Transcript for:
Effective Prompting Techniques for AI

I don't usually overhype myself, but in this video, we're going over the only formula you will ever need to master prompting on ChatGPT and Google Bard. So let's get started. Hey friends, welcome back to the channel. If you're new here, my name is Jeff.

I work full time in tech. And if you're anything like me a couple months ago, you know prompting is an important skill to learn, but you're not exactly sure why some prompts generate outputs that are super generic while others give you precisely what you're looking for. Since then, I've spent hundreds of hours taking prompt engineering courses and applying what I've learned in my daily life.

And so in this video, I'm sharing the six building blocks that make up a good prompt so that you can use this formula to consistently generate high quality outputs. First, it's critical to not only know what the six components are, task, context, exemplars, persona, format, and tone, but also know that there's an order of importance to these six components. To show you what I mean, let's use this simple example. I'm a 70 kg male, give me a three month training program.

The first part is context followed by the task. The reason why the task is higher up in the formula hierarchy is if we just input the task without the context, there's still some sort of meaningful output. But if we just give ChatGPT the context, nothing really happens.

Put another way, it's mandatory to have a task in your prompt. It's important to include relevant context and exemplars. and it's nice to have persona, format, and tone. When you think of writing your prompt, go down this mental checklist. So this formula will act as a constant reminder for you to include just enough relevant information when writing prompts.

And as you'll see in this next part, you do not need all six components in every prompt to have a good output. Now let's break down each building block with specific examples, starting with the task. The rule of thumb is to always start the task sentence with an action verb.

Generate, give, write, analyze, et cetera, and clearly articulate what your end goal is. It could be one simple task, like generating a three month training program, or a complex three step task, like analyzing hundreds of user feedback, sharing the top three takeaways, and categorizing the feedback based on the team responsible for following up. The second component, context, is the trickiest to get right, because technically there's an infinite amount of information you can give.

So, I found asking myself these three questions to be super helpful in coming up with just enough information to get a good result from ChatGPT. First, what's the user's background? Second, what does success look like?

And third, what environment are they in? Back to the workout example we now have. I'm a 70 kg male looking to put on 5 kg of muscle mass over the next 3 months.

I only have time to go to the gym twice a week and for 1 hour each session. Give me a 3 month training program to follow. Could I have added more background information? Of course. Only prioritize the muscle groups that make me look good on Instagram.

But the key to staying productive with ChatGPT and BARD is giving just enough information to constrain the endless possibilities. By the way, although this video is not sponsored, it is supported by those of you who subscribe to my paid productivity newsletter on Google Workspace tips. Link in the description to learn more. Moving over to the exemplars component. It's just a fancy way of saying examples.

Basically all the research on large language models, LLMs, have shown that including examples within the prompt drastically improves the quality of the output. Starting with a simple example, this is a poorly written bullet point from a resume. We can now ask ChachiPT to rewrite this bullet point using this structure.

I accomplish X by the measure Y, that resulted in Z, which is actually best practice, by the way. So actually do this in your resume. For example, I lowered hospital mortality rate by 10% by educating nurses in new protocols, which translates to 200 lives saved per year.

Here's a slightly more complicated example for interview prep. Based on my own resume, write me an answer to the interview question. What's your biggest weakness?

Use the STAR answer framework. Situation, task, action, and results. Here, instead of using a full blown interview answer as an example, which would be overkill, the STAR framework acts as an example structure for ChatGPT to follow. Last example, let's say you need to write a job description. You give some context around the opening and ask ChatGPT to reference an existing job description.

If I use this one I found on LinkedIn, the output will follow the same formatting and use the same professional HRE language, saving me a bunch of time. Main takeaway here. Exemplars are not necessary for every prompt, but including a relevant example or framework will greatly improve the quality of your output. Moving along, the persona component is basically who you want ChatGPT and BARD to be.

And the pro tip here is that think of someone you wish you had instant access to with the task you're facing. If you injured yourself working out, that person might be a physical therapist with experience helping athletes recover. If you're a job seeker, that person might be a recruiter or a hiring manager. If you're working on a creative brief, that person might be a senior product marketing manager who's great at storytelling. Pro tip, you can also name specific individuals, but I found the results to be good only when they're famous enough, like Warren Buffett, Steve Jobs, Jeff Su.

By the way, I just have to share this. We have a team offsite with a superheroes theme. So I asked ChatGPT to draft an email from Batman and it even included things like, please let Alfred know and signed off as the Dark Knight.

So fictional characters work as well and I'm actually gonna use this. The fifth component, format. The pro tip here is to literally close your eyes and visualize how exactly you want the end result to look like. A million likes on my thirst bomb Instagram photo. Damn it, didn't work.

Back to the user feedback example, I don't want to read each sentence, so I asked ChatGPT to take all the feedback and output a table with three headers. The original feedback, the team responsible for following up, and priority. And now I can copy...

this directly and paste it into a Google Sheet. Store by priority and filter by team. Other common formats include emails, bullet points, and code blocks, but the one I found to be the most useful as a full-time working professional is paragraphs and markdown. For example, I just received a lengthy industry report from my director.

First, give me the three key takeaways, then summarize based on topic. Use H2, as section headers. Here is the report. Pro tip, whenever I use ChatGPT to proofread any document, I specify that all changes need to be bolded so I can easily see exactly what has been changed.

Let's quickly go through the last component, tone, before we put all this together in one example. The good news is tone is easy to understand. Use a casual or formal tone of voice.

Give me a witty output, show enthusiasm, sound pessimistic? The bad news is we're usually not very good at recalling the thousands of potential adjectives and adverbs at a moment's notice. So here's a pro tip.

Tell ChatGPT the feeling you're going for. For example, I'm writing an email to a team I haven't worked with before and I want to be taken seriously without coming off as too stuck up and cringy. Can you please give me a list of five tone keywords I can include in a prompt for ChatGPT?

And look, now the actual prompt, I can say use clear and concise language. and write in a friendly yet confident tone. Putting all this together, let's look at this comprehensive prompt.

You are a senior product marketing manager at Apple, Persona, and you have just unveiled the latest Apple product in collaboration with Tesla, the Apple Car, and received 12,000 pre-orders, which is 200% higher than Target. Write an email to your boss, Tim Cookie, sharing this positive news. Task and format. The email should include a TLDR section.

Project background, why this product came into existence. Business results section, quantifiable business metrics. And end with a section thanking the product and engineering teams.

Example structure. Use clear and concise language and write in a confident yet friendly tone. Tone. Note that if I had an existing email to reference, I could delete the instructions around the structure and simply tell ChatGPT the email should follow the exact same format as the one I will share below. And paste.

the email from the fork. By the way, you can compare the output from this prompt to that of a simpler prompt. I just launched a new product, the Apple car.

I received 12,000 pre-orders. Please run email to my boss with this update. There's a pretty big difference in terms of how generic and usable the end result is.

Now that you know the basics of prompting, my next video is gonna take you from beginner to pro, so make sure you're subscribed for that. Check out my top five ChaiGPT productivity tips for work. See you on the next video.

In the meantime, have a great one.