AI Resources for Coaches » The AI Referral Request Workflow

What This Workflow Does

It turns the referral ask from something you remember to do into something that happens on its own. When an engagement wraps or a client logs a win, the workflow pulls the context you already have on that person, drafts a referral request in your voice, and puts it in front of you for approval. Replies land in one place, so you can see who said yes and who needs a follow-up.

You build it once. It runs on every client after that.

Why Automate an Email

Most referral advice tells you to ask more often. Frequency is not the problem. A vague “keep me in mind” hands your client three jobs: remember you later, notice a match, and make the introduction. Most people will not do all three, however much they liked working with you.

A specific ask does that work for them. It names the kind of person you help, references something real from your time together, and gives them something they can forward without rewriting it. Writing that by hand for every client is where the habit dies. That is the part worth handing to a machine.

Before You Start

Time: about two hours to build. After that it runs without you.

To be clear about what “runs without you” means: that is your choice, not the default. You can build in an approval gate so nothing goes out until you have read it, which is how we would suggest starting. The workflow below does exactly that. Full automation is something you switch on later, if you ever want it.

Cost: n8n has a free self-hosted tier and a paid cloud plan. Gmail, Sheets, and Forms are free. The AI step costs pennies per email.

What you will need:

• An n8n account, cloud or self-hosted
• An email account. We use Gmail and Google Workspace in this example, but the same workflow runs on Outlook, Microsoft 365, or any provider n8n connects to. Swap the email node and the rest is unchanged.
• A Google Sheet listing clients and engagement end dates
• An API key from an AI provider (OpenAI, Anthropic, or similar)
• Optional: LinkedIn. Useful for two different things. First, pulling role and company so the ask is specific. Second, sending the request through LinkedIn instead of email, or asking for a public recommendation there rather than a private referral. That is a different play from the one described here. See Advanced Ideas.

▸ What is n8n?

A workflow automation tool. You connect apps visually, as boxes joined by lines, instead of writing code. Think of it as a flowchart that actually runs. Zapier and Make do similar things. n8n is the one with a free self-hosted option and fewer limits on steps.

▸ What is a node?

One box in the workflow. Each node does a single job: fetch an email, call the AI, write a row to a spreadsheet. You chain them together in order.

▸ What is a trigger?

The node that starts everything. It can be a schedule (“check this sheet every Monday”), an event (“a new row was added”), or a manual click. Nothing runs until the trigger fires.

▸ What is an API key?

A password that lets one tool talk to another on your behalf. You paste your AI provider key into n8n once, so it can send requests for you. Treat it like a password. Never share it, and never paste it into a public document or a prompt.

Before You Connect Client Data

The Workflow at a Glance

The approval gate sits in the middle on purpose. The AI writes the draft, Slack tells you it is ready, and nothing reaches a client until you say so. Sending automatically is available, but it is the option you turn on later, not the one you start with.

How It Works, Step by Step

1. The trigger

Keep a Google Sheet with one row per client and an engagement end date. n8n checks it on a schedule and fires when a date passes. If you would rather trigger on a win, add a “recent win” column and let a change in that cell start the run.

2. Pull the context

The workflow searches your inbox for the last few threads with that client and, optionally, pulls their role and company from LinkedIn. This is what makes the ask specific instead of generic. It is also the step that touches client data, so set your limits here: pull what you need for one email, and nothing more.

3. Draft the ask

The AI node takes that context plus your tone instructions and writes one email. Prompts are below.

4. Review before it goes anywhere

n8n posts the draft to Slack with an approve button, or emails it to you. This is the gate. Keep it on for at least the first month, and honestly, most coaches keep it on permanently. It costs ten seconds.

▸ Why the gate matters

The AI has your client real context. That is what makes the draft good, and it is also what makes a bad draft embarrassing. A ten-second glance costs far less than one awkward email to a client you like.

5. Choose what happens next

Two options, and the first is the one we would recommend.

• Save it as a draft. The workflow writes the email into your Gmail or Outlook drafts folder and notifies you. You open it, adjust anything you want, and hit send yourself. The automation does the work; you keep the last word.
• Send it automatically. Once you have watched twenty of these come through and they are consistently right, you can let it send from your own address. Replies still come back to you.

6. Capture what comes back

Include a short Google Form link for anyone who wants to pass along a name. Responses flow into a Sheet, so you can see who referred whom without digging through your inbox.

The Prompts

Copy these, then replace the bracketed parts. Each produces a different kind of ask.

Prompt 1 — The direct ask (use when an engagement wraps)

You are writing a referral request on behalf of [YOUR NAME], a [YOUR NICHE] coach.

CLIENT CONTEXT:
- Name: [CLIENT NAME]
- Role and company: [ROLE, COMPANY]
- What we worked on: [ENGAGEMENT FOCUS]
- A specific result or moment worth referencing: [WIN]

WRITE AN EMAIL THAT:
- Opens by referencing the specific result above, not a generic thank you
- Names exactly who I help, in the words a prospect would use for their own
  problem, not my professional label
- Makes one clear ask: does anyone specific come to mind
- Gives them a forwardable sentence they can paste to that person
- Stays under 150 words
- Sounds like [YOUR TONE: e.g. warm and direct, no corporate filler,
  contractions, short sentences]

Do not use the words "circle back", "reach out", "touch base", or "synergy".
Do not flatter. Do not use em-dashes.

Prompt 2 — The win-triggered ask (use right after a client succeeds)

[Same CLIENT CONTEXT block as above]

The client just achieved: [SPECIFIC WIN, with date]

WRITE A SHORT EMAIL THAT:
- Congratulates them on the specific win in one sentence, genuinely, no gushing
- Connects that win to the work we did, briefly
- Asks whether anyone in their orbit is facing the thing they were facing
  six months ago
- Keeps the ask light. They just did something hard. This should feel like
  a natural next beat, not an invoice
- Stays under 120 words
- Tone: [YOUR TONE]

Prompt 3 — The soft re-open (use for past clients you have lost touch with)

[Same CLIENT CONTEXT block as above]

Last contact: [DATE / roughly how long ago]

WRITE AN EMAIL THAT:
- Acknowledges the gap without apologizing for it
- Leads with something useful to them, not a request:
  [RESOURCE, IDEA, OR OBSERVATION relevant to their role]
- Asks the referral question second, and lightly
- Gives them an easy out so silence is not awkward
- Stays under 130 words
- Tone: [YOUR TONE]

▸ Teaching the AI your voice

Paste two or three emails you have actually sent into the prompt and add: “Match the rhythm and vocabulary of these examples.” That works better than describing your tone in adjectives. The model imitates samples far more reliably than it follows instructions like “sound warm.”

Advanced Ideas

Practical extensions once the basic version is running.

• Score the timing. Add a node that checks how recently you last emailed that client. If it is under two weeks, hold the ask and requeue it. Nothing kills a referral like arriving in a crowded inbox.
• Segment by referral type. Not every client should be asked the same thing. Tag clients as likely to refer peers, likely to refer inside their company, or likely to refer publicly, and branch the prompt accordingly.
• Switch the channel to LinkedIn. For clients you mostly know through LinkedIn, route the ask there instead of email. Same prompt, different node. You can also flip the ask entirely and request a public recommendation rather than a private introduction, which builds proof other prospects can see. Different goal, same workflow.
• Close the loop automatically. When a name comes back through the form, trigger a second workflow that drafts the intro email to the new person and reminds you to thank the referrer. The thank-you is the step most coaches skip, and it is what makes someone refer twice.
• Track what actually converts. Add a column for outcome. After twenty asks you will see which prompt variation and which timing produces real conversations, and you can drop the rest.
• Run it manually first. Before automating anything, use prompt 3 by hand on ten dormant clients. If nothing comes back, the problem is not the workflow, it is the list, and you will have saved yourself two hours.