Ask ChatGPT, with no context at all, to "write a follow-up email to a prospect after our sales demo". Here is roughly what comes back:
Subject: Don't Miss Out on This AMAZING Opportunity!!!
I just wanted to REACH OUT and CIRCLE BACK to see if you had a chance to think about our GAME-CHANGING solution!
This LIMITED TIME offer won't last forever!
Plus a row of emojis, a list of benefits that apply to every product ever sold, and a "limited time offer" nobody on your team ever made.
No prospect replies to that. And yet this is how a lot of sales teams use AI: a rep opens a blank chat after a demo, types one line, cleans up the result for ten minutes, and wonders why AI feels underwhelming.
The problem is not the model. The prompt knows nothing about your product, your prices or how your team writes. A Custom GPT fixes that once, for everyone. Below is how to build one, using a made-up 22-person company, Northwind Scheduling Software, which sells scheduling software to cleaning companies, salons, lawn crews and repair shops.
Follow-ups are where small B2B deals quietly die. A rep runs four or five demos a day. Each follow-up should mention what the prospect actually said, the two or three features that fix it, and one clear next step. Done by hand, that is 10 to 15 minutes per email. Done with a bare prompt, it is fast but generic, so the rep rewrites it anyway.
A Custom GPT with your rules and playbook gets the first draft close enough that the rep only fills in the specifics. And it does not invent a discount, which is the mistake that actually costs money.
Do this on purpose. Open ChatGPT, click the message box, paste the bare prompt and click Send. Read the response. It is useful as a "before" picture for your team: this is what zero context gets you.
Click GPTs in the sidebar (in the current ChatGPT it may say Explore, or go straight to chatgpt.com/gpts). Click + Create.
The builder opens on the Create tab, which is a chat. Skip it. Click the Configure tab, where you control everything directly.
In the Name field: Sales Follow-Up Writer. Description: "Writes short follow-up emails from your sales playbook." Name it for the task, not the company, so reps know when to open it.
This is the part that matters. Click the Instructions field and write rules in four groups. Here is a shortened version of what Northwind used:
Voice and tone
Structure
Content rules
Company context
Two rules in there do most of the work. "Never invent a price or a discount" stops the most expensive mistake. "Use a [bracket] if a detail is missing" stops the GPT from making up a first name or a team size, and makes the gaps obvious to the rep.
Instructions set the rules. Knowledge gives it the facts. Under Knowledge, click Upload files and attach your sales playbook. Northwind's is a short PDF with:
That last item is underrated. One real email that worked teaches the tone better than a paragraph of adjectives.
Keep the file short and current. If your pricing changes and the PDF does not, the GPT will quote the old price with full confidence.
Click the message box in the preview panel on the right and paste the same bare prompt from Step 1, word for word. Click Send.
Northwind's result:
Subject: Next step after today's demo
Hi [First name], Thanks for your time today. Here is a quick recap of what fits your team:
- Online booking, so customers can pick a time from your website
- Text reminders sent before each appointment
- One shared calendar for the office and the crew Next step: a 14-day free trial with your real schedule loaded in. No setup fee, and you can cancel anytime. I can set it up with you on a 20-minute call.
Does Thursday at 10 or Friday at 2 work for you?
Same prompt. The only difference is context. No emojis, no invented offer, one ask with two specific times, and a bracket where the name goes.
The bare prompt is the worst case. In practice, a rep pastes two or three lines from their call notes: "Maria, 3 crews, asked about reminders, wants to stop double-booking, asked about price." The GPT now has something specific to recap, and the price comes from the playbook, not memory.
Then the rep reads it before sending. Every time. The GPT does not know what was said on the call unless someone tells it, and it can still pick the wrong feature to highlight.
The same build works for any team that writes the same kind of message all day: a property manager's tenant replies, a recruiter's candidate updates, an accounting firm's document requests. The pattern is always rules plus one reference file plus a test with the prompt people actually type.
Practice it step by step, free: https://learn.ynteractive.com/content/custom-gpt-sales-follow-up-email-writer