We are surrounded, if not inundated, by artificial intelligence technologies. Since ChatGPT brought AI into the mainstream, we have seen countless generative AI tools taking shape. Some helped us with amplifying our creativity with generating images that were once limited to our imaginations, some allayed our doubts about almost anything under the sun, and most importantly, these became the ultimate tool in the arsenal of professionals around the world. One thing that has been common to all generative AI tools is the way we communicate with them. Of course, as much as language matters to humans, it is important for AI models to take meaning and intent from our words.
Prompting is simply the way we communicate with an AI model, or how we ask these systems to generate desired outputs. I too have felt overwhelmed in the beginning, but ever since, it has been a journey of learning. Over the weekend I decided to take up Google’s Prompting Essentials course, a nine-hour-long programme to help one understand the most effective ways to communicate with AI tools.
I sat through the course, and I have boiled it down to a quick read with insights, frameworks, and some hands-on tips.
Story continues below this ad
Course structure
Google’s Prompting Essentials is broken down into four modules: Start writing prompts like a pro; Design prompts for everyday work tasks; Use AI for data analysis and presentations; and Use AI as a creative or expert partner. Each of these modules builds on the previous one. They start with the basics and end with advanced prompting techniques that guide you to create smart AI agents.
Prompting like a pro
The highlight of Google’s programme is this simple, yet powerful framework – Task, Context, References, Evaluate and Iterate. This method is to make sure that all your prompts are detailed, effective, and easier for the AI models to follow. One can start by simply defining what exactly they want the AI to do, which is the task. Later, provide the context which could include a bit of background details, specific conditions, etc. Next, offer references in your prompt, such as examples of similar outputs. Once the chatbot generates a response, evaluate it to see if it meets your expectations, and if it does not, iterate. To help you remember this framework, Google suggests the mnemonic “Thoughtfully Create Really Excellent Inputs.” The main idea here is that one should always be iterating. From my experience, coming up with a great prompt is rarely a one-and-done task; it is an ongoing process.
Pro tip: You can elevate your prompts by assigning the AI chatbot a persona. For example, use words such as “act as a film critic” or “fitness expert”. You can also guide the chatbot with a format to get the output as a table, a list, or maybe a social media caption.
While the five-step framework is effective, sometimes it may not produce the output you want. This is where the four key iteration techniques come in. These are revisiting the framework, rephrasing, breaking down prompts, and adding constraints. The last one here is constraints, which is a set of boundaries, such as a specific tone, region, or theme, to narrow the focus of the prompt.
Story continues below this ad
Applying prompting to everyday work
Module 2 of the course focuses on how the five-step framework can be effective in real-world tasks. Be it writing emails, draughting intros for newsletters, or summarising documents, these are some of the everyday tasks. However, instead of spending 10-15 minutes writing a note to your gym members about a schedule change, one can prompt AI to manage it in under a minute. For example, “Write a short, friendly email to staff announcing that the Monday-Wednesday-Friday cardio class has moved from 7 a.m. to 6 a.m.”
For client emails or blog posts, the key here is to be specific about the tone and the audience. Try a prompt with ‘write in a tone like you are explaining this to a curious friend. Providing examples of previous work is a great way to get the AI to work up the tone and style you want.
AI as your spreadsheet and slide deck assistant
In the third module, the course explores how AI can help in more technical tasks such as data analysis and presentations. In case you are not very familiar with spreadsheets, prompting AI to calculate sales trends or customer averages can save time as well as reduce errors. Be careful not to upload sensitive company data. Google emphasises the importance of following company guidelines on privacy and data-sharing policies.
After the data is processed, you can ask the AI to find trends, compare columns, and generate slide outlines. AI can be a handy tool to generate visuals or take points for your presentations.
Story continues below this ad
Prompting with advanced techniques
In the final module, one gets to learn to use advanced prompting techniques such as prompt chaining, chain of thought prompting, and tree of thought prompting. Prompt chaining is using the output as the input for the next. For example, one may start by asking for three one-line summaries of a book manuscript. Then, use those summaries to create a tagline, and finally use the tagline to build a six-week marketing plan.
On the other hand, chain of thought prompting encourages the AI model to walk through its logic in a step-by-step manner. This is ideal for complex problem-solving or troubleshooting. The tree of thought prompting goes wider as it asks the AI to explore multiple reasoning paths in parallel. This is more like brainstorming with multiple experts at once. According to the course, all of these methods combined can generate powerful and nuanced outputs. And, in case you ever get stuck, you can try ‘meta prompting’, a method to use AI to help you write the prompt itself.
Multimodal prompts
Most AI tools nowadays, like ChatGPT and Google Gemini, are multimodal, as they can handle input in the form of text, audio, images, code, and more. If you are looking to create a product post for a nail art collection, you can upload a photograph and prompt the AI with “Write a fun Instagram caption for this nail design photo, highlighting that it’s a new collection.” This applies for audio and video, like you can record a music clip and ask the AI to write a short story inspired by the mood of the sound or ask the AI to brand it with visuals to create a themed presentation. The framework – task, context, references, evaluate, and iterate – is applicable across multiple formats.
The course also shows how to create custom AI agents, meaning virtual experts that can support simulations and interviews or give ongoing feedback. In the course, the experts discuss two types – Agent SIM for roleplay-based learning and Agent X, which is a feedback partner that critically assesses your work. These agents can be created in five key steps: assign a persona, provide rich context, define interaction types, set a stop phrase to end the chat, and ask for takeaways at the end. If done well, these agents can be your AI-powered collaborators who learn and grow with you. Based on the course, I created an AI agent, Comma Chameleon, which instantly helps me proofread copies, suggest changes, and brainstorm ideas for features.
Story continues below this ad
Towards the end of the course, Google offers a Responsible AI checklist. It is very important for all who use AI tools to disclose the same to their clients, co-workers, or workspace. Remember never to share private data without permission, and always double-check the outputs. The goal here is to use AI as a tool, not a crutch.
The course Google Prompt Essentials is paid and is offered in collaboration with Coursera. On completion of the course, users will be able to obtain a certificate that can be shared across their social media platforms and even linked to their LinkedIn account. At a time when AI is entering every domain of work, pursuing the course demonstrates an eagerness to learn and adapt. It is most likely to get you noticed by potential employers.
