A contribution from Sophie Hundertmark
Sophie Hundertmark is an expert in the practical use of artificial intelligence with a focus on chatbots, AI strategies and responsible technology integration. She is a researcher and lecturer at the Lucerne University of Applied Sciences and Arts and is currently writing her dissertation in the field of Conversational AI at the University of Fribourg. As a consultant, she supports companies, administrations and educational institutions in the introduction of effective AI solutions. More about Sophie Hundertmark on LinkedIn.
A CustomGPT was used for linguistic and stylistic creation – as well as for translation. This is based on the GPT-5 language model from OpenAI and was developed by Sophie Hundertmark herself.
As a busy consultant, lecturer and researcher with a focus on AI, I am constantly looking for ways to make my day-to-day work more efficient. Of course, I use AI assistants as often as possible to automate routines – after all, I don’t just preach this in my lectures, I also put it into practice myself.
At the same time, I really appreciate the personal exchange. I’m actually a big fan of classic letter post –simply because it always has a very special touch and you immediately notice that someone has taken the time to write it. I also want to preserve this individuality in my digital communication.
Emails are a key means of communication for me – whether it’s with customers, partners, journalists or students. Nevertheless, the high volume can quickly become overwhelming. So I looked for a way to automate the preparatory work as much as possible without my messages losing any of their personality.
The result: a Zapier workflow with AI support that does a lot of the work for me, but always gives me full control over the content and tone of voice. In this article, I’ll show you step by step how it works, which tools I use and how I don’t lose sight of the personal touch and data protection.
The three basic principles before the start
Before I even built a single line in Zapier, I defined three fixed rules:
- No fully automated shipping
I personally check every e-mail that the system creates and send it manually – except in rare special cases, which are clearly marked. - Taking data protection seriously
Zapier is a great tool, but it is not always transparent where data is stored. Therefore, no sensitive or confidential e-mails are sent via this process. - Technology must adapt to me, not the other way around
I didn’t want a complex, error-prone workflow, but a clear, stable solution that does the work for me.
Overview: How my email workflow works with Zapier
My system is based on simple logic:
I don’t answer emails immediately myself, but forward them to my AI assistant with instructions. This is an individually trained AI with a specially developed system prompt that is integrated into my workflow via Zapier.
Step-by-step instructions
Step 1: The manual trigger
- I consciously choose which e-mails are suitable.
- I forward these to a separate e-mail address that only exists for the AI workflow.
- At the top of the forwarded message, I add precise instructions for the reply, e.g.: “Please formulate a reply that responds to Ms. Meier’s keynote request. Mention that I don’t have any appointments until next week and add my link to the portfolio page.”
These brief hints give the AI direction and ensure that the answer is tailor-made.
Step 2: Zapier workflow starts
Zapier continuously monitors the special mailbox. As soon as a new message arrives, the automated process begins.
Action 1 – Generate AI response
- The AI receives the complete e-mail history plus my instructions.
- A system prompt that I developed from over 10 old e-mail histories is running in the background.
- This prompt defines:
- Tonality: friendly, engaging, clearly structured, with a personal touch
- Response patterns for different categories (keynote, workshop, general topics, etc.)
Example: When a workshop is requested, the AI automatically asks for the number of participants and target group and outlines possible content.
Action 2 – Extract data
After the answer has been created, the AI extracts two further pieces of information from the original process:
- Sender e-mail address – for correct addressing.
- Last message without instructions – so that the context is retained in the design.
Together with the generated answer, I then have three variables:
- Answer text
- Original course
- Recipient address
These are saved in the current Zap and can even be used later in other systems (e.g. CRM).
Action 3 – Create e-mail draft
Zapier uses the variables to create a new message in my drafts folder:
- Sender: always my own e-mail address
- Recipient: extracted address
- Subject: taken from the original message
- Contents:
[Generierte Antwort] --- [Originalverlauf ohne Instruktionen]
The human influence remains
I never send these drafts immediately.
Instead, I open it, read the answer carefully and adapt it:
- Make formulations more personal
- Insert additional thoughts
- Adjust dates or content
This final revision ensures that the communication remains authentic and individual.
Special case: Fully automated shipping
There are a few, clearly defined cases in which I deliberately send completely automated messages – e.g. for simple confirmations or standardized responses.
Then I’ll add a clear note:
“This message was created automatically by Sophie’s AI assistant.”
This way, recipients always know that the response was created automatically.
Data protection & security
One important point that I must emphasize is the issue of data protection. Services like Zapier are incredibly useful, but it’s not always transparent where and how data is stored and processed.
For this reason, I only use this type of automation for certain emails and not for confidential or sensitive information. My primary email account, which contains sensitive customer data, is not connected to Zapier. The separate “AI Assistant” account minimizes the risk and keeps the main communication strictly separate.
- No sensitive e-mails are sent via this workflow.
- Separate mailbox as a trigger so that my main account is not constantly scanned by Zapier.
- Clear separation between regular communication and automation.
My learnings from the implementation
- Start small – test with one category first before automating everything.
- A good system prompt is crucial – the more precise it is, the fewer corrections are necessary.
- Automation = preparatory work, not a substitute – the human eye is irreplaceable.
- Plan for data protection from the outset – saves changes later on.
Conclusion
With this system, I process e-mails much faster, save time and still maintain a high level of personal quality.
The AI does the elaborate preparatory work, I take care of the finishing touches and the personal touch.
That’s how I combine efficiency and human proximity – and that’s what makes the difference.
Get started right away
If you’re curious, you can book a call with me now and fully or partially automate your emails in just a few hours.
More information about the offers here.
And if you have any further questions, just drop me a line, preferably by WhatsApp message or email.
FAQ – Frequently asked questions about my AI assistant workflow with Zapier
1. can I also implement the workflow without Zapier?
Yes, but Zapier offers the advantage of very flexible, visual workflow automation. Alternatives would be Make (Integromat), n8n or directly programmed solutions. However, these often require more technical know-how.
2. why do you only create the e-mail as a draft and not send it automatically?
Because final control is important to me. Every e-mail should appear personal, and small adjustments or additions are almost always useful. In this way, I also avoid errors in content or inappropriate wording.
3. what role exactly does the system prompt play?
The system prompt is the “brain” of the AI assistant. It defines tonality, structure and response patterns for different types of email. The more precisely it is formulated, the less you will have to rework it later.
4. can I use the AI assistant for any type of email?
Technically yes, but I recommend using the workflow only for inquiries where data protection risks are low and where recurring patterns occur – e.g. appointments, responses to quotations or general questions.
5. how do you deal with data protection and sensitive data?
I do not run any confidential customer data through this workflow. The trigger runs on a separate e-mail account and only selected messages are forwarded. This keeps the main account protected.
6. which AI do you use for text creation?
In my case, I use ChatGPT 5 Auto via a Zapier integration. It is important that the AI can reliably handle long texts and understand context from complete email histories.
7 What do I do if the AI assistant makes a mistake?
That’s one of the reasons why I don’t automate shipping. If the content of the answer doesn’t fit or seems too generic, I adjust it manually. Over time, the system prompt is optimized so that errors become less frequent.
8. can I use multiple AI assistants for different types of email?
Yes. For example, you can set up an assistant for sales inquiries, one for support and one for internal communication. Each wizard can have its own system prompt.
9. how long does it take to set up such a workflow?
Depending on your experience with Zapier and AI prompts, it can take between 1 day and weeks. Most of the time often goes into creating and fine-tuning the system prompt. And of course it also depends on the data basis.
10. does the workflow really save time?
Absolutely. Even if I still check every e-mail, the automatic pre-formulation saves several minutes per message. Extrapolated over the course of a day, this results in significant time savings – without compromising on quality.
