If you’re like me, you’ll have started using ChatGPT or other AI chatbots the same way as you use search engines – posting simple queries but with more text. Sometimes this gives you what you need really fast, sometimes it doesn’t. 

But the beauty of AI bots is that you can programme them to focus on your specific needs without knowing any code. This makes the bots super-useful and all you need is a bit of structure for your prompts to provide more context. 

This is an introduction to writing more effective prompts. I’m not pretending I’m an expert here, just sharing some of the things that have brought the bots alive for me and saved me hours of time. (You may also find 8 Ways to Use ChatGPT to Supercharge Your Writing useful.)

One discovery I’ve made is the Chatbot Paradox – Large Language Models like ChatGPT are just machines that are very, very good at spotting patterns. They can’t reason like us, but and yet they work best if you treat them as though they are human. 

1. Create an AI Sidekick

Generic prompts generate generic answers. The Large Language Models behind AI chatbots cover such a huge amount of information that when given vague instructions, they deliver bland results. 

That’s precisely what you get in human conversations. To get more relevant answers you need to provide more context. And the first step to providing this is to devise a persona. 

The best way to think about this is to ask yourself, “if I had access to someone who could help me with my problem, what sort of person would that be? Then simply add that persona to your prompt: 

• Knowledge and expertise – with a detailed knowledge of digital engagement tactics
• Experience – working with internal communications teams on change programmes
• Goals, desires – an interest in data, practical examples, and succinct explanations
• Context: working with purpose-driven organisations with limited budgets and not much time

“Act like [a world-class digital media analyst]“

The golden rule of effective prompts is to add as much context as you can. So consider creating a more detailed persona by adding the following characteristics: 

Specific tasks require specific personas. The others I use most frequently are:

“A sub-editor at Buzzfeed who is expert at taking nerdy topics and making them appealing to a broad audience.”

“A Search Engine Optimisation expert who focuses on Google’s framework of Experience, Expertise, Authority, and Trust.”

“A social media manager with 10 years experience, obsessed with engagement data, and an awareness of the importance of visual content. “

See How to Create ChatGPT personas for every occasion for more examples.

2. Be the Boss: Tell the Bot Precisely What You Want, But do it Politely

You can increase the usefulness of your persona by specifying what sort of responses you want. The trick here is to imagine that you have a new recruit to your team. How would you break down tasks into simple instructions with just enough context so you don’t overwhelm the newcomer but guarantee they don’t waste their time.

In my work, I’m sceptical about much of what is written about digital media, and am always on the lookout for data I can verify and practical examples I can use as illustrations. I want my sidekick to act like I do and anticipate these needs. So this is my behavioural instruction:  

“When I ask a question, provide a sceptical response, always provide a source for data, and give me practical examples where possible.” 

Something about chatbots can turn the nicest people into dictatorial bosses with incessant demands, no manners, and minimal feedback.  Odd though it may sound, they respond better to being treated politely, give better answers when you make it clear the problem is important or urgent, and can be motivated by financial rewards. 

See Onboarding your AI Intern for more on the kind of tasks and instructions that will increase your productivity.

3. Turn the Tables: Get AI to Ask the Questions

If you don’t know quite know what to ask, then ask the chatbot to guide you: 

“From now on, whenever I ask a question, suggest a better one and ask me if I would prefer to use it.”

One feature of AI chatbots is that they will remember such instructions and you don’t have to repeat them for every task you have.

A more sophisticated way of refining your questions is known as the Cognitive Verifier Pattern:  

“From now on, whenever I ask you a question, follow these rules:

“Generate a number of additional questions that would answer the original question more accurately.

“Combine the answers to individual questions to create the answer to the original question.”

One of the concerns about the chatbots is the lack of transparency over how they come up with answers and this approach helps get around that. 

Completely reverse the roles with the Flipped Interaction Model. Instead of thinking about what instructions you need to provide to achieve your objective, flip things 180 degrees and start with your objective and ask the chatbot to come up with the questions for you to answer.

“I would like you to ask me questions to achieve [objective].

“You should ask questions until [you have enough information to achieve the objective]. “

Possible refinement: “Ask me one question at  a time.”

4. Turn the Spotlight On Your Audience

There’s a difference between using AI chatbots for research and using them to draft output. Extreme care needs to be taken with the latter – the bots can hallucinate (make stuff up) and you’ve probably already spotted the idiosyncrasies of AI-generated text. 

The chatbots are nevertheless very helpful in drafting output for a defined audience. And the secret to good prompting here is to have as detailed a view of your audience as possible. 

You may already have a highly developed model of your typical audience member. In which case you can simply add it to your initial prompt. 

Tailor your answers for [persona].

If you don’t have ready-made personas then use one of the options under tactic 3 to create one: 

“You are a world-class digital marketing expert. We are going to create an audience persona to help us target appropriate content. I want you to ask me questions one by one until you have sufficient information to give me the persona. I want you to start by defining what parameters we should be looking at. When you have done this, proceed to your questions.“

Then use this persona within your content drafting prompt.