top of page

🛟 RAFT: A Lifesaver to achieve better results from ChatGPT with prompt engineering

At Structize AI we specialise in workflow automation supercharged with Artificial Intelligence. Thanks to our Structz, we make tasks like Data Extraction, Summarization, Categorization or Translation super easy for companies without the need of prompt engineering.

As we are one a mission to stop the copy-paste madness for knowledge workers to save them more time for higher value-added tasks, we want to share in this post how you can build your own Structz - even without the Structize AI platform. Yes, you read this right, just follow these hacks and it should get you already very close - and if not, well you know where to find us 😉.

In this post, we give you 4 hacks to get started yourself with the basics. The basics are really just writing superb prompts. We have dubbed the framework as RAFT 🛟 as it has been a lifesaver for us. We trust it can be one for you too. This is what RAFT stands for:

R : Assign a Role

A : Give it Areas to focus on

F : Define response Formats T : Try different prompts

Let's try two examples:

Example 1: Generating SEO Meta descriptions for a food blog - Entry level

Desired outcome: Generate SEO Meta descriptions for each of your 89 articles on a food blog.

Prompt: Take these 89 articles and imagine that you are an SEO Marketeer (Role) with a website about food recipes (Area). Could you provide me an SEO-optimized text of maximum 160 characters in English per recipe that I have just uploaded? (Format)

Check the answer and if not satisfied, tweak it and try again ♻️ (Try)

Example 2: Financial analysis between different companies - Moderate Level

Desired outcome: You want to analyse profitability growth metrics between 3 companies; Apple, Nvidia & Microsoft using their Annual Reports.

Prompt: Take these 3 Annual Reports and imagine that you are a Financial Analyst (Role) and that  you are specialised in Tech Companies (Area 1). You are focused on comparing profitability between different companies measured in EBITDA (Area 2) .

Create a 3x2  table with each row populated with one company: Apple, Microsoft & Nvidia. Now per row, please provide 2 columns With in Column 1 the EBITDA for 2022 and in column 2 the EBITDA for 2023. Both column 1 and 2 should be a number and in US Dollars. (Format)

You could add a column 3 with growth rates by dividing column 2 by column 1 (in our experience however this goes faster in excel after generating the initial table than directly in GPT, Mistral or Llama).

As always, check your answer and if not satisfied, tweak it and try again ♻️ (Try)

That's it. The basics are just that easy !

I you are interested in scaling your prompts into standardized workflows that run in your organisation and save tons of time? You are at the right place! Get started today:


Commenting has been turned off.
bottom of page