Sales Follow-Up Emails: How AI Writes Them From Your CRM

"Just checking in" is the most ignored email in sales.

You send it because you can't remember what you talked about. The prospect can tell.

Here's how I fixed that. Every follow-up email I send now is written by AI, from the call notes, the proposal and every email I've traded with that prospect. Then I approve it and it goes out.

How do you write a sales follow-up email that gets a reply? Reference what the prospect actually told you: their goal, their last question, what they're waiting on. Connect Claude or ChatGPT to your CRM and it writes that email for every open proposal, from the call notes, the proposal and your email history.

Here's the short version:

• Specific beats persistent. A follow-up that quotes the prospect's own words gets read.
• AI can only write that email if it can see the context. You need a CRM that stores calls, proposals and emails in one place.
• Connect Claude or ChatGPT to that CRM over MCP and ask: "Go through my open proposals and write a follow-up for each."
• Approve every email before it sends. The AI drafts. You decide.
• Put it on a schedule, so nothing falls through the cracks.

Why Most Sales Follow-Up Emails Get Ignored

Because they're about you, not them.

"Just checking in." "Circling back." "Any update?" Every one of those asks the prospect to do the remembering for you.

A follow-up should move the deal forward, not just touch it. That means answering a question they asked, removing a reason they're waiting, or giving them something useful.

What a Good Sales Follow-Up Email Includes

Four things, in a few short lines:

• Their words. What they said on the call: the goal, the worry, the question.
• Where things stand. What they're waiting on, and what you're waiting on.
• One useful thing. An answer, a fix, or a next step they can take today.
• One clear ask. A time to talk, or a decision date.

No recap of your services. No "hope you're well." Short enough to read on a phone.

3 Real Follow-Up Emails AI Wrote From Our Pipeline

Right now we have 12 proposals waiting on a decision. The median one is 13.5 days since we last spoke.

I asked Claude to go through all 12 and draft follow-ups from the call notes and proposals in our CRM. Here are three of them, with names and identifying details removed.

The Prospect Who Asked Hard Questions

A firm unhappy with its current paid ads vendor. On the call, the owner asked about cost per case, how long before you can judge a campaign, and what spend it takes to get started. He also said he'd be "totally game" to film 5 to 8 videos a month.

The AI pulled all of that into the email. No "just checking in" anywhere.

The Prospect Who's Waiting on Cash

A brand-new two-attorney firm. They told us the timing depends on a couple of settlements coming in over the next few months.

Pushing for a decision would kill the deal. So the AI wrote a nurture email instead: no pressure, and one free thing they can do now.

The One AI Told Me Not to Send

We'd met with this prospect two days earlier, on a call where we changed the scope.

The AI flagged it: don't send a follow-up. Send the updated scope from that call instead. That's the judgment a "day 3, day 7, day 14" sequence can't make.

Why AI Needs a Source of Truth

This is where most people get stuck with AI.

They open Claude or ChatGPT, paste in a few notes, and get a generic email back. There's no source of truth. The AI doesn't know who the prospect is, what they said or what you sent them.

You need a system that stores all of it in a database the AI can read.

What the AI Reads Before It Writes

We use Agency OS. Every lead comes in from the booking form, with the qualification answers attached. Every call is recorded and summarized on the lead.

The proposal is built on the lead too. So when the AI opens a prospect, it sees the form answers, the call recordings and notes, the proposal, and the team's notes. With Gmail connected, it reads the email thread as well.

How to Automate Sales Follow-Ups With Claude or ChatGPT

Five steps. The first one is the hard part.

Step 1: Put Every Lead in One CRM

Forms, Calendly, HubSpot: whatever brings the lead in, it has to land in one place. One pipeline, with a stage for "proposal sent."

If you're still choosing, here's our comparison of the best CRM for marketing agencies.

Step 2: Connect Your AI to the CRM

The connection runs over MCP, the Model Context Protocol. It's an open standard Anthropic created for AI apps to connect to tools and data.

Both Claude and ChatGPT can connect to a remote MCP server. The plans differ, so check yours (details in the table below). Agency OS setup: connect Claude to Agency OS.

Sources: Claude Help Center and OpenAI Help Center, checked Nov 7, 2026.

Step 3: Run the Prompt

Here's the prompt I use. Nothing fancy:

"I need you to go through my open pipeline. Find anything that has a proposal open and write a follow-up email. Write them here for me to approve first."

It finds every prospect with a proposal out, reads the calls, notes and proposal, checks the email thread, and drafts one email per prospect. It also tells you which ones not to send.

Step 4: Approve, Then Send

Read every draft. Edit the ones that are off. Then tell it which to send: "Approved. Send 1, 2, 5, 9, 10 and 12."

Never let it send without your approval. Both Claude and ChatGPT can ask for confirmation before actions like sending, and you should keep that on.

Because the AI is connected to the CRM, it can update the last-contact date when it sends. No manual logging.

Step 5: Put It on a Schedule

I have mine email me twice a week:

• Monday: every open proposal and the date of last contact.
• Friday: what closed this week, and what it recommends for next week.

That's brain space I don't spend anymore. Nothing falls through the cracks.

Claude vs. ChatGPT for Sales Follow-Ups

I ran the same prompt in both, side by side, in the video above. Both are really good.

Claude found the 12 leads in "proposal sent," checked my Gmail and wrote the drafts. ChatGPT went wider: 18 outstanding proposals, 11 in the active pipeline and 7 in nurture.

I skew toward Claude right now because I've built so much in it, so it has more context about me. I use it for the heavy agency work: proposals and deliverables. ChatGPT is my personal assistant for email.

Why a Bloated CRM Isn't the Answer

You can try to automate follow-ups inside a big sales platform. Sequences, triggers, templates.

I think that's a dead approach. A timed sequence doesn't know the prospect just told you they're at a conference until next week.

All you need is a simple platform that holds the data, and an AI connected to it. As the AI gets better, your follow-ups get better with it.

And because the CRM is shared, every rep, partner or teammate can connect their own AI to the same source of truth.

Sales Follow-Up Email Mistakes

• ❌ "Just checking in." Say something new.
• ❌ Recapping your services. They read the proposal.
• ❌ A fixed sequence for every prospect. Context beats cadence.
• ❌ Pushing a prospect who told you they're waiting. Nurture instead.
• ❌ Letting AI send without approval.
• ❌ AI with no source of truth. Generic in, generic out.
• ❌ Logging follow-ups by hand. Let the connected AI update the CRM.

Sales Follow-Up Email FAQs

What should a sales follow-up email say? Reference what the prospect told you: their goal, their question, or what they're waiting on. Add one useful thing, like an answer or a next step, and end with one clear ask.

How many times should you follow up after sending a proposal? Until you get a real yes or no. Let the context decide the timing: a prospect waiting on cash needs a nurture email, not a weekly push. The full sales process is in how to sell SEO services.

Can AI write sales follow-up emails? Yes, if it can see the context. Connect Claude or ChatGPT to a CRM that stores your call notes, proposals and emails, and it writes a specific follow-up for each open deal.

Should AI send follow-up emails automatically? No. Let it draft, then approve each email before it sends. Both Claude and ChatGPT can ask for confirmation before taking actions like sending.

Is Claude or ChatGPT better for sales follow-ups? Both work well. Claude is my pick for heavy agency workflows like proposals. ChatGPT works well as a personal email assistant. On ChatGPT, full MCP with write actions is available on Business and Enterprise/Edu plans, while Pro users can connect custom MCP servers with read and fetch permissions.

Do I need a CRM to automate follow-ups with AI? Yes. The AI needs a source of truth with every lead, call, proposal and email in one database. Without it, you get generic emails.