how to automate follow-up emails with AI

You know that moment when a lead fills out your form, you mean to follow up, and then, three busy days later, you're staring at the reminder like it's judging you. This post will show you how to automate follow-up emails with AI so leads hear from you faster, without you writing every message by hand. It sounds technical, but honestly, this is more like connecting a few dots than building a spaceship.

What you need to automate follow-up emails with AI

Zapier — a tool that passes info from one app to another, like a front desk person carrying a sticky note to the right room. It has a free plan, and paid plans start around the cost of a couple lunches a month.

HubSpot or Pipedrive — your customer list and deal tracker, so the AI knows who to follow up with. Both have paid plans, and HubSpot also has a free option for basic use.

OpenAI or Anthropic — the AI service that writes the draft email based on your prompt and lead details. These usually charge by usage, and for short emails it's often fractions of a cent per draft.

Gmail or Outlook — where the draft gets saved or sent. If you already use one, you're set.

Time required: About 15 to 30 minutes the first time. Less once you’ve done it once.

Skill level: If you can fill out a form and copy and paste, you can do this.

How to automate follow-up emails with AI step-by-step

1. Pick the kind of follow-up you want to automate

Open your CRM and choose one simple follow-up moment, like “new website lead came in” or “quote sent but no reply after three days,” and you should see the stage or tag you want to use.

Start with one lane, not your whole sales process. A salon might start with consultation requests. A contractor might start with estimate requests. A coach might start with discovery call leads. Keep it boring and obvious. That's usually the right choice.

Write down the trigger in plain English, and you should have one sentence that says exactly when the follow-up should happen.

For example: “When a new lead is added to HubSpot and hasn’t been contacted yet.” Or: “When a deal in Pipedrive moves to Proposal Sent and no reply comes in after 3 days.”

2. Connect your apps in Zapier

Head to Zapier and create a new automation, which Zapier calls a “Zap,” and you should see a screen asking for a trigger app.
Choose HubSpot or Pipedrive as the trigger app, and you should see a list of trigger events.
Select the trigger event that matches your note from Step 2, and you should see a setup screen asking you to connect your account.
Click to connect your CRM account, and you should see your account listed as connected.
Test the trigger with a sample lead, and you should see lead details like name, email, company, and any notes pulled into Zapier.

This is the part where people think, “Okay, now it gets weird.” It doesn’t. All you’re doing is making sure Zapier can grab the lead details so the AI has something useful to work with.

3. Tell the AI what kind of email to draft

Add a new action step in Zapier for OpenAI or Anthropic, and you should see a form where you can choose the model and enter a prompt.

You might be thinking this sounds complicated. It's not. The prompt is just instructions. Like leaving a note for an assistant who writes pretty well but needs very clear direction.

Before the blockquote, here’s a prompt you can copy. It’s written to keep the email personal, short, and usable for small business follow-ups instead of sounding like a robot in a blazer.

Write a short follow-up email for this lead.

Business type: [your business type]
Brand voice: friendly, clear, helpful, not pushy
Goal: encourage a reply or next step

Lead details:
Name: {{lead name}}
Company: {{company name}}
Service interested in: {{service}}
Original inquiry: {{message from lead}}
Previous email or note: {{last contact note}}

Rules:

Keep it under 120 words

Use simple language

Mention one specific detail from the lead's message if available

Do not sound salesy

End with one clear next step

Do not make up facts if details are missing

Write a subject line and body

Format: Subject:
Body:

You can swap in your own business type and tone. If you run a restaurant, you might say “warm and quick.” If you’re a freelancer, maybe “casual and sharp.” If details are missing, the AI should stay general instead of inventing things. That line matters.

Paste your lead fields from Zapier into the prompt, and you should see those little variable bubbles inside the text.
Choose a current model like GPT-5.4, Claude Sonnet 4.6, or Gemini 3.1 Pro if your setup supports it, and you should see the model selected in the action step.

These newer models are better than old drip campaigns because they can use actual context. That means the draft can reflect what the person asked, what page they visited, or details from a CRM note. In some setups, teams also use website content or LinkedIn profile details to make the message more tailored. It feels less like a mass email and more like you remembered the conversation. Because, in a way, the system did.

4. Save the email as a draft instead of sending it right away

Add another action step for Gmail or Outlook, and you should see options to send an email or create a draft.
Select “Create Draft,” and you should see fields for subject, body, recipient, and sender.

This is the human-in-the-loop part. Fancy phrase, simple idea: the AI writes it, you approve it. For high-ticket work — think kitchen remodels, legal services, coaching packages, big catering orders — this is the safest move.

Map the AI subject line to the draft subject field, and you should see the subject field filled with the AI output.
Map the AI body text to the draft body field, and you should see the body field filled with the AI output.
Map the lead’s email address to the recipient field, and you should see the recipient field filled in.
Test the draft step, and you should see a real draft appear in Gmail or Outlook.

That’s your first win. You’ve gone from “remember to follow up” to “open draft, read, send.” Which is a much easier task when the phone is ringing and someone just spilled coffee near your laptop.

5. Add timing so the email goes out when it should

Insert a delay step in Zapier if you want the follow-up to wait a day or two, and you should see options for hours, days, or a specific date.
Set the delay to match the kind of lead you handle, and you should see the schedule saved in the workflow.

A good starting point: one day for hot inbound leads, three days for quote follow-ups, and maybe seven days for a gentle check-in. You don’t need a perfect system on day one. You need something that actually runs.

6. Personalize the message without making it creepy

Grab one or two useful details from your CRM notes, website form, or previous email thread, and you should see those details available as fields in Zapier.
Add those details to the prompt, and you should see the AI start using them in the draft.

Some tools and teams also pull in website content or a LinkedIn profile to make the email more tailored. That can work well, especially for B2B services, but keep it tasteful. You want “I saw you asked about patio enclosures” not “I noticed your third post from last February.” Big difference.

7. Give the AI your real brand voice

Type three short lines that describe how you actually write, and you should see a cleaner brand voice instruction in your prompt.

Here’s a simple brand voice add-on you can paste into your prompt. It helps the AI sound more like your business and less like it learned manners from a corporate brochure.

Match this brand voice:

Write like a real person

Be warm and direct

Keep sentences short

Never use hype or pressure

Sound helpful, like you're following up after a real conversation

If your salon is upbeat, add “friendly and upbeat.” If your plumbing company is more no-nonsense, add “clear and practical.” Small tweaks go a long way.

Send yourself three test drafts using different sample leads, and you should see whether the voice stays consistent.

This matters because most bad AI email isn't bad because it's AI. It's bad because nobody told it how to sound. Give it a lane, and it usually stays in it.

8. Keep your follow-ups out of spam

Limit your automated volume to under 30 to 50 emails per day per sending address, and you should see a more natural sending pattern.

This one matters more than people think. Google and Yahoo are much stricter now, and generic, repetitive email patterns can trip spam filters fast. AI can write a mountain of emails. That doesn’t mean you should send them like a fire hose.

Ask the AI to vary sentence structure and openings in your prompt, and you should see drafts that don’t all sound cloned.

Here’s a short line you can add to your prompt for that:

Vary sentence length and phrasing so emails sound natural and not repetitive across leads.

Use one real sender name and a normal reply-to address, and you should see the email look like it came from an actual person.

Dedicated outreach tools like Smartlead, Reply.io, and Instantly can help with deliverability if you’re doing more outbound volume. They usually cost around $30 to $100 a month and include sending controls that basic workspace tools don’t. If you just want help drafting inside your inbox, native options like Google Workspace with Gemini or Copilot for Microsoft 365 are often enough, usually costing roughly $20 to $30 per user each month depending on plan and setup.

When something goes wrong

These are normal speed bumps. Not disasters.

The email draft sounds like a robot wrote it

This usually happens because the prompt is too vague. Try adding 3 to 5 voice rules and one short example of an email you’ve actually sent before. The AI needs your style, not just your topic.

The AI makes up details about the lead

This happens when the model tries to be helpful with missing information. Add a rule that says, “Do not make up facts if details are missing,” and save drafts for review instead of auto-sending.

The emails stop getting replies or land in spam

This often happens when too many messages sound too similar or you send too many from one address. Lower the daily volume, ask for more variation in the prompt, and keep the message short and specific.

What to do next

Your best next step is simple: automate one follow-up only, then watch it for a week. Pick the message you forget most often, set up the draft workflow, and let that be your test run. If you want to take this further, [Owen wrote a helpful walkthrough on using AI in your CRM without making it a mess](PENDING: using AI inside a CRM for small business follow-up and lead management).

FAQ

Can I automate follow-up emails with AI for free?

Sort of. You can get close with free plans from tools like HubSpot and Zapier, but most reliable setups eventually need a paid plan or usage-based AI account. The good news is the actual cost per drafted email is usually very low.

Is it better to use Gmail or a dedicated outreach tool?

If you’re following up with inbound leads or warm contacts, Gmail or Outlook drafts are usually plenty. If you’re doing larger outbound campaigns and care a lot about deliverability controls, tools like Smartlead or Reply.io make more sense.

Will AI follow-up emails hurt my deliverability?

They can if you send too many or if every message sounds the same. Keep volume under about 30 to 50 emails a day per sending address, make the wording vary, and avoid blasting generic copy. Think “steady conversation,” not “leaf blower.”

Do I need to know how to code to do this?

Nope. Good question — most people wonder this. A no-code setup with Zapier or Make can handle the whole workflow without writing code.

Should I let AI send the emails automatically or save drafts?

For low-stakes follow-ups, auto-send can be fine once you’ve tested it well. For expensive services, tricky sales conversations, or anything with pricing back-and-forth, save the draft and review it first. That extra minute is usually worth it.