How to get better outputs from generative AI

CIM’s Laura Bracher interviews AI expert Dave Birss and Saurabh Wani, content and growth marketer at AI generation platform Writesonic, about how you can maximise the use of your AI tools. In this article, we outline different ways to approach and pinpoint your prompts for stronger outputs. 

As a copywriter, I admit I was a little alarmed the first time I used ChatGPT. There's no denying that it's a groundbreaking technology with the potential to change how we work. 

Plenty claim to have experienced amazing productivity gains from AI. A study by researchers at MIT found that ChatGPT lowered the time taken to complete a mid-level professional writing task by 40%, and overall output quality rose by 18%.

I can't say that's been my experience.

Generative AI is clearly a powerful tool, but my encounters haven't been particularly fruitful, and there are a few good reasons for that.

Firstly, much of the content I write is ghostwritten articles for thought leaders, which requires original ideas that AI can’t currently predict. Secondly, for the work where AI might assist me, like article research, preparation, and creating sales copy, I suspected my prompts needed some work. 

I spoke with two AI experts about how I can increase my productivity using AI.

What is an AI prompt?

AI prompts refer to specific input instructions or queries given to a language model to generate desired outputs. These prompts are used to interact with the likes of ChatGPT, CoPilot and Gemini to receive text-based responses. The quality and relevance of the generated responses depend on the specificity and clarity of the prompts given.

This rise in generative artificial intelligence has caused some companies to hire "prompt engineers" who are responsible for training AI tools to deliver better, more relevant responses. But prompts aren't just for training specific language models. Even I act as a prompt engineer when I ask ChatGPT to write something for me.

How can marketers benefit from better prompts?

According to Dave Birss, who helps organisations use AI more effectively, we need to change our thinking. 

"The best way to use AI is to augment your skills rather than simply using it to reduce effort. When you use it well, it can make you capable of previously unachievable things. It can help you plan and build a prototype of a digital tool. It can help you analyse complex data. It can help you understand your audience better. 

Yes, you can use it to answer emails faster. But it's so much better when it gives you superpowers." 

Saurabh Wani, content and growth marketer at AI generation platform Writesonic, supports this view that the real advantage of generative AI is its ability to help us think outside the box. 

"Diversifying the usage of AI platforms and experimenting with various prompt types means marketers can find new strategies for customer experience that are smarter and better equipped to address customer concerns. Ultimately, improving proficiency in using AI prompts allows marketers to elevate their marketing endeavours and maintain a competitive edge in the industry."

Garbage in, garbage out

As mentioned before, my experiences with generative AI haven’t been great. Most of the time, the answers have been far from perfect. They often miss the mark, but I will hold my hands up and admit my prompt writing has been pretty poor so far. So, how am I getting it so wrong? According to Birss, I need to be more specific.

“One of the mistakes that people often make is thinking of tools like ChatGPT as a knowledge machine rather than a thinking machine. It has a heck of a lot of knowledge, but you can't rely on that information to give you the best output,” he says.

"The information it has is general knowledge. But to solve your problem, you need to feed it with specific knowledge. That means writing a more thorough prompt that gives the bot all the information and context to provide the best response.

The rule you should remember is that you can't give the AI less information than you'd give a human to do the same job and expect excellence. The age-old law of “Garbage In, Garbage Out” applies here."

How to write better generative AI prompts

We don't know what's been fed into these chatbots as part of their training. And no matter how good the prompt, there's always a chance that the AI will make things up, so fact-checking is always needed.

But there are things we can do when crafting prompts to ensure the best possible outcome.

1. Give it a character to play 

Birss highlights that AI chatbots are particularly good at stepping into character (explaining quantum theory in the style of Snoop Dogg is one example I've seen), so you should start a prompt by telling it who you want it to play. 

"You might choose an industry expert, a frustrated client, Socrates or an alien visiting Earth for the first time. ChatGPT is particularly good at stepping into character to give you a response from an interesting point of view. This helps you focus on the kind of response you're looking for and what might be valuable to you. You can be as imaginative as you want with this. For example, you could ask for a response from the point of view of a water droplet being squeezed through freshly ground coffee beans into a mug of artisanal coffee and you'd get something interesting and maybe even valuable."

2. See your prompt as a brief 

To get the best output with generative AI, you need to see your prompt as a brief, which means no single sentences, as Birss explains.

"Just like any marketing project, - the quality of the response is dictated by the quality of the brief you start with. My prompts are paragraphs and sometimes pages long. If your prompt is a sentence, you should rethink your approach.”

Wani echoes this, explaining that supplying background information and being focused will result in a more accurate response.

"To enhance the AI's response, offer context by providing relevant background information. This approach helps guide the AI's understanding and ensures a more accurate response. Being specific with your questions also aids in obtaining accurate and detailed answers. Rather than asking for general information, you can ask a focused question. By doing so, you prompt the AI to provide specific and informative insights."

3. Build a prompt library

Prompts involve a lot of trial and error, and when you find one that works well, you don't want to forget about it or type it repetitively, which is where a prompt library comes in handy.

"Good prompts often take a fair amount of effort and experimentation. You want to make the most of that by reusing your valuable prompts – and maybe sharing them with your colleagues," explains Birss.

"I've been helping my clients build prompt libraries for their teams that give everyone access to tested and approved prompts that help them produce better work. This involves assigning responsibilities and training everyone on basic AI skills," he adds.

A prompt library could include a section for work-related prompts, for example, creating key takeaways or notes from meetings.

4. Include examples

Wani adds that including examples in your prompt provides the AI with guidance on generating the desired content. 

"Including examples in your prompt further assists in guiding the AI's understanding and generating content that meets your requirements. This allows the AI to comprehend your desired outcome more effectively."

5. It's about the journey, not the destination

And finally, Birss highlights the importance of developing skills and being prepared to change your approach.

"Keep developing your skills. There are no experts in this field yet. It's too new. There are only experimenters and explorers. That's what I am. The world of generative AI is changing at a crazy pace. So, you need to keep up with developments. My approach has evolved over the last few months – so I'm starting to work on a new advanced prompt engineering course that builds on my first LinkedIn Learning AI course. Learning how to write prompts is most definitely a journey rather than a destination."

Putting prompts into practice

While each prompt must be slightly different depending on your goal, Birss has developed his "CREATE" framework as a multi-purpose formula to help people create more effective prompts. CREATE stands for the different elements you should consider in your prompt:  

• Character 
• Request 
• Examples 
• Adjustments 
• Type of Output 
• Extras 

Here's what that might look like in ChatGPT using Birss' CREATE framework and some of the abovementioned elements.

You are an expert copywriter who creates emails that engage readers with clear, witty, conversational prose. I would like you to write an email of fewer than 130 words that encourages the reader to sign up for a webinar that's taking place at 10am on the 23rd of March 2024. Include two email subject line options. Describe what the reader will learn at the webinar using this text:

Here's what it came up with: 

While it may require some editing, it's a great starting point. If there are elements that you want ChatGPT to change, there are some extra statements to help shape the results, which Birss covers in his prompt guide.

ChatGPT alternatives

Although ChatGPT is one of the most recognised generative AI tools available, you may be wondering what else to explore now that generative AI has hit the mainstream. 

Wani cites ChatSonic and Jasper AI as two alternatives. ChatSonic is "an AI-driven chatbot and writing assistant that provides assorted templates and utilities to streamline the writing process". 

Jasper AI "emphasises generating SEO-friendly material". Both platforms offer free trials (at the time of writing). 

For Birss, ChatGPT remains a firm favourite, but he also suggests Copilot, Gemini (both free), and Claude (paid) as worth trying. 

"Bing AI – This is basically ChatGPT with access to the live web. The standard version of ChatGPT has a knowledge cutoff of December 2021, and its web browsing feature is temporarily turned off at the time of writing. Bing gives you really good responses that can draw on the latest information from the web. It's built into Microsoft's Edge browser  – but it's about to be released for other browsers, too. 

"Claude by Anthropic – This is a good alternative to ChatGPT. It's really powerful and gives great responses. It also has more focus on ethics. It's definitely worth checking out. 

"Google Bard – I'm still looking for something that Bard can do better than ChatGPT. I haven't discovered it yet, but I'll keep looking. 

Given the nature of generative AI, Birss says users need to stay cautious: "With any publicly accessible AI tool, you need to be careful with the data you enter. Your inputs and responses can be used as training data and – in theory – appear in responses given to other users. Anthropic say they don't do this by default, but you still need to check your data policies. Clients of mine are creating their own chat bots that run on their own servers. That allows them to keep control of their data and the data of their clients."

Final thoughts

In its current form, generative AI should be viewed as an assistant, not a replacement. It can't currently produce content that's fit for publishing straight off the bat; however, for those already skilled in marketing and writing, learning how to use these tools properly clearly has its benefits. 

If it's AI you're looking to master, we have a great range of AI courses waiting for you...