The renewal is 80 days out, the champion who signed it has changed jobs, and your last three emails went unanswered. Nobody has said anything is wrong. That is the problem.
Page one here is split between software vendors and thin directories, and almost all frame renewals as prediction: feed in data, receive a risk score. Prediction is the wrong frame for a rep with 40 accounts and a quarter to get through. Triage is the right one. The scoring model is covered in the customer health framework; this page owns the triage and the conversation, and the motion-by-motion set is in AI prompts for customer success managers.
Triage first: ranking a book of business by renewal risk
Before any individual account, sort the book. Export every account with renewal date, ARR, last buyer-side contact, open support tickets, and whoever signed last time.
Rank these accounts by renewal risk using only recorded facts. For each, give the two facts driving the ranking and name the one piece of information I am missing that would change it. Do not score accounts where the data is too thin — list those separately as unknown.
The separate unknown list is the point: a model asked to rank 40 accounts will rank all 40, including the eight it knows nothing about.
The evidence rule: what the account did, not how it felt
Every renewal prompt on this page carries the same constraint, because without it the output is sentiment analysis of your own notes.
For this account, separate what happened from what I concluded. List the buyer-side actions in date order — logins, tickets, meetings they booked, people they added — then list my interpretations separately. Mark any interpretation with no action behind it.
Run it on the account you feel best about. The gap is usually wider than expected.
Prompts that read usage and support history for the real story
Usage data tells you about adoption breadth, which predicts renewals better than satisfaction does.
Here is 12 months of usage by feature and by user. Tell me whether this account's usage is broadening or narrowing, how many distinct users are active now versus six months ago, and whether use is concentrated in one team. Say what each pattern implies for renewal and flag where the data is ambiguous.
Here are this account's support tickets for the year. Group them by underlying cause rather than by product area, and tell me which causes were resolved and which recurred. Then name the one theme a buyer would raise in a renewal conversation.
Narrowing usage concentrated in one team is the most reliable warning in the data: the account has one point of failure and you already know their name.
Preparing the renewal conversation itself
The conversation is not a negotiation about price until somebody makes it one. It is a review of whether the thing worked.
Build the value case for this renewal using only outcomes the customer has stated or that appear in usage data. For each claim, cite the evidence. Where I have no evidence for an outcome they expected at purchase, say so and write the question that would get it.
You are the buyer in this renewal. Here is what the account did and did not do this year. Ask me the six hardest questions, one at a time, starting with the weakest part of my case, and do not accept an answer with no evidence in it.
The second prompt is the rehearsal, and worth running out loud. Roleplay drills are in AI sales roleplay prompts, and the recurring touchpoints that should have happened before now are in Post-Sale & Renewal Micro-Work.
When the champion has left: rebuilding the case from scratch
A departed champion is the single most common reason a healthy-looking account does not renew.
My champion at this account has left. From these notes and the usage data, identify who else has shown evidence of using or benefiting from the product, what each of them appears to use it for, and which one is most likely to carry a renewal internally. Say what evidence is missing for each.
Draft a first message to a new stakeholder who inherited this relationship and was not part of the original decision. Assume they do not know what we do, have no loyalty to the purchase, and will not read past four sentences.
Then verify the map rather than trusting it — stakeholder prompts for that live in Deal Strategy & Stakeholder Management.
Separating a price objection from a value objection
These require opposite responses, and they arrive wearing the same words.
Here is what the buyer said about cost. Based on their usage, stated outcomes and who is in the conversation, assess whether this reads as a budget-process constraint, a value objection, or a negotiating position. Give the evidence for each reading and say which test would distinguish them.
A budget-process constraint is a timing problem with a procurement answer. A value objection is a renewal you may already have lost. Commercial mechanics for the first are in Negotiation, Procurement & Closing.
Expansion inside a renewal, without risking the renewal
Based on usage, which team at this account is using the product in a way that suggests an adjacent need, and what evidence supports it? Then tell me whether raising expansion before the renewal is signed puts the renewal at risk, and why.
Usually the answer is to sign first and expand in week two. Ask anyway — occasionally the expansion is why the renewal gets approved. The growth plays that follow are in Account Management & Customer Growth.
Take one prompt free, with no account, at /free-prompt. Pro is $12 per month or $89 per year, and lifetime access is $149 — See pricing.
Why a model will tell you an account is healthy when it is not
An assistant assessing your book has no contact with your customers. It has your account of your customers, written by you, mostly on days when the call went well. Ask it whether an account will renew and it reflects that account back with more confidence than you had when you wrote it.
The accounts that churn without warning are rarely the ones with angry tickets. They are the quiet ones, where usage narrowed to one team, the champion moved on, and nobody logged a complaint. A model reading those notes finds no negative signal and reports none.
Hence the three habits in every prompt here: constrain the model to buyer-side actions, require a citation rather than a judgement, and make it list the accounts it cannot assess instead of assessing them anyway. A model that names the four accounts it has no evidence about has told you where Monday goes.
The rest of the catalogue, by role and by stage, is mapped in AI prompts for sales, and the pipeline-side version of the same evidence discipline is in ChatGPT prompts for pipeline review. Browse the library → or read the full taxonomy.
Frequently asked questions
What are good ChatGPT prompts for customer renewals?
Four carry most of the weight: a triage prompt that ranks your book by recorded risk and lists separately the accounts it cannot assess, an evidence prompt that splits buyer-side actions from your interpretations, a usage prompt that says whether adoption is broadening or narrowing, and a buyer roleplay on your value case.
How do I use AI to predict churn?
Use it to triage rather than predict. Have it rank accounts on recorded facts, name the two facts behind each ranking, and flag what is missing. The signal that holds up is adoption breadth — how many distinct users are active now versus six months ago — not satisfaction scores or how the last call felt.
Can AI tell me which customers will not renew?
No. It has no contact with your customers, only your notes about them, which were written on your good days. What it does reliably is point out which accounts have no recorded buyer-side activity, which have usage narrowing into one team, and which lost the person who signed — which is the same information in a form you can act on this week.
How do I prepare for a renewal conversation with AI?
Build the value case from stated outcomes and usage data only, with a citation on every claim and an explicit list of outcomes you have no evidence for. Then have the model play the buyer and ask the six hardest questions. The answers you cannot give are your prep list.
What should I never let AI decide about an account?
Whether to discount, whether to escalate to an executive, and whether a relationship is strong enough to carry a hard conversation. Those depend on history and tone the model has no access to. Let it rank, cite and challenge; keep the commercial and relationship calls yourself.