✨ Use this ChatGPT method for better results

There is a lot of (wrong) information on the internet on how to prompt ChatGPT to get good results. Luckily, OpenAI dropped an informative video on exactly how ChatGPT works, and what the proven methods are that will get you better results.

The optimization flow of ChatGPT

There are four ways to improve the performance of ChatGPT.

1. Prompt Engineering: This is like playing the right notes. You give ChatGPT clear and specific instructions on what you need.

2. Retrieval Augmented Generation (RAG): This gives ChatGPT a library of information to reference, so it can pull in the right facts when answering your questions.

3. Fine Tuning: This is outside our scope as it’s highly technical, but it involves training ChatGPT on specific types of data to make ChatGPT answer in a specific way.

4. A combination or all of the above: Depending on your scenario, you can create a mixture of these methods.

Let’s go over them.

Prompt Engineering

Prompt engineering is the practice of creating inputs for ChatGPT(or other AI tools) that will produce the best possible results. Within prompt engineering, there are multiple proven methods:

• Specific Instructions

Be as specific as possible. Give it some thought before you write out your request, it’s smart, but it won’t be able to read your mind.

❌ Vague instructions:

You will be presented with a news article. Extract useful information in a structured format. 

✅ Clear instructions:

You will be presented with a news article. Your task is to identify any opinions expressed about the government, and their sentiment. 

• Give it time to think / Chain-of-thought prompting

Sometimes, it helps to ask ChatGPT to explain its reasoning or take it through the steps of a process.

This not only increases the chance of an accurate response but also makes it easier to spot where things might have gone off track if the answer isn't quite right.

In the example below ChatGPT would’ve gotten a simple math question wrong if chain-of-thought was not applied.

• One/few-shot examples

Show it examples of what you’re looking for. For instance, "Here are three examples of marketing emails. Can you write a fourth one following a similar style?"

Often this works better than describing what you are looking for, as your instructions can be ambiguous or wrongly interpreted.

💡 Tip: To get the best results with examples you can try adding between 1 and 4 examples, depending on how big the examples are.

Retrieval Augmented Generation (RAG)

Retrieval Augmented Generation (RAG) is a fancy term, but it basically means providing ChatGPT with a custom file it can reference. When you ask a question that requires specific knowledge, RAG helps ChatGPT to look through its database for related information to come up with a response.

For example, I wrote this section on prompt engineering using RAG. I uploaded a transcript of the video to ChatGPT and used that to write this.

It sounds difficult, but there are easy ways to apply this (but you do need the paid plan of ChatGPT)

1. Upload a file (e.g. a PDF) to ChatGPT to be able to talk with it.

2. Create your own GPT and upload custom files it can reference.

3. Create your own GPT assistant and upload custom files there.

Another reason to get ChatGPT Plus! (I promise they didn’t pay me to say this)

Emails are one of the most time-consuming ways we spend our time at work. The average worker spends 28% of their workday reading and answering emails. With AI we can bring this down—a lot.

Let’s take a look at the different ways we can use AI to do this.

ChatGPT

ChatGPT is great at drafting professional emails, you just need to provide the input. There are two ways you can use ChatGPT for emails.

1. Via a simple prompt

A basic prompt with raw input that you want to be mentioned in the email should do the trick. How this can look like:

Compose a friendly and concise email to Judith at UltraCorp. 

Begin with a brief update on the upcoming campaign starting next Thursday, the 25th of January. Include essential details like the €50K budget and a reminder to review the attached file for campaign specifics. Conclude by requesting her final approval before the end of the day on Friday the 19th of January.

Follow the tone and structure of this email:
[Example email]

In this prompt we describe what we want ‘a friendly and concise email’ with the ‘input’ the email should contain. Alternatively, you can also add an example of a previous email you have written to give ChatGPT a reference.

This works great because:

• You only have to provide the input without being worried about grammar or structure.

• You can easily give feedback to optimize the email.

2. Custom GPTs

If you have a paid account, you can create a GPT (a custom ChatGPT) that you can design to answer your email.

For example, you can create a GPT called ‘Work Email GPT’ that is private and only accessible to you, with instructions about who you are, the company you work for, and your writing style. You could even add a document for extra content (RAG) that contains relevant information to answer emails quickly.

Grammarly

Grammarly, is a (free) tool to help you write beautifully without spelling mistakes. And since recently it also contains a lot of helpful AI features.

It’s integrated into your browser and can assist you on every page where you write, like emails.

Once you have Grammarly installed, here’s how you can use Grammarly to create complete emails in seconds:

1. Open or create an email.

2. Click on the 💡icon.

3. Write in keywords what you want the email to contain.

4. Hit generate and insert your email.

It’s that easy. You can even explain your writing style beforehand so that it sounds more like you.

Automations

With an automation platform, like make.com, we can create automations that handle our emails for us.

Next week, we’ll include 3 different automation guides in the email (and on the website) on how you can:

• Automatically create a draft reply for every new incoming email

• Automatically can label emails

• Automatically let AI read emails for you and send you a summary on the bits that are relevant for you (for example newsletters)

Stay tuned!

Which AI tools were released last week? Here are our favorite picks.

• Modelize.ai - Creates automated AI workflows. For example, ask ‘I want to sell toys for babies online’ and Modelize will do automatic market research, write a business and branding plan and come up with marketing tactics.

• Guidde - The easiest way to create how-to guides and SOPs.

• Tabilize - Make interactive data dashboards from your spreadsheets.

• ubique - Send personalised sales videos at scale.

• Podsqueeze - Automatically turn your podcasts into short clips, blog posts, social posts, and much more.

What did you think of this email?

You can add more feedback after choosing an option, or you can reply to this email directly.

Enjoyed this newsletter? Forward it to a friend and have them sign up here.

See you next Thursday!