Customise AI to automate your processes

Victoria says: “Define an SEO process and then automate it with custom GPTs or no-code tools like Make.com.”

How do you determine the processes to focus on?

“Every person who does SEO and has done on-page optimization or an SEO audit 50, 100, or 500 times has found ways to approach this problem.

We have checklists, we have protocols, and we have templates for these tasks. Once you have the process, the checklist, and the protocol, it's very easy to give those instructions to AI in the form of custom GPTs or automations.

If you define those steps manually, it's very easy to then have AI assist you, or automate the process.”

Is there any way that you would select which process to automate first?

“I generally start with the things that I do the most, which require the most repetition.

For example, on-page SEO or SEO briefs are something that I'm doing constantly. That would be the best candidate. That was the first candidate for me, along with keyword research. We're using that all the time as well.

Those were the first two SEO processes that I automated with AI.”

How do you define those processes to make sure that they're broken down, step-by-step, for the automation?

“If you don’t want to write it down, the easiest way is to record it. Record yourself doing keyword research, or an on-page optimization/SEO audit. When you're recording, you're going to do all of the things that you do subconsciously or semi-consciously.

Most video recording technologies have an AI note taker or automated transcription. You can use any of those. If you have a transcription, you probably want to take it to an AI to say, ‘Define the process based on this,’ so you don't just have a huge block of text.

Transcribing is automatic. Then, for the step-by-step, you can ask ChatGPT, Claude, or even Fireflies.ai. Fireflies has an AI embedded, so you can request things directly in that platform instead of going to ChatGPT to paste the transcript.

Fireflies is an AI note taker, and it works across every platform. If you use Zoom, you're going to have the transcript of those notes in Zoom. However, if you use Zoom, Teams, and Google Meets, you might need a note taker that can take notes in all of those and centralise them in one place.”

How do you select the AI to use for carrying out the automation?

“After I have the steps defined, the first thing I do is to train a custom GPT with the instructions.

The instructions are those steps, and also the way I want that output to be delivered. Does this output have to be a formatted document, or does it have to be a spreadsheet? What does it have to be?

Custom GPTs are GPTs that you can train. If you have the ChatGPT Plus subscription for $20 a month, you can create your own bots. Those are the best $20 you will spend in your life, I would say.

For a custom process like the one I just referred to, if you want to train a bot to do that, you can do it on custom GPTs. In the same way that, if you want your bot to write like Shakespeare, David Bain, or Victoria Olsina, you can train it with my output, your output, or Shakespeare's books, sonnets, poetry, and plays.”

How do you set up a custom GPT and train it to be as efficient and effective at this automation process as possible?

“In order to train a custom GPT to set up an automation or anything, there are five things we need to do.

First, you need clear instructions. What is your assistant's purpose, audience, rules, and tone? What do you want it to do, and what are the things you want it to avoid? That's the first step. There's something called the CLEAR framework for prompting, which is very useful for this.

The second step is to add your content. For me, this is key. Custom GPTs have a knowledge base, and you can drag and drop files into that knowledge base. If you want an SEO audit to be written in the style of Shakespeare, you would drag and drop Romeo and Juliet, King Lear, and Hamlet, and then it would be written in the style of a Shakespeare play. You give it ideal samples of how this will be delivered, and a style of writing as well.

The third step would be to set technical parameters, like the response length or whether you want this to be delivered as text or a spreadsheet. The fourth step is to test and refine. That’s very important. Test it with realistic queries, see the answers, then tweak the instructions, or maybe assess whether the knowledge base requires more content for high-quality results.

Finally, you can share it, which is why I prefer custom GPTs to Claude projects. I can share a custom GPT with you, and it's ready to go. If I were using Claude projects, I would have to share the instructions with you, and you would have to train it in your own Claude space.

With custom GPTs, I can build a custom GPT for a client and share it with them. It's ready to go and be used by hundreds of people if you want.”

What makes you use custom GPTs instead of a service like Make.com?

“I use custom GPTs when I have to iterate a lot. For example, with open questions.

I have a client who does Bitcoin loans. An open question for that client could be: ‘Based on keyword research, would you call this product Bitcoin loans or Bitcoin finance? Let's discuss.’ Discussion requires back and forth, and you need a conversational bot to do that.

With Make.com, you need a process that is already defined. You give it the keyword, and then say, ‘With this keyword, query NeuralWriter or Surfer SEO and give me the content brief for this query.’ There are no decisions made. There is no discussion.

I generally go to Make.com when my process is 100% defined, I know what happens after each step, and there's no room for discussion. Both of them have a use case. Custom GPTs are better for newer processes that haven't been totally defined yet. For Make.com, you need the end-to-end process.

Also, it's much harder to do it on Make.com. Depending on the length of the automation, it will take a much longer time to test and ensure that each module works correctly. With a custom GPT, you can train it in around 15-30 minutes, depending on the complexity.

With Make.com automation, I have never done one in less than an hour. There are a lot of variables to consider, and things break. When you are defining models on Make.com, you realise, what if X happened? What if I don’t have Y variable? A lot of issues will come up when you’re trying to automate something end-to-end, so it’s very time-consuming.”

What is a Model Context Protocol?

“You can picture MCPs (Model Context Protocols) as the evolution of APIs. With APIs, we can connect data (like Keywords Everywhere, for example) to custom GPTs.

I have a custom GPT that does AI keyword research, and it’s connected to the Keywords Everywhere API. That means that, if I give it the query, the country, the currency, and the intent, it will start doing the keyword research for me. Now, I have to connect that API manually, and I did that by creating a JSON code (which I wrote with the help of AI).

MCPs are a connector. You just copy a URL or a very short script, and it will make your AI ingest and work with the data from any platform that has an API. It's a simplified API. You can do the same thing, but in one, you have to create the API connection, and in the other, you just click a button or enter a URL. This was developed by Anthropic, which is the company that created Claude.

MCPs are like a universal adapter for plugs. You have the data and, with this, you can convert it to any LLM. It's really cool.

The custom GPTs that I like the most are the ones that are connected to APIs, and I made those connections. MCPs are not revolutionary for me, but they're really useful. They are much cooler and much more on-trend than custom GPTs right now, but you can do the same thing. It's just that one needs a bit more coding.”

How do you keep up with what is on-trend and the technology that will have the biggest impact on your day-to-day life at the moment?

“I have been watching videos about AI and SEO every day for the last three years, on every platform that I'm on. On LinkedIn, Twitter, and YouTube, I'm part of different communities.

By osmosis, whatever social media I open and whatever I do, something about AI will come up. For example, the other day, Gemini released Nano Banana. There's not a social media I’ve opened where they are not talking about Nano Banana, which is a really impressive photo model.

I open YouTube, and I see people showing it. I open Instagram, and I see people showing the same photo model. Everywhere I open, I see Nano Banana because it was launched last week.”

Victoria, what's the key takeaway from the tip you shared today?

“Define a clear process and then automate that process with custom GPTs, if you need an assistant only, or with low-code tools like Make.com, if you need an end-to-end process ready.”

Victoria Olsina is an AI Consultant, and she runs her own agency. Find out more over at VictoriaOlsina.com.