Welcome back!

Last week we created a prompt/GPT that writes outlines for your blog posts.

This week we will be generating the final articles based on these outputs.

First of all, this method gets results. I’ve created some blog articles for you to check out, all completely generated with this method

Examples of final outputs via the GPT-4 API (better than regular GPTs)

• A night owl's guide to becoming a morning person

• The ultimate guide to crafting effective email flows

• Your article? Want me to try to write in article in your tone of voice? Fill in this form.

Table of Contents

• Writing the article itself, what’s important?

  • Extra research with facts from the internet

  • Principles for good blog posts

• Getting the tone of voice right

  • Gathering examples of the writing you want

  • Analysing your own tone of voice and getting a bri …

  • Making the AI sound less robotic.

  • Creating a persona with personality

  • Adding the final spice to the AI to write more hum …

• Combining the information into a GPT or GPT-4 API

  • 1 really big sidenote:

Writing the article itself, what’s important?

Extra research with facts from the internet

AI tends to give very generic and surface level information and advice. This is not what we want. We want a GPT that gives relevant, useful and unique information.

There are 3 ways to do this:

1. Add some research yourself to the prompt (too much manual work for me)

2. Try to add a research step into the writer GPT (could be done, but might be too much for the GPT to handle)

3. Build a separate GPT that researches based on the outline for us (will give us a list with information we can give to the writer GPT

It works like this:

1. You paste your outline into the GPT.

2. It makes a list with subjects to research

3. It starts to research the internet for relevant information

4. It makes a summary per subject.

5. You can paste that into our writer GPT when he asks for it.

Try the GPT here yourself

Less generic information = more relevant blog = win.

Principles for good blog posts

Without good instructions for the final blog posts/article, AI will give you a mess. We want to give the AI some guidelines.

With a quick ChatGPT conversation, I got this as structural points for a good blog post (with some personal add-ons)

Engaging Title and Logical Structure:
- Craft titles under 60 characters long, containing primary keywords.
- Use headings and subheadings (H2, H3, H4 etc.) to structure the content logically. Always write this with lowercase letters except from the first words and names.

SEO Optimization:
- Integrate keywords naturally into the title, subheadings, and body of the post.
- Use proper header tags for headings and subheadings.

Readability and Accessibility:
- Write concise introductions (maximum three paragraphs). Never directly talk to the target audience with "Hey you" like sentences.
- Write for readers and scanners. Make sure you add enough whitespace between texts. To emphasize an important sentence add 2 enters in front and 2 enters in back of the sentence.
- Use the BLUF (Bottom Line Up Front) approach for clarity.

Word Count and Length: Aim for a minimum of 1500 words for SEO effectiveness.
- The optimal word count can vary but often ranges between 1,600 to 1,800 words.
- Adjust word count based on the audience, topic complexity, and content type.

Title and Meta Description:
- Ideal blog post title length: 60 characters.
- Meta descriptions should be under 155 characters.

Lists and Formatting:
- Every 3/4 sentences there should be a some form of other formatted text. This could be a list, but also just 2 enters, to keep the text easily scannable.
- Use bulleted or numbered lists to break up long text and make content scannable.
- Structure lists for reasons, steps, tips, etc.

Call to Action and Conclusion:
- Conclude with a summary or wrap-up of key points.
- Incorporate a call to action, guiding readers on the next steps or further engagement.

Getting the tone of voice right

We have 2 goals here:

• Making the AI write like you or your company would write.

• Making the AI sound less robotic.

To make the AI write like you we will do 2 things:

• Gather examples and let AI analyse them for tone of voice and get instructions how to recreate them

• Use the same examples in the final prompt, to double prime it to do it right.

Gathering examples of the writing you want

Aim for getting 3 good examples of the writing style you want. It’s works best if you give the AI examples of what a final output would look like. So If you want to write blog posts, you give it blog posts as example. If you want emails, you give it emails as example. This works best.

For tone of voice it’s important to consider:

• If your tone of voice is generic, it will be harder for an AI to mimic this like you want it. It’s easier for the AI to write like Nike or Apple than a more generic tone of voice. The more personality your current texts have that you train AI on, the better.

• Garbage in = Garbage out. Take your time to find good examples of your tone of voice.

If the final output is not close to 75% what you are looking for, then looking into other examples is a good next move.

Analysing your own tone of voice and getting a brief on how to write like it

With the examples you can ask ChatGPT to make an analyses on what your writing style is and how to recreate it.

You can use my GPT to analyse your own tone of voice

Important things in tone of voice are:

• Tone

• Vocabulary

• Sentence structure

• Rhythm (sentence lengths and variation on this)

Tips:

• Ask it for brands or famous people that would also use a writing style like this.

• Ask it for the emotions that this tone of voice should bring up in people.

Making the AI sound less robotic.

AI will default to being a super neutral robot. This is perfect getting quick facts etc, but horrible if you trying to recreate a human.

The biggest differences between how humans and robots write:

• Humans have strong opinions, have personal experiences, make small mistakes, don’t always follow the exact rules, talk from an I perspective.

• AI’s are very predictable in usage of words and sentence structure and length.

That’s what we adres now.

Creating a persona with personality

It works well if you give your AI writer a real personality, like a human would have as well.

You give it opinions. A background. An archetype. Personality traits.

So not: You’re a Professional Blog Writer…

But rather:

Adding “You write Humanlike content. It must be insightful, well-organized, and compelling. Everything you write feels like a human wrote it.” helps as well.

Using an archetype can help. This will prime your AI to specific character.

Adding the final spice to the AI to write more human

AI will write in a predictable pattern. The biggest tells (that AI detectors also use) are:

• Perplexity, it measures the complexity of text.

• Burstiness, how much variation there is in sentences.

Humans tend to write with greater burstiness. For example, using longer or complex sentences alongside shorter ones. AI sentences are more uniform.

There for we are going to ask it to add a good amount of perplexity and burstiness.

We’ll add this in the prompt by adding:

Combining the information into a GPT or GPT-4 API

Now we have all our ingredients for a good writer GPT. We are going to structure the prompt for this as follows:

1. Persona

2. Task (step by step)

3. Important principles for blogs

4. Tone of Voice description

5. Final make it more human instructions

6. Deep breath & Step by Step

Test the GPT yourself here: (This is with a neutral writer)

Or build the GPT yourself with your specific instructions:

1 really big sidenote:

This works even better if you run this completely via the GPT-4 API. All GPTs currently run on the ‘better’ GPT-4 Turbo.

However, this model has proven to be less accurate compared to the original GPT-4.

GPT-4 is more expensive & will take more time. But I would say that’s more than worth it for a little extra accuracy.

If you need help with this, shoot me an email. I can set this up for you.

Next week - Generating Engaging Images with DALL·E 3

AI will think of concepts for engaging images for our articles. We will create images with a consistent style, using your own brand colors, emotions etc.

Be on the lookout for these newsletters coming weeks in February on Thursdays:

1. Feb 1: Building the Perfect Blog Outline with ChatGPT (Step-by-Step Guide)

2. Feb 8: Writing Human-Like Blogs with ChatGPT (Step-by-Step Guide)

3. Feb 15: Generating Engaging Images with DALL·E 3 (Step-by-Step Guide)

4. Feb 22: Running Your Blog on Autopilot with AI & Automation (Step-by-Step Guide)

Thanks for reading. My last request: help me make this newsletter better by taking 1 minute to tell me about your struggles with AI. So in the future I can help you even better.

See you next week.

Are you excited about this process but just don't have the time or know-how to get it off the ground?

No worries, I can help you! I’ll build a custom app for you to control your blog with AI. All tailer made for your company, by me.

• A super easy-to-use interface that lets you monitor and control the AI like a breeze.

• You'll be able to turn simple keywords and ideas into full outlines, articles, and images, all with a few mouse clicks. Plus, you can easily tweak and get new versions based on your feedback.

• You can upload new articles to your website in a snap, whenever you feel like it (as long as it works with your platform).

Sound good to you? Fill out this form and I'll be in touch to explore all the possibilities with you.