Skip to main content

AI for Marketers: Your Own Assistant · Part 1: Getting started

Chapter 1.1 · 1 of 32 · 5 min

The difference between ChatGPT and your own AI assistant

You may be thinking: ChatGPT already exists, so why install a separate assistant? It is a fair question, and an example makes the answer clear.

Free to read

You may be thinking: ChatGPT already exists, so why install a separate assistant? It is a fair question, and an example makes the answer clear.

Say you have a client called Sarah Lin, who came from Fiverr and runs an online flower shop. Three days ago you discussed a GA4 setup with her. You quoted 150 dollars, and she is hesitating. Today you want to send a follow-up.

To do this in ChatGPT, you first have to write who this client is, which platform she came from, what she asked for, what you quoted, what her objection was, and how you like to write. In the time that takes, you could have written the message yourself. And tomorrow you will do the same for another client. ChatGPT does not know your clients, because it has never been given a way to know them.

Your own assistant reads your CRM. You only write "Write a follow-up to Sarah Lin, she's hesitating about the price". It works out the rest from the CRM by itself, because the CRM has a row for Sarah Lin with her platform, need, budget, the price you quoted and notes from your earlier conversations.

A different place to talk

ChatGPT runs in a browser. You have to open a tab, stay logged in, and be at your computer.

You talk to your own assistant from Telegram on your phone. That means on your commute, while waiting at a client's office, in bed at night. Whatever you send on Telegram reaches the assistant running on your computer. You can also talk to it from the terminal, which is handy when you are working at your computer.

It also acts on its own

This is the real difference. ChatGPT never tells you anything on its own initiative. It answers when you ask, and that is all.

Your own assistant sends you the follow-up list every morning by itself. Without you asking anything, you get a Telegram message at 8 a.m.: today is the day to follow up with these three people, and here is what a short draft for each could look like. It creates rows in Notion. And, with your permission, it runs commands on your computer: reading files, doing calculations, that kind of work.

Note the word permission. Before doing anything risky, it will ask, and you reply yes or no. Module 7 has a separate chapter on this mechanism, and I will urge you strongly to keep it switched on.

Where the data lives

I work in privacy, so for me this section is not a matter of curiosity. It is my profession.

Whatever you type into ChatGPT is stored on OpenAI's servers, inside their account, under their rules.

With your own assistant, the memory and keys live on your computer, on your hard disk. The CRM lives in your own Notion workspace. Only what you send along with a specific question goes to the AI model. The whole CRM is not uploaded and parked anywhere.

But saying "nothing goes anywhere" would be a lie, and I will not say it. When you paste a client conversation and say "add this to the CRM", that text goes to the AI provider's servers, because the model has to read the text to understand it. On a free Gemini key, Google may use that text to improve its products. So in Module 7 I set a clear rule: on the free key you practice with made-up data, and you move to the paid route before you work with real clients' personal information.

Which program this is

This assistant runs on an open-source program called Hermes Agent. Open-source means the code is open for anyone to inspect, and you do not depend on any company's servers. The program sits on your computer and talks to the AI model using your key.

Hermes Agent by itself is an empty framework. On top of it we will install a ready-made marketer template called marketer-starter. Inside it you get a personality that talks in a mix of Bangla and English (that is how I work; write to it in English and it should answer in English), a Notion CRM skill, the draft and "sent" rules, the morning follow-up job, and command approval switched on. So you do not have to teach the assistant from zero. In Module 3 it goes in with one command.

Steps

  1. Think back to your last three client conversations. Consider where their details are scattered right now: Fiverr inbox, WhatsApp, email, notebook. Write the list down on paper, not only in your head.
  2. Decide which details would be useful to have in your CRM. In Module 4 we will use a ready-made list, but thinking about it yourself first will make that list easier to understand.
  3. Read the "Where the data lives" section of this chapter once more. The whole of Module 7 stands on that foundation.

What not to do

  • Do not think of this assistant as a replacement for ChatGPT. It is a tool for a different job. ChatGPT works well for polishing text or random questions. This assistant's strength is that it knows your clients and acts on its own.
  • Before you start the install, do not assume it will run everything by itself. You will have three jobs every day: pasting, sending yourself, and saying "sent". Module 4 covers this in detail.

Proof

You can write, in your own words and in three lines, the difference between ChatGPT and your own assistant, and what it means for the assistant to be able to read your CRM. If you cannot, read this chapter again, because every later decision rests on it.

Want the full chapter list, or to jump to another chapter?

All chapters in the book