Automating Customer Emails with GPT + Zapier
Reply to every new enquiry in under two minutes, personally, without writing a word. Here is the exact GPT and Zapier workflow — prompts, templates, and fixes included.
What Email Automation Actually Is
Automation is not a robot pretending to be you. It is a set of rules that says: when this happens, do that. A customer fills in your contact form. Within seconds, that information is passed to an AI model, a specific reply is written, the email is sent, and the contact is filed in your CRM with a follow-up reminder.
The important word is personalized. Traditional autoresponders send the same paragraph to everyone, which customers recognise instantly. Adding GPT to the middle of the workflow means every reply references what that person actually asked for, in your tone, at the speed of software.
- Trigger: something happens (a form submission, an order, a booking)
- Filter: incomplete entries and obvious spam are dropped
- Generation: GPT writes a reply grounded in those details
- Delivery: the email is sent from your own address
- Record: the contact and the message are stored in your CRM
Why It Matters for Small Businesses
Speed wins enquiries. When several businesses are quoting for the same job, the one that replies first is usually still in the conversation when the others arrive. Most small businesses reply in hours or days because the owner is on a job site. Automation closes that gap without hiring anyone.
Concrete benefits
- Every lead gets a reply, including evenings and weekends
- Response quality stops depending on how tired you are
- Nothing falls through the cracks — every enquiry lands in the CRM
- You reclaim the 30 to 60 minutes a day spent on first replies
- Follow-ups happen on schedule instead of when you remember
Measure one number before and after: median time from enquiry received to first reply sent. If that number does not drop dramatically, the automation is not configured correctly.
The Workflow: Lead Capture to CRM
The complete path, with the tool responsible for each stage.
| Stage | What happens | Tool |
|---|---|---|
| 1. Capture | Visitor submits a form with name, email, and job details | Website or HubSpot form |
| 2. Trigger | Zapier detects the new submission within seconds | Zapier |
| 3. Filter | Spam and incomplete entries are dropped | Zapier filter |
| 4. Generate | GPT writes a personalized reply from the form data | ChatGPT step |
| 5. Send | The email goes out from your business address | Gmail or Outlook |
| 6. Record | Contact created, email logged, deal opened | HubSpot |
| 7. Follow up | A task is created for day 3 and day 7 | HubSpot or Zapier delay |
Steps 1 to 6 run in under a minute with no human involved. Step 7 is where you re-enter the process, with all the context already written down for you.
Tools You Need
- A form Zapier can read — HubSpot Forms, Google Forms, Typeform, Webflow, or your website form
- A Zapier account (the free plan is enough to build and test)
- ChatGPT access — either Zapier's built-in ChatGPT app or an OpenAI API key
- An email account Zapier can send from — Gmail or Outlook
- A CRM to receive the record — HubSpot's free tier works perfectly
Use the Zapier ChatGPT integration rather than raw API keys for your first build. You can always switch to the API later when you want finer control over model and temperature.
Zapier Setup Guide, Step by Step
- 1Create a new Zap and choose your form as the trigger
Pick the app your form lives in and the event 'New Form Submission'. Connect the account and select the specific form.
- 2Test the trigger with a real submission
Fill in your own form with realistic answers. Zapier needs live sample data to map fields correctly — a blank test record causes most later failures.
- 3Add a Filter step
Continue only if the email field contains an @ and the message field is not empty. This alone removes most spam before it costs you a task.
- 4Add the ChatGPT conversation action
Paste your prompt into the message field and insert the form fields as variables. Choose a current GPT model and keep the temperature low, around 0.4, for consistency.
- 5Add the Gmail or Outlook send-email action
Set To as the form email field, write a subject line that includes their job type, and map the body to the ChatGPT output. Add your signature manually so it is always correct.
- 6Add the HubSpot create-or-update contact action
Map name, email, phone, and the enquiry text into properties. Use create-or-update so returning customers do not duplicate.
- 7Add a delay and a follow-up task
Delay three days, then create a task titled 'Follow up with [name] about [job]'. This is the step that recovers the most revenue.
- 8Turn the Zap on and watch the first ten
Check Zap History daily for the first week and read every generated email before trusting the workflow unattended.
Do not send to the customer on your very first test run. Point the email step at your own inbox until the output looks right three times in a row.
Sample GPT Prompts
New enquiry reply
You write customer emails for [Business Name], a [type of business] in [location]. Tone: warm, direct, no corporate filler. Short sentences. Write a reply to this enquiry: Name: {{name}} Service requested: {{service}} Details: {{message}} Preferred timing: {{timing}} Rules: - Under 130 words - Reference their specific request in the first sentence - Give one clear next step (reply with a preferred time slot) - Never state a price, availability, or delivery date - Do not use the words "reach out", "delighted", or "valued customer" - End with the sign-off: {{owner_name}}, {{business_name}}
Quote follow-up
Write a short follow-up email to {{name}}, who received a quote for {{service}} {{days_since}} days ago and has not replied. Tone: friendly, zero pressure, one question only. Under 80 words. Do not repeat the quote amount. Offer to answer questions or adjust the scope. Do not apologise for following up.
Post-purchase thank you
Write a thank-you email to {{name}}, who just booked {{service}} for {{date}}. Include: confirmation of what was booked, what happens next in two bullets, and how to reach us. Warm but efficient, under 110 words. No discount offers.
Example Email Templates
These are real outputs from the prompts above, lightly edited. Use them as the quality bar for your own workflow.
Hi Sarah, Thanks for getting in touch about the kitchen rewire at your flat in Didsbury — you mentioned you would like it done before the new units arrive in September, which is very doable. The next step is a quick look at the current consumer unit so the quote is accurate rather than a guess. Could you reply with two time slots that suit you next week, morning or afternoon? I will confirm availability and send a fixed price after the visit. Best, James Harper Electrical
Hi Sarah, Just checking in on the quote I sent last week. No rush at all — I know these decisions take a bit of thought. Is there anything about the scope you would like me to adjust or explain? Best, James
Hi Sarah, You are booked in for the kitchen rewire on Thursday 14 August. What happens next: - I will message the evening before to confirm arrival time - Please clear access to the consumer unit and kitchen sockets Anything comes up, just reply to this email. Best, James Harper Electrical
Best Practices
- Review every email for the first two weeks before switching to auto-send
- Never let the model invent prices, dates, or availability — pass those as variables
- Keep replies under 150 words; long automated emails read as spam
- Ban filler phrases explicitly in the prompt, not just in your head
- Send from a real person's address, never noreply@
- Log every generated email in the CRM so the next conversation has context
- Add a human escape hatch: reply and you get a person, not a bot
- Re-read your prompt monthly as your services and pricing change
Give the model three examples of emails you have actually sent and liked. Real examples improve tone far more reliably than stacking adjectives in the instructions.
Troubleshooting Guide
| Symptom | Likely cause | Fix |
|---|---|---|
| Zap does not trigger | Form not connected, or only a test submission exists | Re-test the trigger with a real submission and confirm the correct form is selected |
| Email body is empty | The GPT step output field is not mapped | Open the email step and map the body to the GPT step's response field |
| Reply sounds generic | The prompt lacks form data or brand context | Insert the actual form fields into the prompt and add three lines of brand voice guidance |
| Wrong price or promise | The model was allowed to invent facts | Pass prices as variables and forbid the prompt from stating figures not provided |
| Duplicate emails | Trigger fires on both create and update events | Add a filter so the Zap only runs when the contact is newly created |
| Zap hits task limits | Too many steps per lead | Combine steps, filter spam earlier, or upgrade the plan |
When something breaks, start in Zapier's Zap History. It shows the exact data that entered each step, which almost always reveals whether the problem is the form, the prompt, or the field mapping.
What It Costs to Run
| Component | Plan | Monthly cost |
|---|---|---|
| Zapier | Professional (multi-step, 2,000 tasks) | $19.99 |
| ChatGPT | Plus, or usage-based API | $20, or roughly $1-5 in API usage |
| HubSpot | Free CRM | $0 |
| Existing Gmail or Outlook | $0 | |
| Total | - | About $25-40/mo |
For a business where a single won job is worth a few hundred, the workflow pays for itself the first time it catches a lead that would otherwise have gone cold over a weekend.
Frequently Asked Questions
Do I need to know how to code?
No. Everything in this guide is built with Zapier's visual editor and plain-English prompts. The only technical step is connecting your accounts, and Zapier's built-in ChatGPT integration removes the need for an API key entirely.
Will customers know the email was written by AI?
Not if it is specific. The tell is generic filler like 'thank you for your interest in our services'. When the email references their exact request, timeline, and location, it reads as a fast human reply — which, after your review, it effectively is.
Should I send automatically or review first?
Review first for at least your first 50 emails. Send the draft to yourself or a Slack channel, approve it, then switch to auto-send once the quality has held up across edge cases.
What happens if the AI writes something wrong?
Constrain it. Never let the model invent prices, availability, or promises. Feed those values from your form or a lookup step, and instruct the prompt to say 'we will confirm this shortly' whenever a fact is not supplied.
Is this GDPR compliant?
It can be. Collect explicit consent on the form, state in your privacy notice that AI assists with drafting replies, keep personal data to the minimum needed, and use providers with appropriate data processing terms. A transactional reply to an enquiry is normally covered by legitimate interest; marketing sequences need consent.
How many emails can I automate on the free plans?
Zapier's free plan allows 100 tasks per month, and a multi-step workflow uses several tasks per lead. Expect roughly 20 to 30 automated leads a month before you need a paid plan.
Key Takeaways
- Automation is trigger, filter, generate, send, record, follow up
- Speed of first reply is the metric that matters — track it before and after
- GPT supplies the wording; your form and CRM supply the facts
- Review manually for the first 50 emails, then let it run
- Constrain the prompt hard: no invented prices, no filler, under 150 words
- The whole workflow runs for about $25 to $40 a month
Practical AI workflows, every Friday
Step-by-step automations you can build in an afternoon — written for small business owners, not engineers.