3. Ten tips
This section is organised into ten subsections offering advice under the following titles: Be polite, Be friendly, Be communicative, Be precise, Be specific, Be natural, Be persistent, Be patient, Be positive and Be you. In reality, of course, all these attitudes are connected to one another. For example, being communicative can be part of being friendly and is probably helped by being precise. But in the context of a conversation with an AI assistant, it is useful to think of what each approach can contribute individually, and so this section gives specific examples of good practices driven by these attitudes.
3.1 Be polite
When we work with someone new, one of the first things that can help us talk to each other is politeness, simply because it offers common ground for two strangers to share and build on: we may not know anything about our interlocutor, but we do know that if we are polite to them, they will probably show the same positive attitude to us. Something similar happens in our conversations with AI assistants, who aren’t human but whose responses are determined by both the content and the tone of our prompts, and for whom negative content or tone can be disrupting. In the words generated by one of these assistants: “When users are courteous, it makes the conversation more enjoyable and productive. Being polite can also make it easier to communicate your needs effectively, ensuring I provide the best possible assistance.”
Being polite in a conversation with an AI assistant does not mean saying “please” and “thank you” all the time. But it does mean treating the assistant as a partner in an interaction. Think of it like this: after three or four exchanges in any conversation geared towards producing a text, you might well want the assistant to refer back to an earlier moment, or repeat or change something; or you might want to prepare it to do something else entirely. These things are all much easier to do if the conversation resembles a real interaction between two participants who are seeking to accomplish a task together. The basic advice “Be polite”, then, is saying be consequent with your side of the interaction.
3.2 Be friendly
When the assistant doesn’t produce what you want, rephrase your request rather than pick an argument. Therefore, not
That’s wrong, I didn’t ask for the text as a list.
but
Instead of bullets, now could you write it for me as a running text?
Offer the assistant choices within a task.
What would you like first, the text or the questions?
Don’t overload the assistant with instructions. Give a first instruction and then, after receiving the first answer, fine-tune it with further instructions.
3.3 Be communicative
Ask the assistant to ask you questions.
I need to write a short text about the importance of my colleague’s research into sustainable practices in agriculture in the Baix Ebre, but I’m not sure where to start. Can you help me get ideas by asking me questions about his work?
Assign the assistant a communicative role and purpose.
You’re a member of the administrative staff in a university department, and you have to write a short notice asking the students who have finished class not to make noise in the corridors. The notice that is there now just says “Please be quiet” and that hasn’t worked. You want the words Please be quiet followed by a short sentence to make the students reflect more.
Signpost backwards and forwards in the conversation.
Can you change these three sentences in the same way you changed the sentence that started “All applicants should contact Dr Puig…”?
When you’ve written this list, please call it List B so we can keep it for the next task I want to give you.
3.4 Be precise
First of all, this means being precise and explicit in how you proceed. Come to the conversation with a clear plan of the text you want the AI assistant to produce. Also, be aware that what you ask for (and how you ask for it, that is, the words you choose) and what you don’t ask for will be equally important in the result you get. So the instruction to “analyse the file” that you upload gives much better results than “read the file”, for example, and asking for “exactly five” is more likely to return that number.
Define classes and concepts that may be clear to you but might sound vague to the assistant.
Use British English. That means British spelling. That also means British English for any truncated words. For example, write Mr and Attn without a final point. Please also use British English for the date.
Make the email formal. That means explaining each of these three points in a short paragraph that is concise and avoids contractions.
Make sure each sentence you write leads clearly to the next.
In the second file, I think there are some sentences that contain the words recommend and recommends. Can you find those sentences and write them for me here?
Don’t use jargon. Therefore, not
I want the first paragraph to briefly describe SRH provision for the region’s hill tribes, the second to summarise their general WASH infrastructure and the third to propose four ways in which the UN’s NWoW is accomplishing triple nexus objectives.
but
I want the first paragraph to briefly describe the provision of sexual and reproductive health commodities for the region’s ethnic groups, the second to summarise their general water, sanitation and hygiene infrastructure and the third to propose four ways in which the UN’s drive for countries to jointly commit to collective outcomes are accomplishing humanitarian, development and peacekeeping objectives.
At a basic level, try to write accurately, avoiding abbreviated language, spelling mistakes and bad grammar. These may not always confuse the assistant, but they can have a cumulatively negative effect on both the assistant’s output and your development of the task. They can also make the text ambiguous. Therefore, not
Replace 12.9 x 13.6 + 2 more sentences on expenses (first task)
or
Now include + info cf wboards, blackbirds, overheads, etc. (last text)
or
Add details about the values that provide the extreme conditions of the terrain in January and February.
but
Replace Section 12.9 by Section 13.6 and add two more sentences on expenses using the information from the first task.
and
Now include more information to compare whiteboards, blackboards, overhead projectors and any of the other items that may be relevant from the last text.
and
Add details about the values provided by the extreme conditions of the terrain in January and February.
3.5 Be specific
Make the prompt as specific as possible. For example, instead of just saying that you want the text to be formal, say that it is addressed to the “head of department at a government institution”, whether it should include or avoid contractions, and how many words and paragraphs it should have.
Draft an email from me, a member of the rector’s office at a Catalan university, to a California university representative (Jennifer Lee) I’ve just met at a conference in Venice. Ask her for more information about her university’s mobility programmes and if we could meet up (by videoconference) to discuss them further. The email should be around 200 words. Divide it into three paragraphs: the first introductory paragraph to remind her who I am, the second to ask about the programmes, and the third to propose a date to talk again. Use standard US English and contractions.
————
Subject: Follow-up on mobility programs
Dear Jennifer,
I hope this email finds you well. It was a pleasure meeting you at the conference in Venice last week. I’m [Your Name], a member of the rector’s office at [Your University] in Catalonia. I enjoyed our conversation about international education and the potential for collaboration between our institutions.
I’m particularly interested in learning more about your university’s mobility programs. Could you provide me with more detailed information on the types of exchanges and partnerships you offer? We are keen to explore opportunities that could benefit our students and faculty.
Would you be available for a videoconference to discuss this further? I’m flexible with timing and can adjust to your schedule. How about we aim for a meeting next week? Please let me know your availability, and we can set up a convenient time.
Looking forward to your response.
Best regards,
[Your Full Name]
[Your Position]
[Your University]
[Your Contact Information]
3.6 Be natural
First, when someone says “be natural” they often mean “use common sense”. Indeed, in one of our most natural human activities, we use the commonsensical system of following a train of thought to reach a result, which is more effective than arbitrarily acting on different thoughts. When prompt engineers recommend engaging AI assistants in chain-of-thought procedures, they are basically encouraging us to make the assistant use that natural and very productive procedure. So, point the language assistant towards chain-ofthought procedures to get it to do something that comes naturally to us, but not to it.
Do this as a single instruction, before the task.
Work on this problem step by step.
Or by chaining parts of the task together, at different moments.
Now write those three sentences in just one. And then wait for more instructions.
Label that as “Group 2” and keep it for later.
Or by chaining the whole task, as in this role-play.
You: You’re a member of the admin staff in a university.
[The assistant’s first response, showing readiness and asking for more context]
You: At a Catalan university.
[The assistant’s second response, asking for more context]
You: I’d like you to write to a group of researchers explaining the details of a trip to Tarragona. But please wait now for more instructions.
[The assistant’s third response, agreeing to wait]
Right, I’d like this to be a short email message. First…
Second, “be natural” is also synonymous with “be spontaneous”. Often, even though we have a clear idea about what we want the assistant to produce before we start the conversation, we may ask for changes as the text takes shape because the assistant has suggested something that we like but hadn’t considered ourselves. Taking these suggestions on board can make the whole procedure more creative.
Nice. I really like the addition of the timeline here. In fact, that makes the third group in the last paragraph unnecessary, so can you get rid of that but keep that text coherent?
And because language assistants borrow from such vast data sets of language phrases, in the same way that our natural inclination towards trains of thought can be reproduced in their chains of thought, there are moments when we can choose spontaneity instead and, giving the assistant just one well-written prompt, be rewarded with a fairly complete text ‘just like that’. Spontaneously, you might say.
Write me a 6-line poem that is also an announcement on the wall of a university classroom reminding students to bring their own laptop chargers to class. Make the poem rhyme and make the last line funny.
————
Remember to bring your charger, dear,
Without it, your laptop might disappear.
No power, no notes, oh deary me,
So pack that cord and be worry-free.
Don’t let your battery run out of steam,
Or you’ll be stuck in a tech-less dream!
Third, “be natural” can also mean “be straightforward and direct”. This does bear on conversations with AI assistants because not being straightforward with them can often undermine their output. For example, as an AI assistant’s main remit is to provide the user with a result, testing it to do something you know is impossible will probably backfire, as the assistant will go on doing its best to provide even an inadequate answer.
Here are the observations of one user who discovered this.
“I had read that this particular AI assistant finds it challenging to sound British. The text I gave my assistant (the first text below) didn’t actually have any non-British English elements, but I asked it to make the text more British-sounding anyway, to see if it would change anything. Of course it did change some things just to keep me satisfied (the second text below), but the changes themselves didn’t contribute anything new.”
Hi Gustav,
I hope you’re doing well.
I’m really sorry, but I won’t be able to join you for a coffee tomorrow. I have a meeting, and I can’t move it around.
Maybe next time!
Raoul
————
Hi Gustav,
I hope you’re well.
I’m really sorry, but I won’t be able to join you for a coffee tomorrow. I have a meeting, and I can’t rearrange it.
Maybe next time!
Raoul
Finally, the advice to be straightforward and direct also means that if you realise that you’ve made a mistake in some part of the task development, just own up to it and make it part of the conversation.
Whoops, I think I’ve asked you to do those two parts the wrong way round. Can we go back to the introductory section, and I’ll give you new instructions?
3.7 Be persistent
Don’t necessarily expect the assistant’s first answer to be the best or only answer. Use it for brainstorming. Take its first answer and fine-tune it with more context or a request for specific changes.
Write a warning that the room will be occupied tomorrow.
————
Notice: Room Occupancy
Please be advised that the room will be occupied tomorrow. Kindly make alternative arrangements if you need to use the space.
Thank you for your understanding.
————
Make it friendlier and add the date.
————
Friendly Reminder: Room Occupancy
Just a heads-up that the room will be occupied tomorrow, October 16th. Please plan accordingly and make alternative arrangements if needed.
Thanks for your understanding and cooperation!
————
Make the date format like the one used in the Interuniversity Style Guide for Writing Institutional Texts in English.
————
Friendly Reminder: Room Occupancy
Just a heads-up that the room will be occupied tomorrow, 16 October. Please plan accordingly and make alternative arrangements if needed.
Thanks for your understanding and cooperation!
————
Use something else instead of friendly reminder.
————
Heads-Up: Room Occupancy
Just a heads-up that the room will be occupied tomorrow, 16 October. Please plan accordingly and make alternative arrangements if needed.
Thanks for your understanding and cooperation!
3.8 Be patient
Chain prompts together to slowly build up to what you want. For example, first assign the assistant a role, then provide the context and, finally, give the exact instructions for what you want it to do. Don’t worry about its superfluous responses to each individual prompt.
You’re a college administrator.
————
It sounds like you might be looking for some help with college-related matters. Whether it’s about admissions, course selection, campus facilities, or anything else, feel free to ask! How can I assist you today?
————
At a Catalan university.
————
That sounds interesting! Are you studying or working at a Catalan university? If so, which one? I’d love to hear more about your experience there. 😊
————
You need to write to students to explain the pros and cons of the use of generative AI — wait for more instructions.
————
Got it! I’ll be ready to help you draft that message. Just let me know when you’re ready to provide more details or specific points you’d like to include. 😊
————
I want a short email message with, first, the three most important pros and, then, the three most important cons; the tone should be friendly, but not overly so.
————
Subject: Understanding the Pros and Cons of Generative AI
Dear Students,
I hope this message finds you well. As we continue to explore new technologies, it’s important to understand both the benefits and potential drawbacks. Here are some key points about generative AI:
Pros:
Enhanced Creativity: Generative AI can assist in creating new content, from art and music to writing and design, sparking creativity and innovation.
Efficiency: It can automate repetitive tasks, saving time and allowing you to focus on more complex and meaningful work.
Personalization: AI can tailor educational content to individual learning styles and needs, providing a more personalized learning experience.
Cons:
Quality Control: AI-generated content may sometimes lack accuracy or quality, requiring careful review and editing.
Ethical Concerns: The use of AI raises questions about data privacy, intellectual property, and the potential for misuse.
Dependency: Over-reliance on AI tools might hinder the development of critical thinking and problem-solving skills.
I encourage you to consider these points as you engage with generative AI in your studies and projects. If you have any questions or need further information, feel free to reach out.
Best regards,
[Your Name]
3.9 Be positive
Avoid saying what not to do; instead, say what you want the assistant to accomplish in positive terms. Focus on the desired outcome by using verbs that imply negation but do not include the word not. For example, use avoid rather than do not and exclude rather than do not include. Therefore, not
Describe some of the expenses this research project will involve, but don’t mention the expenses that come under (a), (b) and (c) and don’t talk about overall figures.
but
Describe some of the expenses this research project will involve, but exclude mention of the expenses that come under (a), (b) and (c) and avoid talking about the overall figures.
3.10 Be you
Role-playing can be an excellent way to adapt the assistant’s answers to your needs. Provide roles for both the author and the recipient to set the boundaries for the generative AI precisely.
You work with international students at a Catalan university. You need to warn them that their exams are in two weeks’ time. Tell them they need to prepare and if they have any questions they can contact you.
————
Subject: Reminder: Exams in Two Weeks
Dear students,
I hope this message finds you well. I want to remind you that exams are just two weeks away. This is a key time to focus on your preparation and ensure you are ready for this important part of the semester.
If you have any questions about the exam format, content, or need clarification on any topic, please don’t hesitate to reach out. You can contact me directly via email or visit during office hours.
Wishing you the best of luck with your studies!
Best regards,
[Your Name]
[Your Position]
[Your Contact Information]
If you don’t want the variety of English that your AI assistant generally defaults to, you can try to adapt the role accordingly.
Thanks for that, but please adapt it slightly: you’re from the UK and you like to use a more informal tone when talking to students.
————
Subject: Heads up: Exams in Two Weeks!
Hi everyone,
Just a quick reminder that your exams are only two weeks away. It’s crunch time, so make sure you’re on top of your revision and feeling confident about the material.
If you’ve got any questions—whether it’s about the exam format, what to focus on, or anything else—don’t be shy. Drop me an email or pop by during office hours, and I’ll be happy to help.
Good luck with the prep—you’ve got this!
Cheers,
[Your Name]
[Your Position]
[Your Contact Information]
As you can see, the assistant can take your instructions to an extreme. In this case, the text could be deemed too informal.
Essentially, you are engaging with an interlocutor who processes and uses the writings and utterances of countless other human beings to simulate, mimic and imitate human conversation. You might even consider that it functions as a sounding board, and whatever you put or don’t put into the conversation pretty well determines what will then come out of it.
All these approaches must surely be considered basic good practices in any conversation with an AI assistant.