Is It Still Worth Making a Newsletter in the Age of AI?

In a world where AI can find information, summarise articles, and bring answers directly to people, do we really need another email landing in someone's inbox?

Good question.

And I don't have the answer yet.

But I do have a reason to try.

Why am I even thinking about a newsletter?

We're building a business around AI Thinking, and part of our strategy is to share the process of building it.

It's not about what we do. It's about using something anyone can connect with, like trying to build a business, and sharing the thinking behind it. And how that thinking is reflected in the way I use AI.

As AI is a way of thinking, how can you make someone see thoughts? Only when they are shared around an idea you can connect with.

So here we are. We call it Thought Documentary. :)

And I thought maybe there are still people out there who prefer this kind of interaction and love getting a newsletter. I still see people reading newspapers, you know, the ones made of paper, even though I don't.

So let's see.

Now we need a way to send it

This is where Alex comes in, my lovely ChatGPT assistant, and an opportunity I can't miss to share what I'm here to do.

How do you share thoughts with an LLM whose responses come from its training data and vendor configuration?

Let's take our newsletter as an example.

I could open Google, search for newsletter platforms, visit their websites, compare features and prices, read reviews, and try to figure out which one would work best for us.

Or I could have a conversation with Alex.

And that's what I did.

But the interesting part isn't asking AI which newsletter platform is best.

It's knowing that if I ask for the best, its definition of best will be based on its training data, vendor configuration, and what it knows about me. My specific preferences and what it believes will be best for me.

MailerLite or Mailchimp?

Alex's first recommendation was MailerLite.

For the logic behind its answer, it pointed to the free plan.

And I asked, "What about Mailchimp?" knowing they have a free plan too.

Only to reveal its logic, without asking it directly.

Asking an LLM something directly can change its answers, pivoting it towards agreeing with me, which would prevent me from understanding what the logic was to begin with.

So we looked at the differences.

MailerLite's free plan offered 250 subscribers and 2,500 emails per month, while Mailchimp's free plan offered 250 contacts but only 500 emails per month.

And Alex gave me an example.

If we have 200 subscribers and send a newsletter every week, that's 800 emails a month.

Which means we would already exceed Mailchimp's free sending limit, while still being within MailerLite's.

Now that makes sense. :)

The interesting part wasn't that Alex recommended MailerLite.

It was understanding why it recommended MailerLite. In this case, I realised Alex was prioritising the free plan as one of the most important considerations for me.

Alex got that from the time we had spent researching token consumption and the cost of using AI.

So it made sense to me why Alex might have given so much weight to getting something for free.

But this wasn't the same decision.

And now that I understood what was influencing the recommendation, I could direct the conversation differently.

How do you direct an AI agent's thought process?

Once I understood what was influencing Alex's recommendation, I wanted it to follow my way of evaluating software.

For me, when comparing software, I prefer looking at independent review sites like G2 Crowd and Capterra.

So I directed Alex to use these sources to compare MailerLite and Mailchimp.

The point wasn't to get Alex to change its recommendation.

It was to give it a thought process to follow instead of picking things on its own, deciding what is important for me without asking.

And here's what we found.

Ratings and review counts reflect the date of our comparison. These may change over time.

We looked at how people rated both platforms, how many reviews they received, and what users had to say about working with them.

And we ended up choosing MailerLite.

Now, if I want Alex to carry this thought process into every future recommendation, I have two options.

Creating a skill or saving a prompt.

Where is the best place to start?

Pay attention, because no one will tell you this. :)

Don't start by defining your own instructions.

Because this is a thought process, and once you've followed that process through questions and answers, the best place to start is by asking the LLM to describe to you the method of how you reached a decision.

And see what it got.

Why?

Because you've already gone through the process together.

You've asked questions, challenged answers, brought in information, and made choices.

Now you want to see what the LLM understood from that conversation.

Did it recognise why you directed it to independent review sites? Did it understand what information mattered to you and why?

Or did it simply conclude that you prefer MailerLite? :)

That's a very different thing.

And this is why I wouldn't start by asking Alex to create a skill or write a reusable prompt.

First, I want to see whether it understood the method, not just the result.

If something is missing, I can correct it.

If it has added something that wasn't part of my thinking, I can question it.

And once the method reflects the thought process I want it to follow, I can decide how to use it again.

Want to try it yourself?

Here is the prompt I would start with, after you've completed a decision-making conversation with your LLM.

Look at the conversation we just had about [decision or subject].

Describe the method we followed to reach our decision.

Identify the questions I asked, the information I directed you to use, the assumptions I challenged, and how we compared the alternatives.

Separate the method from the final recommendation.

Show me the steps in the order we followed them, so I can review whether you've understood my thought process before we turn it into a reusable instruction.

Don't add steps or criteria that weren't part of our conversation. If something is unclear, ask me.

Now look at what it gives you.

Does it reflect your thinking?

If not, keep the conversation going until it does.

And only then decide whether you want to save that method as a prompt or create a skill that will follow it whenever you need to make a similar decision.

The thought process comes first. The instructions come after.

And that's how you start building an AI that works with your thinking, rather than leaving it to decide what matters to you.

As we say in AI Thinking, shape it or be shaped by it. :)

Heads up!

The people at MailerLite, thank you for having us.

We decided to give you a try.

Was it the right decision? We genuinely hope so.

To everyone else, we'll let you know. The more we use it, the more we'll tell you how it goes.

Stay tuned for the rest. :)