Important Things I Have Figured Out About Using AI

AI. It is the marketer’s little helper. For now at least.

I wonder how redundant we will become as AI becomes ever more capable and powerful. But, as I say…For now it gives online marketers a huge opportunity.

I have found myself using it more and more over the last year or more. I even upgraded to a paid ChatGPT account as I could finally see that it what it enabled me to accomplish far outweighed the $20 + VAT a month that it costs me.

Along the way I have made some interesting discoveries that help me help the AI to help me (if you follow all that). 

A quick disclaimer – I don’t know the inner workings of these AI models. I am not saying that all of these tips are strictly necessary. These are simply observations that I have had and seem to improve the output for me.

Here I will pass on a few of the things I have spotted and maybe it will help you too.

One of the cool things is that you can actually see glimpses of how the AI is reasoning out your prompts. 

I guess a good few people miss that – they set it to do something and while it’s doing it’s “extended thinking”, they switch browser tabs or make a coffee rather than look at the thing working stuff out.

I could actually see what was wrong with my prompts as I saw the process spool through to it giving me an answer.

This sharpened up my prompt writing no end. I could anticipate what I needed to ask, how detailed and specific I needed to be and I had to make sure I didn’t leave something out.

Something I always do (and this comes from creating my own dedicated AI apps), is to clearly and explicitly define the role of AI in the chat right at the very beginning. This sets out the stall, so to speak. It will frame what you ask it based on that role moving forward.

This is very important for “agent mode” but I find it gives far better results in general chats too. Obviously you don’t need to set a role for it when asking simple, short questions or tasks. But if you are doing something bigger, it helps massively.

I finish this initial role prompt with this “Please confirm that you understand your role and focus for this chat”

The next insight refers back to a story I always tell students who are starting to create their own digital product for the first time.

The story is about a family of Neolithic hunter-gathers. They normally ate nuts and berries and whatever they could kill. So birds, squirrels, maybe a small deer if they were lucky.

One day, and don’t ask me how they managed it,  they killed a fully grown woolly mammoth.

As they surrounded the carcass in awe, one of the youngsters broke the silence; “How on earth are we going to eat something so big?”

To which an elder replied; “One bite at a time, son. One bite at a time”

And this applies when you start creating a project using AI.

If you dump the whole task on it all at once, it never seems to get it right. If you break things down into steps, though, it seems to cope a lot better.

Here’s an example. If your goal is to create a video course on a topic and you want AI to write the content that will be made into your videos, the last thing you would do is ask it to create a video course.

You would first tell it what your topic or subject is and ask it to research the top problems that people experience in doing whatever your topic is. You may already have in mind the problem you want to solve but AI may come up with something even better. 

You then pick a route and ask the AI to create a 6 part (or however many you want) framework – lesson headings as bullet points with one sentence to define what each lesson covers. 

Then, one by one, you ask it to flesh out the lessons.

When you have this, you tweak as you see fit and then ask it to make a script for a video for each lesson. When you have that you can ask it for an action plan and checklist for each lesson that are course handouts.

Then, until ChatGPT can do this for you, you head over to a tool like Pictory and very quickly and easily make the videos from your scripts.

When asking it to start on one of these “micro-steps” I find it useful to get it to confirm that it understands what I am asking of it. It will then answer with it’s interpretation of the brief which you can accept or correct as need be. This saves a lot of back and forth.

Sometimes, these chats get quite long and are spaced out in different sessions. Maybe the robot can carry on but you need to have a poo or eat dinner and sleep at some point.

I always find that when I return to a general chat in a brand new browser session (even on a different day) I start with asking the AI to review what we have achieved so far, where we have reached and confirm that it is fully up to date with the project.

This seems to re-focus the AI to the task in hand. As the chat expands and moves through the different phases, it can seem to lose a bit of focus so it’s always worth doing from time to time.

Lastly, when you have your project all completed. Ask it for some sales copy bullet points and a prompt for a AI chat to create sales copy for the project. This will help you if you want to use a different chat (or agent mode chat) to help you write the sales page.

Conclusion

So here are the things I suggest doing:

• Watch the model’s “thinking” to spot prompt issues and tighten your instructions.

• Be explicit and specific—anticipate details so you don’t leave gaps.

• Define the AI’s role at the start of a project and ask it to confirm understanding.

• Tackle big projects “one bite at a time”—break work into clear micro-steps.

• Use a structured workflow (e.g., course creation: research problems > pick angle > outline lessons > expand > script > action plans/checklists ? produce).

• At each micro-step, have the AI restate the brief to align before proceeding.

• In long or multi-session chats, re-sync by asking for a recap/status to refocus the AI.

• When the project is done, generate sales-copy bullets and a reusable prompt for a separate sales-page chat/agent.

  I hope this helps you. Comment if it does!