paste evidence, not vibes. Contract dates, ARR or MRR, health signals, usage trends, open tickets, recent call notes, Slack excerpts, and CRM stage. If a fact is missing, say “unknown.”
- Forbid invention. Tell the model not to invent numbers, discounts, legal language, competitive claims, or customer quotes.
- Separate facts from judgment. Ask for observed facts first, then interpretation, then recommended next steps for a human to review.
- Require sources or gaps. Every claim should point to pasted evidence or be labeled as an open question.
A usable system prompt footer for any of the fifteen:
Use only the account data I paste. Do not invent ARR, usage, discounts, quotes, or dates. Keep observed facts separate from interpretation. Flag missing information as open questions. End with next steps a human owner must review before acting.
Kickoff Prompts (1–3)
Use these in the first two weeks after close, when the handoff is still messy and the success criteria are still negotiable.
1. Kickoff brief from the sales handoff
Paste: opportunity notes, sold scope, promised outcomes, known stakeholders, contract start, and any open technical risks.
Prompt: “Turn this sales handoff into a kickoff brief. Sections: account snapshot, sold outcomes, in-scope vs out-of-scope, stakeholder map, known risks, and a 14-day checklist. Do not invent stakeholders or outcomes. Mark gaps as questions for the AE.”
Forbid: invented go-live dates, discounts, or “must-have” features that are not in the paste.
2. Success criteria draft for day-1 alignment
Paste: buyer’s stated goals, product modules sold, and any existing KPI language from the deal.
Prompt: “Draft 3–5 measurable success criteria for the first 90 days. For each: metric, baseline if known, target, owner, and check-in date. If baseline is unknown, write ‘baseline TBD’ and ask what system can supply it.”
Forbid: invented baselines and ROI percentages.
3. Stakeholder map from messy notes
Paste: CRM contacts, call attendees, Slack mentions, and email threads.
Prompt: “Build a stakeholder map with role, influence (high/medium/low), current sentiment if evidenced, and one recommended touch. Only use people named in the paste. List unknown roles we still need.”
Forbid: invented titles or fake executive sponsors.
Health Check Prompts (4–6)
Use these weekly or biweekly when you need a clear read without staring at five dashboards.
4. Weekly health narrative
Paste: current health score and drivers if available, 30-day usage trend, open tickets, last call summary, and renewal date.
Prompt: “Write a 150-word health narrative. Lead with the current state, then the top two supporting signals, then one risk, then one recommended action. Cite the pasted evidence for each claim.”
Forbid: inventing a health score or changing the risk level without evidence.
5. Adoption gap finder
Paste: licensed seats or workspaces, active users, feature adoption metrics, and segment norms if you have them.
Prompt: “Compare sold scope to actual adoption. List gaps as: unused capability, shallow usage, or concentrated power users. For each gap, propose one low-effort experiment and one deeper intervention. Do not treat missing data as zero.”
Forbid: fake benchmarks.
6. Support-to-CS translation
Paste: open and recently closed tickets with themes, severity, and ages.
Prompt: “Cluster these tickets into themes. For each theme: customer impact, whether it is a product, process, or enablement issue, and whether CS should open a save conversation. Flag any ticket older than SLA that coincides with falling usage.”
Forbid: inventing root causes not present in the ticket text.
Save Plan Prompts (7–9)
Use these when risk is real and you need a plan a manager can approve, not a motivational paragraph.
7. Risk evidence pack
Paste: timeline of usage drop, sponsor changes, ticket spikes, unpaid invoices, and any cancel language.
Prompt: “Build a risk evidence pack with timeline, primary risk hypothesis, contributing factors, and strongest counter-evidence that the account is still recoverable. Separate facts from hypotheses.”
Forbid: inventing a cancel date or competitor switch.
8. 30-day save plan
Paste: the evidence pack output, available playbooks, and internal owners.
Prompt: “Draft a 30-day save plan with weekly outcomes, owners, customer-facing actions, and exit criteria for ‘saved,’ ‘still at risk,’ or ‘likely churn.’ Do not propose discounts unless discount authority is in the paste.”
Forbid: unauthorized pricing, legal concessions, or fake executive commitments.
9. Difficult conversation outline
Paste: the risk pack, customer’s last stated concerns, and your desired outcome for the call.
Prompt: “Outline a 30-minute rescue call: opening, three evidence-backed questions, two options to propose, and close with clear next steps. Keep tone direct and respectful. Do not invent customer quotes.”
Renewal Brief and QBR Prompts (10–13)
10. Renewal brief for the owner
Paste: contract end date, ARR, expansion history, health, usage, open risks, and stakeholder map.
Prompt: “Write a one-page renewal brief: snapshot, value delivered with sources, risks with sources, commercial options already in the record, and recommended next steps. If commercial options are missing, list questions for RevOps or the AE. Do not invent pricing.”
11. Mutual action plan seed
Paste: renewal date, open blockers, and success criteria.
Prompt: “Draft a mutual action plan table with action, owner (us vs customer), due date, and dependency. Leave dates as relative weeks if exact dates are unknown.”
12. QBR outline
Paste: last QBR notes, current metrics, wins, open issues, and roadmap items already shared with the customer.
Prompt: “Create a QBR outline with slide titles and speaker notes. Sections: outcomes vs goals, adoption, support themes, risk and opportunity, and asks. Only include roadmap items present in the paste.”
13. Executive one-pager
Paste: the renewal brief and QBR outline.
Prompt: “Compress into a one-page executive summary: value, risk, ask, and decision needed. Max 200 words. No jargon. No invented ROI.”
Churn Review Prompts (14–15)
14. Lost-account post-mortem
Paste: exit survey, usage history, tickets, call notes, and billing events.
Prompt: “Propose one primary churn reason and up to three contributing reasons. For each, cite evidence. Note where the exit survey conflicts with behavior. List what early signal we should have watched.”
Forbid: blaming a single person without evidence; inventing competitor names.
15. Pattern extractor across several losses
Paste: three to ten coded post-mortems.
Prompt: “Cohort these losses by primary reason and segment. Return a table of reason, count, ARR if provided, and the earliest common signal. Recommend one play change for the top two reasons.”
Why Prompts Break Without Account Context
Even a great prompt fails when the human has to hunt for the paste pack. People skip steps. They paste last week’s note and omit the unpaid invoice. The model fills the silence with confident fiction. That is how “AI for CS” earns a bad reputation.
Context fixes the failure mode in three ways:
- Live data beats memory. Contract, usage, tickets, calls, and health should come from systems, not from what the CSM remembers on Friday afternoon.
- Shared instructions beat private prompt libraries. When every CSM keeps a slightly different version of “renewal brief,” quality drifts and onboarding never sticks.
- Sources beat vibes. Managers will not act on a save plan that cannot show where each risk claim came from.
In Quivly, Ask Quivly answers plain-language questions across customers, revenue, health scores, tickets, calls, conversations, usage, and billing, with a reminder to verify before acting. For work you repeat, Skills save reusable AI instructions for tasks such as renewal reviews, QBR prep, onboarding, and churn save plans. A Skill states the inputs, evidence to gather, output structure, and how to handle missing information, so the team runs the same brief every time. If you prefer to stay in Claude, ChatGPT, Cursor, or another MCP-capable client, the Quivly MCP server lets those tools query the same organization-scoped customer data instead of relying on a manual paste.
Turn Your Best Prompts Into Shared Skills
Do not keep the fifteen prompts as a personal Notion page forever. Promote the ones your team actually uses.
- Pick one high-frequency task: renewal brief, weekly health narrative, or save plan.
- Rewrite the prompt as instructions another teammate could run on a different customer.
- Spell out required inputs, evidence to gather, output sections, and forbidden inventions.
- Require sources and an open-questions list.
- Publish it as a shared Skill, then revise from real runs instead of theoretical prompt tweaking.
That is the difference between “AI prompts for customer success” as a blog bookmark and AI prompts as an operating system. Start with the fifteen above. Keep the forbid rules. Then move the winners into reusable shared instructions so every account gets the same standard of preparation.
Want to see this on your own accounts? Book a demo with Quivly.
