Search for referral prompts and you land in two places: template packs for designing a referral programme, which is a marketing job with a budget and a tracking link, and single-line prompts that write a referral email. Neither helps the actual case, which is one happy customer, one introduction worth having, and goodwill you can spend exactly once.
These prompts assume the relationship is already good and the only question is whether you ask well. Expansion on that same happy account belongs to ChatGPT prompts for upsell and cross-sell, and the wider post-sale motions sit in AI prompts for customer success managers. This page owns the ask itself, and the uncomfortable fact that most reps never make it.
The moment to ask is a moment, not a stage in your sequence
Referral asks get scheduled: ninety days after close, or whenever the quarter looks thin. That is why they land flat. The real moment is specific and it passes.
Here are my notes, support history and recent correspondence for this account. Identify the moments in the last ninety days when this customer expressed satisfaction unprompted, and quote the evidence for each. Then tell me whether any of those moments is still live enough to reference this week, or whether it has passed.
A referral asked three days after somebody praised a result reads as a continuation. The same ask in month four reads as a favour request from a stranger.
Prompts that find who can actually refer you, and to whom
Satisfaction is not the same as usefulness. The customer who loves you may know nobody you want, and the quieter one may sit in the association that runs your market.
Here is my customer list with role, company, tenure and how our relationship stands. For each one, assess two things separately: how likely they are to say yes, and whether the people they could plausibly introduce me to match my ICP. Rank by the combination, not by satisfaction, and name anyone with no useful network even if they are my happiest customer.
That second column is the one models skip unless you force it. Account-side craft sits in Account Management & Customer Growth, and list-building is covered in how to build a lead list with AI.
Asking: easy to say yes to, easy to decline
A bad ask is open-ended and slightly needy. A good one is narrow, names the person, and gives an exit that costs nothing.
Draft the referral ask for this customer. Name the one specific person or type of person I want to meet, say in one sentence why that introduction makes sense from their side, and give them a clear way to decline that costs them nothing. Under 120 words. Do not thank them twice and do not use the word "favour".
Now read it as the customer and tell me which sentence makes them hesitate.
Models write warm and over-grateful copy by default, and over-gratitude is what signals you expect this to be hard. Messaging craft is in Outreach & Messaging.
Write the introduction for them, because they will not
This is the whole difference between a referral you get and one you are promised. A customer who has to compose the intro will not send it. One who has to forward a paragraph usually does, today.
Write the forwardable introduction in my customer's voice, not mine: three sentences they could send to their contact with no editing, explaining who I am and why this is worth twenty minutes. Keep their register rather than mine, and leave a bracket where they would add one personal line.
Take one prompt free, no account needed. Pro is $12 per month or $89 per year, and lifetime access is $149 — See pricing.
The bracket matters. It makes the note theirs rather than a press release you asked them to sign.
When the referral is to somebody wrong for you
It happens often, and the reflex is to take the meeting anyway. That costs an hour and, worse, tells your customer this is what you wanted.
My customer has offered an introduction to this person at this company. Assess the fit against my ICP honestly and say whether this is worth a meeting, worth a short call, or worth a polite redirect. If it is a redirect, draft the reply that declines the specific introduction while asking for a better-matched one, without implying their suggestion was poor.
Qualification discipline carries over here; the inbound version is in ChatGPT prompts for lead qualification. Pipeline creation generally sits in Prospecting & Pipeline Creation.
The thank-you that makes a second referral possible
Most reps thank the customer and then go silent, which quietly ends the arrangement. The close of the loop is what earns the next one.
The introduction my customer made has now had its first meeting. Draft a short note back to them: what happened, specifically enough to be real, what I am doing next, and nothing that asks for anything. Under 80 words.
Reporting back without asking again is the move. The structured version of these post-sale touches lives in Post-Sale & Renewal Micro-Work and in the QBR playbook.
When asking will cost you more than the referral is worth
Goodwill is finite and it is also collateral for the renewal. Occasionally the right answer is not to ask this account at all this quarter.
Here is what is currently open with this account: renewal date, outstanding issues, anything I have already asked them for in the last six months. Tell me whether a referral ask now risks something more valuable, and name what that is. If the answer is to wait, say what would have to change first.
An account mid-escalation, or one you asked to be a reference six weeks ago, is a no for now rather than a no.
Why a model will tell you every happy customer is ready to refer you
Ask a model which of your customers are ready to refer you and it will say most of them, because the only evidence it has is evidence of satisfaction. Satisfaction is not willingness, and willingness is not a useful network.
It cannot know that your contact is job-hunting and will not spend political capital, that their company has a policy against vendor references, that the person they would introduce you to left last month, or that they already referred you once and nothing came of it.
Being specific is what makes an ask credible, and the types of specificity that earn trust apply to the people who already know you. Use the model for assembling evidence nobody collected and for writing the note in somebody else's voice. Keep the decision about whose goodwill to spend where it belongs.
The rest of the catalogue, by role and by stage, is mapped in AI prompts for sales. Browse the library → or read the full taxonomy.
Frequently asked questions
What are good ChatGPT prompts for asking for a referral?
Four do the work: an evidence prompt that finds the moments this customer expressed satisfaction unprompted and whether they are still live, a ranking prompt that scores willingness and network quality separately, a narrow ask that names one person and gives a free exit, and a forwardable introduction written in the customer's voice.
When should I ask a customer for a referral?
Within days of a specific moment when they said something good unprompted, not at a scheduled point in a sequence. Pull those moments out of your notes and support history with their evidence, then check whether any is still recent enough to reference this week.
How do I ask for a referral without being awkward?
Make it narrow and easy to decline. Name the one person or type of person, give one sentence on why the introduction makes sense from their side, and offer an exit that costs them nothing. Then read the draft as the customer and cut the sentence that makes them hesitate.
Should I write the introduction email for my customer?
Yes, and it is the step that decides the whole thing. A customer who has to compose the note will not send it; one who has to forward three sentences usually does. Write it in their register, not yours, and leave a bracket for one personal line so it stays theirs.
What do I do with a referral that is not a good fit?
Assess it against your ICP before you accept the meeting, then decide between a meeting, a short call and a polite redirect. If you redirect, decline the specific introduction while asking for a better-matched one, without implying the suggestion was poor.