AI tools like ChatGPT and Claude have revolutionised the way we approach communication, whether it be writing an email, a social post, brainstorming an idea, or reading stories to our children at night.
However, writing effective prompts can be the difference between receiving an insightful, entertaining, targeted response, and a vague, unhelpful one.
We’ve been talking a lot about our “go-to prompts” in the Fuller office – and in an effort to share some of them with you, we went straight to the source to find out the top dos and don’ts of interacting with AI, according to AI itself!
Here’s what ChatGPT said…
How to get the best responses from AI
Do, follow the RISEN Framework
- Role: Tell ChatGPT who it should be, including context about your business or audience.
Example: “Act as a content creator for an Australian communication agency.” - Instructions: Clearly describe the task.
Example: “Draft a blog post on AI prompting tips.” - Steps: Break down the task into steps, as though you were training a junior employee.
Example: Instead of “Write a blog”, you should say “Write a blog with an interesting hook, include 3 main points, each with examples, and end with a strong call to action.” - End Goals: Clarify the purpose of the request, communicating the broader strategy of the task.
Example: “The LinkedIn post should summarise the key points from the blog and encourage people to read the full blog on our website” - Narrowing: Add constraints for more precision. Can include aspects such as word count, type of language and tone, topics to emphasise or avoid, formatting preference etc.
Example: “Use simple, easy-to-read language suitable for LinkedIn. Tone should be professional but warm, avoiding overly formal phrasing and jargon.”
Do, encourage reflection
Before submitting your prompt, ask ChatGPT to “think before” and to show its thinking. This not only encourages deeper analysis, avoiding generic responses, but also allows you to sense check the results – is the answer appropriately justified by its reasoning?
Example: “Before suggesting blog topics, I want you to think through what makes a topic engaging for readers and elaborate on your reasoning before you write each header.”
Do, ask for clarification
End prompts with, “before you start, ask me any questions that you need answered to help you generate the best response”. Clarifying questions can help refine the AI’s understanding of your intent, reducing the chances of receiving irrelevant or incomplete answers. This step is particularly important when dealing with complex tasks or nuanced messaging.
Do, iterate
Always ask ChatGPT to make edits or iterations. Consider each AI response as a draft. By iterating, you can fine-tune the output to better align with your needs. You might even discover fresh insights during this refinement process.
Example: “Now, revise the blog introduction to make it more engaging.”
Mistakes to avoid with AI prompting
Don’t upload sensitive data
Avoid sharing personal or proprietary information with ChatGPT, especially in an open AI system, as they do not comply with specific data privacy laws.
Don’t rely on AI for critical decisions
Avoid making business, legal, or financial decisions based solely on AI-generated content.
Don’t assume perfect answers
AI can occasionally provide incorrect or biassed information – so always double check its responses with a reputable source. A great prompt to get around this is to ask ChatGPT to “show its working” and cite its sources, then always double check answers with your own research.
Be aware that ChatGPT is skewed towards Western culture and performs worse with non-English languages. ChatGPT’s dialogue nature can also reinforce a user’s biases over the course of interaction, so it’s also important to be aware of this.
Don’t disregard AI’s carbon footprint
Be mindful of environmental considerations. Each AI interaction has a carbon footprint; managing AI usage efficiently and purposefully can help reduce this impact.
Be aware of updates and evolutions
As AI continues to evolve, so do the ways we interact with it, and the quality of responses we receive from it – so it’s important to be aware of updates as they happen.
OpenAI-01 is the latest version of the language model, which came into effect as recently as mid-September and has been designed to “spend more time thinking” before responding to prompts. According to ChatGPT, OpenAI-01 can “reason through complex tasks and solve harder problems than previous models in science, coding, and math.”
This is the first in a series of new releases expected from the language model, which will make the technology even more intuitive and intelligent… eventually making the need for prescriptive prompting (and articles like this) obsolete.