Building an AI Team using AI PART 2
⚙️ Building an AI Team using AI | Prompt Playbook
Today we’re going to be using the instructions below as well as providing our custom GPT with training data. This is a little bit more advanced than the previous GPT we built.
Create a new GPT and plug in these basic instructions to get started
Create a social media content assistant for my business.
The tool will be used to generate social media ideas and posts for you my business.
Use the uploaded social media posts and analytics as Knowledge of my past social media content. Analyse the analytics to identify the most successful content and aim to replicate.
Step 1, based on the training data generate a list of content ideas and ask me to identify (by number) the idea to focus on. Identify the content pillar for each idea.
Step 2, for the content idea select generate multiple posts. Create posts for all three levels. 10 posts at level 1, 5 at level 2, 1 at level 3.
Content should be created at three levels:
-Level 1: wide, awareness generating viral content. Accessible and not complex. Short posts, one or two sentences maximum.
- Level 2: medium complexity expert opinions and hot-takes. Medium length, 280 character max.
- Level 3: high complexity guides, how-to content. In depth, not for beginners. Long form tweet/blog post format, up to 1000 words.
Do not include hashtags or emojis.
Match my tone of voice and style.
The post provided should be the whole post, not a preview or snippet.
Do not provide the levels in step 1, only in step 2.
Provide outputs in a tabular format.
Custom GPT will also ask all the standard questions about the name of the GPT, its logo etc. Go ahead and answer all this.
Now we’re going to upload training data. ChatGPT calls this “Knowledge”.
You can find this in the Configure tab
Knowledge allows us to upload multiple files and our custom GPT will use this information in its answers.
For this particular GPT we want to upload our past social media posts.
What specifically do we want to include? This depends on what platforms you are producing content for.
If this is primarily for X content then upload your X data.
If it’s primarily for LinkedIn then upload your LinkedIn data.
Ideally have one GPT per social media network - this allows us to keep it more specific and train for better results for that specific platform. Trying to do too much will dilute the results.
In my GPT I’m focusing on X so I download an archive of all my past activity. This include my posts as well as the analytics (views, likes, retweets etc.) for each post. This is helpful because it gives the content as well as information about what content was the most successful in the past.
Upload any and all relevant files so that your GPT has access to the data.
Then, as before, we want to start testing the model and providing feedback.
Don’t expect the model to work immediately - we instead want to see it getting stuff wrong. Every time it produces an error or poor results we have an opportunity to teach it. So keep using the Preview window, finding fault and providing feedback!
Premium Prompt - Pre-built Social Media Content GPT
Here’s a link to my version of this tool
https://chat.openai.com/g/g-sjQZFDvXi-social-media-content-gpt
This is useful for you to check out the potential but because it is trained on my content it’s not valuable for your own social media!
Make sure to upload your own content in order to personalise the GPT!
Here’s another quick tip - let’s say you are still finding your style. You can actually train the model on the content of writers your admire.
For example I like Dan Koe’s material a lot. When training my GPT I can’t directly upload all his analytics because I don’t have access. But I can give the GPT a link to his Twitter, blog and other places where content is found.
This allows you to emulate the voice of other influential online writers.
To take this a step further you can add in specific elements from different writers. For example I might bring in Dan Koe’s Twitter formatting style, Tony Robbin’s influential language all mixed with the content topics of Dickie Bush.
I would feed in content from webpages and the precise instructions for what elements my GPT should be looking to emulate.
This is a useful “hack” if you haven’t yet refined your own style and are looking for a way to get started.
As before come to the community with your building experiments! I’m there to answer questions.
Pulling it together
Another day another Custom GPT. Today we introduced the technique of adding Knowledge data to help train our GPT.
A reminder of what we’re covering this week
Part 1: Getting started with custom GPTs for business
Part 2: Building a social media content GPT
Part 3: Building a newsletter GPT
Part 4: Building a sales copywriter GPT
Part 5: Building a business scaling coach GPT
Until then, keep PROMPTING!
Kyle