Prompting is the technical name given to asking a question of an AI chatbot. You might know the expression "ask a stupid question - get a stupid answer" and the computer science acronym GIGO - garbage in, garbage out. Both are relevant to prompting. There's an old trope - you can find examples on the Internet - of adults following children's instructions literally - here's a dad following his kids' instructions to make a peanut butter and jelly sandwich. Prompting generative AI can be a bit like that, with you as the child (sorry!) and the AI as the adult (except unlike in the video, the chatbot doesn't know any better). You need to be clear and exact in your instructions. AI chatbots are pretty stupid; they'll do what you tell them to do, so you must be quite specific about what you want. Here are some tips on improving your prompting.  

Prompting Tips

Define Your Objectives Clearly

Decide what you want from the chatbot before you start prompting. 

Be Clear and Specific

Just like when you ask someone a question, the more specific you are, the better the answer will be.
Example:

• Less Effective: "I need information about World War II "
• More Effective: "Can you tell me the causes of World War II with reference to the Japanese need for economic independence?"

Ask One Question at a Time

Chatbots are pretty stupid and asking more than one question at once will reduce the quality of the results they provide. If you have multiple questions, break them down into separate prompts. Humans can handle the multiple questions below, chatbots not so much.
Example Queries:

•  Multiple Question: "What are some healthy dinner recipes?  Can you suggest a dessert? Is it low calorie?"
•  Single Questions: "What are some healthy dinner recipes? "Can you suggest a low-calorie dessert?"

Stay on Topic

If you want to change the subject, start a new chat. Chatbots get confused easily and might be influenced by your last messages, using the same tone, structure and context.   

Provide Context

If your question relates to a specific situation, providing context will help the chatbot output a more relevant response
Example Queries:

• Without Context: "I need help with my social science essay."
• With Context: "I need help finding sources for my essay on the impact of social media on youth for a Level 5 college media literacy course"

Use Complete Sentences

While chatbots, like people, can respond to short phrases, using complete sentences can improve clarity and ensure your request is processed correctly (just like with people!)
Example Queries:

• Short Phrase: "Best study apps."
• Complete Sentence: "What are the best study apps for college students available for Android phones?"

Give Examples

Also called "few-shot prompting", if you provide a few examples of a task, you're showing the chatbot what you want in terms of output, tone, style and structure. 

Examples: 

• "Here are some effective study tips for learners:
  •  Create a study schedule to manage your time effectively.
  •  Use active learning techniques, such as summarising information in your own words.

Now, can you provide additional study tips?"

• "Here is a 200-word summary of World War 1 [summary]. In the same style, write a 200-word summary of World War 2" 

Be Patient

If a chatbot can't provide a suitable answer, try rephrasing the question or providing additional details. Experiment with different prompt styles. 
Example Queries:

• Initial Prompt: "I need information on nutrition."
• Rephrased Prompt: "Can you provide information on the benefits of a balanced diet?"

Explore Follow-Up Questions

Look at the answer the chatbot gives you. There might be some scope to ask follow-up questions for more information or clarification.
Example Queries:

• Follow-Up: "Can you give me more details about the health benefits of eating more fruits and vegetables?"

Think about Personas

Suggest a persona (a role) that the AI should embody, to guide the style, tone, and focus of responses.
Example personas:

• maths teacher
• salesman

Example query:

"Explain Pythagoras' theorem in the style of a maths teacher." 

Some people say that suggesting a persona might not make any difference to the accuracy of the result, but you can certainly use a persona to change the style and tone of a chatbot's answer.

Take It Slowly

If your question is complex, break it down into step-by-step instructions.

Example query:

"Summarise this article (step 1), list the key takeaways as bullet points (step 2) and then turn the summary and key takeaways into a short blog post (step 3)"

Think About Format and Output

Chatbots can create their responses in a variety of different outputs, such as (but not limited to!):

• plain text
• report
• table 
• presentation
• bullet points
• flowchart

Think about the best format for the answer to your question.

Example query: 

"Give me 10 tips for better sleep, formatted as a list". 

Don't Enter Private Data

AI tools remember queries and other inputs and use these to help with training, so avoid including anything you wouldn't want to share with the world - private, sensitive, or confidential information - in your prompts.

Be Aware That Your Conversations Are Stored 

Chatbots usually let you see a history of your conversations. You can go back to a previous conversation and continue it. Chatbots will also let you export and delete your history and turn off the saving of future conversations. 

Don't Use Chatbots as Search Engines

In general, AI chatbots don't search the web (though this is starting to change) - they primarily rely on their training data to generate responses. However, some advanced chatbots can be designed to search the web, but chances are you're not using an advanced chatbot. While chatbots are trained on huge amounts of data, they miss vast swathes of the web. Also, chatbots hallucinate, so the results a chatbot returns may not all be real. Don't ask for sources, either. Chances are some will be made up.

Know When to Seek Human Help

This is perhaps the most important thing to know about prompting and using chatbots. Generative AI chatbots can be useful, no doubt, but there will be times when you will need more personalised  - and knowledgeable - assistance. If your question is complex or requires detailed help, a human will be able to provide a better answer than a chatbot.

Frameworks

There are several different prompting frameworks. A prompting framework is a set of guidelines that provide instructions and context to the chatbot to ensure that its output is relevant and accurate. Many frameworks are unnecessarily complicated or are designed for advanced chatbot users. The CLEAR framework is worth looking at, however.

The CLEAR Framework

The CLEAR framework is a structured approach designed to help users build effective chatbot queries. CLEAR stands for: 

• Context: Provide relevant background information or context for your question. This helps the chatbot construct answers relevant to the specific situation or topic you are inquiring about.
• Language: Use clear and concise language. Avoid jargon or overly complex phrases to ensure that the chatbot can easily interpret your request.
• Expectations: Clearly state what you expect from the chatbot. Specify the type of information or assistance you are looking for, whether it’s a direct answer, a list of resources, or step-by-step guidance.
• Action: Indicate any specific actions you want the chatbot to take. This could include searching for information, providing examples, or guiding you through a process.
• Refinement: Be prepared to refine your question based on the chatbot’s responses. If the initial answer isn’t satisfactory, rephrase or add more details to clarify your request.

A Gentle Reminder

Chatbots are stupid and are prone to making stuff up. It's good practice to verify anything a chatbot tells you. 

Image source: Third Space Librarian AI Prompt Engineering: It’s the new literacy