Creating effective prompts when using artificial intelligence
Generative artificial intelligence, just like other tools used in education and research, can assist in achieving your study outcomes. However, you need to use generative artificial intelligence effectively and critically to get the most out of it. This starts with developing clear prompts that provide the software with enough information to allow it to produce the best response to your request. In addition to this, you’ll need to ensure that you are thinking critically about the response and how you use this effectively in your studies.
Top tips for crafting a strong prompt
View
 | Precision is key The best prompts provide the software with plenty of information. The more details you can provide, the more useful the response will be. |
 | Prompting is part of a cycle Prompting the software is just one part of using generative artificial intelligence. You also need to consider the output, and how you may need to refine or redesign your request based on this. |
 | Topic expertise is important Generative artificial intelligence can help support your learning, but it still makes mistakes. You need to critically analyse the response it gives you to ensure the information is accurate. |
 | You need to think critically You need to think critically about your use of generative artificial intelligence. Consider how you can change your prompt to get more targeted outputs. Also, ask yourself if the response is correct and if it aligns with what you have learnt in your studies. |
Effective AI use
Using generative artificial intelligence involves multiple stages. You should start by developing a detailed prompt. From here, you need to critically analyse the response the program gives you to determine if it is useful, and if the prompt needs to be reviewed. From here, you can consider running the updated request again, and then using the response to support your study.
Craft prompt
The first step in effective use of generative artificial intelligence is to consider how you are crafting the prompt. Just like searching for materials in a database, or using a search engine, the instructions and detail you provide the software determine the quality of the output.
Your prompt needs to be detailed, and precise. The more detail and context that you can provide, the more useful the first output will be. Strong prompts include:
- Topic (be as narrow as possible)
- Specific requirements (e.g. tone, structure and format, number of items provided)
- Points you want covered or examples used (if known)
- Audience or perspective the piece is written from (if required)
(Fitzpatrick, 2023; Mewburn, 2023; Scharth, 2023)
As you start using the software, experiment with different prompts to get an output that you are happy with. Consider using dot points, ample description (important for generating images), and breaking up your requests across multiple prompts so the program can build on its last response. Instruct the program like you would a person - the more information you provide, the better it performs (Mewburn, 2023).
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
Write three exam questions on cities and sustainability.
The answer from the program reads:
- In what ways can cities promote sustainability and reduce their environmental impact? Discuss at least three key strategies and provide specific examples to support your answer.
- How does urban planning play a crucial role in creating sustainable cities? Identify and explain three fundamental principles or concepts that urban planners should consider when designing sustainable urban environments.
- Sustainable transportation is a critical aspect of creating eco-friendly cities. Compare and contrast two different modes of sustainable transportation commonly used in urban areas, highlighting their advantages, disadvantages, and potential impacts on urban sustainability.
The information hotspots read:
- This section of the prompt provides some detail, which is good. In asking for a certain number of exam questions, the prompt controls how many questions the generative artificial intelligence will create.
- It is good to see this prompt identifying a topic. However, this prompt could be more targeted. Including restrictions like level of study (e.g. first year), discipline (e.g. engineering), or type of sustainability (e.g. social sustainability) could help produce more targeted results.
- These questions are a good base for starting revision, however, there are some limitations. The program has assumed that the exam questions need to be short answer style questions. It also has assumed that sustainability means environmental sustainability, and has not asked questions on other pillars such as social or economic sustainability.
Accessible version
This screenshot contains an example of a generative artificial intelligence query.
The question posed reads:
Write three essay questions exploring the intersections of social sustainability and cities. The questions should be at the level of a first year geography student, and be achievable in a two hour exam.
The answer from the program reads:
- Discuss the concept of social sustainability and its significance in urban environments. In what ways can cities promote social sustainability, and what challenges do they face in achieving this goal? Provide specific examples to support your arguments.
- Analyze the impact of urbanization on social inequalities in cities. How does rapid urban growth contribute to the creation or exacerbation of social disparities? Discuss the role of housing, transportation, and access to basic services in shaping social sustainability in urban areas. Provide relevant case studies to illustrate your points.
- Evaluate the role of community engagement in fostering social sustainability in cities. How can citizen participation and grassroots initiatives contribute to creating inclusive and resilient urban communities? Discuss the benefits and challenges of community-led projects in promoting social cohesion, equity, and well-being. Support your arguments with real-world examples from different cities around the world.
The information hotspots read:
- This time the prompt asks for a specific style of question. You could also try asking for short answer questions, or multiple choice.
- This topic is more specific. It is asking about one of the pillars of sustainability (social sustainability), as well as where it intersects with cities.
- Again, this provides more context. This helps ensure that you are not going to be given essay questions that require research as it is unlikely you will be doing research in an exam.
- Adding the level of study helps provide context for the answer. It ensures the program gives you some level of sophistication in the questions it designs.
- Through providing a more detailed prompt, the response is more tailored:
- The software has not assumed that sustainability means environmental sustainability.
- The questions are essay questions, not a collection of short answer questions.
- These are achievable in an exam as you can use examples from the unit, therefore requiring no extra research.
Analyse and refine
Critically reviewing the output from any generative artificial intelligence tool is an important step. You need to consider both the suitability of the content, as well as the accuracy. Remember, generative artificial intelligence does not know what it is saying. Rather, it is programmed to determine what word will most likely be next. As such, generative artificial intelligence does, at times, provide inaccurate information, called a hallucination, and you need to read carefully to be able to identify these.
You need to have a good understanding of the topic area you are asking the software about to ensure that the outputs are correct, and useable. Ask yourself, if this information is accurate, whether the resources exist, and if there are biases in the output? You should also consider whether generative artificial intelligence is the right tool to use. Though it has significant capabilities, this does not mean it is the best option for the task you are asking it to do. Remember that being a university student involves learning how to think and solve problems creatively.
To improve the output, you can consider refining your prompt. Consider trying the following:
- Limiting the output (e.g. in one paragraph, define . . . )
- Refining the scope (e.g. use a location, or discipline)
- Narrowing the topic (e.g. cities, rather than urban areas)
- Targeting a particular audience (e.g. second year undergraduate students)
Also, some software allows you to continue to engage with the prompt in a conversation-like interaction. This helps you use the current response as a foundation to keep working from. You could ask for further examples, ideas, or a different format, for example. You can also use generate again options. These use the same prompt and the current response, to try and develop a different answer.
If the output is not useable, return to part one of the flow chart. Consider how you can reframe what you are looking for to get a useable response. For example, are there different terms you can use?, Is the request better asked across two prompts?
Though generative artificial intelligence is powerful, also consider if it is the best option, or whether using another strategy to get what you need is better. Using a library database, or looking at unit materials might be a better option.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
I need to write an essay on natural disasters for a third-year geography unit. Provide a list of five natural disasters that occurred between 2000 and 2019 that I could use as an example in this essay. Provide no further details apart from the hazard, location and year.
The answer reads:
Certainly! Here's a list of five natural disasters that occurred between 2000 and 2019:
- Hurricane Katrina - Location: United States (Louisiana, Mississippi, Florida) - Year: 2005
- Indian Ocean Earthquake and Tsunami - Location: Indian Ocean (primarily affected Indonesia, Sri Lanka, India and Thailand) - Year: 2004
- Fukushima Nuclear Disaster - Location: Japan - Year: 2011
- Haiti Earthquake - Locations: Haiti - Year: 2010
- Cyclone Nargis - Location: Myanmar (Burma) - Year: 2008
These natural disasters have had significant impacts on the affected regions and offer rich examples for your essay. Remember to delve into the causes, effects and response measures taken in each case.
The information hotspots read:
- This prompt provides a lot of detail for the software. It includes the use (an essay for a third-year unit) as well as the topic (natural disasters). This prompt also includes a series of limiters to control the number of items to include in the list (five natural disasters), a date range (2000 - 2019), as well as asking for no further information than what was asked for.
- Most of the examples here are useful case studies to practise with in preparation for the task.
- This is an example of a hallucination. A nuclear disaster is not an example of a natural disaster. This is why having a strong understanding of a topic, and definitions, is so important.
- This part of the answer was unprompted. As you can see in the next example though, qualifying statements like this are not uncommon. It is a good example of how certain generative artificial intelligence programs act more like chat-bots, than just text generators. It is also a good reminder of the importance of researching the output to ensure that the information is correct.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
List two journal articles in APA 7 style that would be helpful for me to get an understanding of how mobility is researched in human geography literature
The answer reads:
Sure! Here are two journal articles in APA 7 style that would be helpful for understanding how mobility is researched in human geography literature:
- Article 1:
Author(s): Cresswell, T. (2006).
Title: On the Move: Mobility in the Modern Western World.
Journal: International Journal of Urban and Regional Research.
Volume(Issue): 30(4).
Page Range: 923-939.
DOI: 10.xxxx/xxxxxxx - Article 2:
Author(s): Sheller, M., & Urry, J. (2006).
Title: The New Mobilities Paradigm.
Journal: Environment and Planning A: Economy and Space.
Volume(Issue): 38(2).
Page Range: 207-226.
DOI: 10.xxxx/xxxxxxx
Please note that the DOIs provided are placeholders and need to be replaced with the actual DOIs of the articles. You can find the articles by searching their titles or authors on academic databases or by using the DOIs provided.
The information hotspots read:
- This is another example of a strong prompt. The prompt contains a topic (mobility), context (in human geography literature), and required output (two journal articles) in a particular format (APA 7).
- The second hotspot contains an image which is a screenshot of the warning provided by ChatGPT about the inaccuracies that may be present in its responses. It reads “While we have safeguards in place, the system may occasionally generate incorrect or misleading information and produce offensive or biassed content. It is not intended to give advice”
The screenshot is followed by the text:
This article is a hallucination as this article does not exist. However, Tim Creswell has written a book of the same title in 2006. Here, you can see the artificial intelligence has used some truths, but then matched these with the prompt’s request to provide a journal article. This is a good example of why we need to critically engage with the response. This author is well established in this field, and therefore this article looks like a truthful answer. - Overall, the software has done a good job of being able to identify highly relevant researchers to mention here. This is because the programming associates certain words with each other. In his case, this is authors and topics. Further, older materials are more likely to appear in responses such as these. This is because they have the chance to have been used more heavily in academic discussions, and therefore, may be more prominent in the collection of texts the software was trained on.
- Unlike the first reference, this journal article does exist. All the details here are correct.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
Where does the research take place?
The answer reads:
David Bissell's research for "Transit Life" encompasses multiple locations, as he aims to capture diverse commuting experiences and perspectives. While the specific locations may vary, the research is likely to have taken place in various urban and suburban areas where commuting and transit are prevalent. Bissell's interviews and observations may have been conducted in cities with well-developed transportation systems, such as London, New York, Tokyo, or Sydney, among others. The intention is to encompass a range of settings and contexts to provide a comprehensive understanding of transit experiences in different cultural and geographical contexts. The specific locations visited and studied by Bissell are not explicitly mentioned in the available information.
The information hotspots read:
- This prompt was part of a conversation with a generative artificial intelligence program. The first prompt asked for a book summary which generated a response that did not mention where the research for the book was completed. This prompt was asked as a follow up.
- The software here is very confident in suggesting that it encompasses multiple locations. But later in the response, this confidence diminishes.
- Here, the response starts to be more cautious. It has transitioned from “multiple locations”, to its “likely”.
- Here is a piece of truth that helps make the response sound more accurate. This book drew on substantive qualitative research to develop its insights. These truths make the response seem accurate, but you need to read carefully to spot errors.
- The program has inadvertently got the answer correct here - this research took place in Sydney. However, this is not because it knew the answer, but rather it knew of an association between major cities and transport networks with high levels of commuters.
- This is a strong example of how these types of programs are not search engines, but rather software designed to give you an answer to your question, regardless of its accuracy. The research for this book took place in Sydney. The location is something mentioned in the blurb and readily available on the internet. This knowledge would have most likely been part of the text this program was trained on.
Use output
Now that you have an output that you are happy with, you can keep working on your task. Generative artificial intelligence can be helpful for:
- Brainstorming ideas about a topic
- Creating exam practice questions
- Summarising articles or topics for you to critique
You can also use the responses as a foundation to build your unit knowledge.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
I am preparing for an exam. Please provide definitions of the four types of tectonic plate boundaries. Do not tell me which definition is for which type of boundary.
The answer reads:
Sure! Here are the definitions of the four types of tectonic plate boundaries without indicating which definition corresponds to each boundary:
- This type of plate boundary occurs when two tectonic plates move away from each other, resulting in the creation of new crust. Magma rises from the mantle, fills the gap, and solidifies, forming new oceanic lithosphere.
- This type of plate boundary is characterized by the collision of two tectonic plates. When they converge, one plate is forced beneath the other in a process known as subduction. This boundary often leads to the formation of volcanic arcs and mountain ranges.
- This type of plate boundary involves two tectonic plates sliding past each other horizontally. The movement can be smooth or abrupt, causing earthquakes along the boundary. This type of boundary does not involve the creation or destruction of lithosphere.
- This type of plate boundary occurs when two tectonic plates move toward each other and collide. Neither plate is subducted, but rather they crumple and buckle, forming mountains. This type of boundary is associated with intense seismic activity and the formation of mountain ranges.
Remember, it's important to understand the characteristics and processes associated with each type of boundary in order to identify and differentiate them correctly during your exam. Good luck with your preparation!
The information hotspots read:
- Asking questions like this can be useful to test your knowledge of certain subjects. This type of prompt is helpful when preparing for an exam.
- You can also ask the program to give you the answers, and check to see if it is correct based on your understanding of the topic.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
Can you please provide another five examples in the same format?
The answer reads:
Certainly! Here are five more examples of natural disasters that occurred between 2000 and 2019:
- Great East Japan Earthquake - Japan (2011)
- Cyclone Nargis - Myanmar (2008)
- Wildfires in California - United States (2018)
- Tōhoku Earthquake and Tsunami - Japan (2011)
- Hurricane Sandy - United States and Caribbean (2012)
These additional examples should give you a broader range of natural disasters to analyze in your essay. Remember to conduct thorough research on each event to gather more information and insights for your essay.
The information hotspots read:
- In asking the question again, you get another set of examples that you can work with to revise, or practise essay responses with.
- However, there are still hallucinations in this response. The Great East Japan Earthquake and the Tōhoku Earthquake and Tsunami are the same event.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
Write three exam questions on cities and sustainability
The answer reads:
- In what ways can cities promote sustainability and reduce their environmental impact? Discuss at least three key strategies and provide specific examples to support your answer.
- How does urban planning play a crucial role in creating sustainable cities? Identify and explain three fundamental principles or concepts that urban planners should consider when designing sustainable urban environments.
- Sustainable transportation is a critical aspect of creating eco-friendly cities. Compare and contrast two different modes of sustainable transportation commonly used in urban areas, highlighting their advantages, disadvantages, and potential impacts on urban sustainability.
The information hotspots read:
- Though this prompt can be improved, the response here is still useful. Ask yourself how might output still be useful for me?
- Instead of coming up with examples, you could answer these questions using case studies from your unit. This means you can still use this response.
Accessible version
The screenshot above contains an example of a generative artificial intelligence query.
The question posed reads:
Can you write three short answer questions on the same topic?
The answer reads:
- What is the concept of a "green building" in the context of sustainable cities, and how does it contribute to environmental sustainability?
- Explain the concept of "urban heat island effect" and its implications for urban sustainability. Provide two strategies that cities can employ to mitigate this effect.
- What is the importance of community engagement in achieving sustainable urban development? Provide two examples of community-based initiatives that have successfully contributed to enhancing sustainability in cities.
The information hotspots read:
- Though the first prompt in this conversation was not fantastic, if the topics the generative artificial intelligence managed to identify are still useful, you can keep refining your prompt.
- Here, using the same themes, the program it has been asked to create the output in a different format.
NoteIf you are using generative artificial intelligence in your assessments, make certain that it is permitted for use, and allowed to be used in the way that you are using it. You will also need to acknowledge its use. |
References
Bell, G., Burgess, J., Thomas, J., & Sadiq, S. (2023). Rapid response information report: Generative AI - language models (LLMs) and multimodal foundation models (MFMs). Australian Council of Learned Academies. https://www.chiefscientist.gov.au/sites/default/files/2023-06/Rapid%20Response%20Information%20Report%20-%20Generative%20AI%20v1_1.pdf
Fitzpatrick, D. (2023, February 5). AI won’t replace teachers, but it will replace teachers who don’t use AI. The AI Educator Sunday Newsletter. https://preview.mailerlite.io/preview/282063/emails/79169655273424140
Mewburn, I. (2023, May 2). Using ChatGPT (ChattieG) to write good. The Thesis Whisperer. https://thesiswhisperer.com/2023/05/02/usingchatgpt/
OpenAI. (2023). ChatCPT (May 24 Version) [Large language model]. https://chat.openai.com/chat
Scharth, M. (2023, February 23). How to perfect your prompt writing for ChatGPT, Midjourney and other AI generators. The Conversation. http://theconversation.com/how-to-perfect-your-prompt-writing-for-chatgpt-midjourney-and-other-ai-generators-198776
Taking it further
-
Using artificial intelligence
This resource is designed to help students understand when it is appropriate and ethical and when it is not appropriate or ethical to use artificial intelligence technologies in their university studies.
-
What is critical thinking?
Learn what critical thinking is, what it looks like and how you can demonstrate it.
-
Prepare for timed assessments
Learn how to make your study for all types of timed assessments more efficient. Discover strategies for open-book and closed-book exams, and what to do before, during and after a timed assessment.
-
Understand assessments
Improve your grades with this key resource, which will help you understand how to approach different assessments and analyse your assessment tasks.