Part 5 - Personal GPT writing agent

This final long form content generation suggestions is going to get a bit advanced.

BUT if you put in the work you'll have a super valuable asset.

We're going to be using ChatGPT's new GPT Builder.

ChatGPT allows us to create our own GPTs - our own little AI models that perform specific tasks for us.

I'll keep this high level to show you the potential, get you excited and give you an overview of how to set this up.

You will need a paid ChatGPT account for this.

First, navigate to GPT Builder and describe what sort of GPT you want to build.

In this case we'll tell it something like "a writing tool that uses my past writing to prepare new long form posts"

We are then going to give our GPT all our written work. We’re going to upload past writing and use it to “train” our GPT.

How you do this depends on where your content is hosted.

The simplest yet most time consuming is always going to be copy pasting all your posts, blog articles, newsletters etc. into a Word or Google Document. Get everything in one place and then upload that document into your GPT.

Ideally though we can download copies of our work automatically.

For example

• if your newsletter is in beehiiv you can download a backup of all your content from Settings.

• If you write on X a lot you can request an export of all posts from Settings, Privacy.

• If you have a Wordpress blog you can go to Tools, Export, Posts.

I could go on! How you get your posts really depends on where they are. Generally though there will be some sort of export tool function built into your publishing tool.

Once you have a copy of your writing we’re going to upload it.

Head to configure and look for the Knowledge section to upload your past posts. You can upload multiple files.

For this example I’ll just upload my beehiiv newsletter data. For best results upload your blog articles, long form tweets and more.

This will allow your GPT to be "trained" on your style, tone of voice, vocabulary usage etc. all based on past writing. Obviously the more you have written in the past the better!

Once we’ve set up the basics and uploaded our data we will give your GPT the instructions it needs to build your writing tool.

Here’s the interface you’ll be working with

On the left you will tell ChatGPT what sort of tool you want to build.

On the right is a preview of your tool.

First we need to give some basic instructions on the left hand side to configure the tool. This is very simple - we don’t need to code. We just need to tell ChatGPT what we want the tool to be like.

This is an extension of all the prompt engineering work we’ve been doing before.

To start simply answer the questions ChatGPT asks you. For example

Here ChatGPT has asked me what it should do when it needs information - should it just wing it and make stuff up or ask me?

Because the veracity of information is important to me in this case I’ll say to ask for clarifications.

ChatGPT will continue to ask more questions like this to get the basic parameters of my tool in place.

Once that’s done we need to start using your GPT by testing on the right hand side. The right hand side is a Preview of how the tool will act - it’s here we test the tool and find areas for improvement.

For example if I asked for a blog article now ChatGPT produces it in American English. That’s the default.

I personally write in British English though so on the left hand side (Create) I would tell ChatGPT to only use British English. That request will be saved into my model and from now on the tool will only use British English.

Optimising (or “tuning”) the model is a process of using the preview, finding areas where it can improve, adding the improvement suggestions in the left hand Create panel, testing the model more, finding more improvements etc. etc.

We do this until the outputs of the Preview model produce what we expect (long form articles) to the quality we require.

With your own GPT you can extremely rapidly produce long form content on hundreds of topics, all trained to your past writing and with your saved requirements and improvements.

Don’t expect the final outputs to be 100% perfect. Again, as with all AI writing, I’d recommend still using the output as your first draft and rewriting yourself.

GPTs Space

Creating your own GPTs is a very new, very exciting development which only just dropped.

I’m playing around a lot with them at the moment and will be writing a lot more about custom GPTs.

Because custom GPTs are such a game changer I’m opening up a new space in the community where we can talk about them, share progress, give advice, ask for feedback etc. etc.

This idea came from Joshua, one of the community members. Make sure to follow him on X: https://twitter.com/joshuahaleio Thanks Joshua!

Pulling it together

We’ve covered a whole slew of different long form content generation techniques this week. Hopefully at least one or two will be useful in your workflow.

Don’t feel that you need to incorporate them all - instead think of this as a menu of options for you to try and choose from.

As long as you’ve come away with one technique that makes long form content easier then I’m happy!

A reminder of what we covered this week

Part 1: Co-writing using ChatGPT

Part 2: Format/Framework Matrix

Part 3: Expansive writing

Part 4: Niche-jacking

Part 5: Personal GPT writing agent