Transcription
Most people are stuck on the wrong side of using AI without realizing it because AI model change new features every week. So people try everything and eventually get overwhelmed. I have been there before using ai, but never felt like I was truly getting the most out of it. So in this video, I'll share the path to get started with AI and which transform completely how I use it.
Let's. The first part is learning how to communicate with AI and communication all starts with effective AI prompting. After spending two years prompting AI every day, the most critical components for an effective prompts are really these three.
Clear tasks. The whats. What exactly do you want? Is it a proposal, a landing page, a social visual? If you're confused, AI will get confused too.
Relevant context, the why. What are the background details AI needs to understand your situation. Instead of brain dump everything to ai, think about why this context matters.
Output format, the how. How a good output should look like for your task. A table? A word document file? Bullet points? So like this example about a performance review conversation, ChatGPT is still able to give us a decent response, but if we intentionally mention the clear tasks, the relevant context, the output format, you will see the response improved dramatically.
Bonus tip is you can always let AI to ask you question to uncover what are the context that you should give in order for it to do its job. You can also add enhancement prompt elements like persona, examples, constraints for more precision control. But the core three elements I mentioned are all you need for 80% of task.
Besides these core components, here are four more techniques, which are my favorite, and they apply to all sorts of tasks.
Technique number one, few-shot prompting, giving examples. So when I ask Gemini to build a landing page with just a copy, it creates something decent, but looks standard. But what if I give two style screenshot examples this time to output transform dramatically capturing the exact premium aesthetic that I want and stand out.
Technique Number two, perspective shifting, asking AI to shift its perspective and give you multiple angles on the same task, like this marketing landing page copy. We can ask AI to critique the same copy from the perspective of a business owner or from an enterprise executive. And then immediately you can identify weak points and compelling angles to emphasize.
Technique number 3, self evaluation loop. Like this task about a content strategy task. After AI give you a response, ask it to think deeply and create an internal quality rubric and critique the output against the standard. Immediately AI will return back a total of 10 pages, really detailed strategy document instead of a standard draft.
Technique number four, reverse prompting. So give AI your desired output, and let it write the prompt for you. Like I can show Gemini a social visual that I love and ask it to reverse engineer the visual formula, and then I can reuse that prompt to generate new visuals in the same style. So instead of guessing what to ask for, show the results that you want.
Another bonus tip. Use the official prompt optimizer. So different AI models have their own prompting rules. So you can go to the official prompt generator. Just describe what you want in plain language or your initial prompt so it would generate and optimize the prompt that works best with that model. I'll put all the links below.
Now that you can communicate with AI. But here's where most people waste the time. They treat every AI the same. The truth is, yes, you don't need 20 AI tools, but you do need to choose AI models strategically. ChatGPT, Claude, Gemini, anyone of these models already handle 70% of what you need. So master one core model first and that expand when you actually need more specialized capabilities.
From my experience for ChatGPT is like your all-in-one AI assistant with generous usage limit that let you iterate without hitting capacity. It handles a right range of daily tasks from deep research, copywriting, creative idea generation, image generation, and even have a built-in agent mode for autonomous multi-step work. I find it best for high volume iteration and general purpose tasks where you don't need to worry about hitting the limits.
As for gemini, I found it excels when things get big and when you need more visual impact. Its huge token context, window handles, super long reports, hour long recordings. It also has deep integration with Google Workspace and products. Just one click import and export. Its multi-model capability is also superior than other models, analyzing both visuals and audio at the same time. gemini image and video generation models also produce some of the most realistic outputs on the market at speed. So I find it best to do large document analysis, building Google Workspace workflows, and create data visualization when you need both visual beauty and analytical depth.
As for Claude, it captures voice better than any other. AI has a human sounding tone, sentence structure, and vocabulary pattern. The attention to detail is also unmatched for complex instruction. Projects, external app connectors and skill features allow you to build specialized workflow, while its artifacts feature create polished documents and dashboard. So I find it best for strategic thinking with brand voice consistency and complex multistep workflow that needs to connect multiple external system.
And once you master one AI model, you will learn other AI models much faster. So just pick one use for at least 30 days. Build real intuition before expanding.
Now you understand how to craft more effective prompts. But if you're figuring out which AI model to use for what has you stuck, then this is for you. I partnered with HubSpot to create a free resource showing how I actually use different AI tools like ChatGPT, Gemini, Claude, and perplexity. It's called the AI toolkit I use every week. I've put the down link below. This resource break down my favorite use cases for each AI model as a consultant and creator, like why I use Claude for certain tasks, but switch to Gemini for others. You'll see the specific scenarios where each tool shines in my workflow. What I'm most excited to share are the actionable prompts I have included for work scenarios like research or strategy. You can use them as inspiration to develop your own approach. You can grab it for free in the description. And thank you HubSpot for partnering with me on this and for sponsoring this video.
Now, even with great prompts and the right model, there's still a problem. You end up repeating yourself constantly, and this is why. Next. Context management is such an important piece If you want to truly master ai. I used to spend at least 10 minutes re-explaining the project details, the background every time, and then I learned these four context techniques.
The first technique system prompt, also called custom instructions, are like giving AI your persistent personal preference without repeating them. Your role, working style, response preference, how you want response to be formatted. You can set it once and then apply to every chat. Most model have this built-in, so make use of it.
The second technique chat memory. So most modern AI chat bots have this built-in chat memory for context retrieval from past conversations. So you can just turn it on and then you can explicitly ask AI what it should remember about your project and goals, or manually import to its memory. You would be surprised how AI can often find linkage when relevant that makes it more useful. Bonus tip is when your chat is getting long, like this case about a data analysis task in Gemini, ask it to create a TL;DR summary. This is especially necessary for AI models that don't have cross chat memory built-in yet.
The third techniques is the use of project files for ongoing projects. Always set up a focused project. Upload your key documents like brand guidelines, project brief templates that would benefit by reusing. Like this case. I need Claude to generate the client proposal so it can reference them across multiple chats. Just give the task, and AI can automatically retrieve the relevant context without you needing to give the details.
And fourth is to make use of the external connectors. So most AI chat bots now support connecting AI models to external file system like your Google Drive, Gmail, Notion CRM, to retrieve more relevant context outside of AI's own knowledge when completing a task. So this is a huge productivity gain to make sure AI is truly useful for you.
Okay, you have master communication, but here is the problem. AI just confidently make things up and you didn't even notice. I have experienced it all the time. It sounds 100% certain while being completely wrong. So the next part Refine with AI, everyone has access to the same models now. What separate you from the 90%, you verify what is true and you make it yours.
First up, verification. Level one is source anchoring, which is the most critical. So when AI makes a claim, don't just accept it, ask for the direct quote and exact location from external source. Like when I'm doing the research with ChatGPT and it makes some claims about AI marketing usage statistic. So ask to verify from the independent source and it successfully check is from a SurveyMonkey report and I can double confirm myself as well. So you'll be surprised how many of AI's claim come from nowhere and this greatly minimize hallucination.
Level two chain of verification. So don't settle for the first response. Instead ask AI to generate a list of verification question and then answer each one independently. So like this case, I forced Gemini to come up with a list of verification question about its response, and then ask it to answer each of them and generate the final improved response with a much higher accuracy. So this is like the self evaluation prompting technique, but laser focus on catching the factual errors, not just improving qualities.
Level three cross model verification. So let AI critique other AI output. Different models have different training bias. So one model, will often catch artists blind spots. So like here, instead of accepting Claude models response, I gave it to Gemini to critique the response and it did an excellent job spotting issues and improvements, I completely overlook. And sometimes if you find both models disagree, that might be a signal to verify manually.
Okay, you have verified the facts, but here's what I see all the time. People think AI did the work and move on. So this is the next step. Most people skip AI plus human collaboration, not just accurate content, but we need it to be memorable.
Method one, the collaboration process. This is about working with AI from the start to truly create something that is distinctive. The thinking has to coming from you. Add your own thoughts from the beginning, not afterwards. So first, let AI analyze your best writing examples to identify your unique voice and prepare a voice style guide that you can use in the future in any ai. Then anytime you can bring down your ideas to AI through voice dictation or write your wall thoughts. It could be in bullet points, messy format, doesn't matter. The goal is to have unfiltered thinking that is coming from you. And then the magic happens when you ask AI to organize and expand your thoughts, add more depth, find on all these connections while still preserving your writing style. So this will make a huge difference as now the polished copy doesn't feel like everyone else as the perspective is coming from you.
But that is still not enough. So method two, The Distinctiveness Checklist. This is your quality filter before publishing.
First add the YOU Factor, it's not repeating your thoughts, but your specific results experience mistake that you made, your own unique story that others or AI can't mention. This is what make it stand out from the crowd.
And second trim those what I call AI words. So AI overuse a certain phrase, sentence structure like ever-changing unlock, robust. So build a list of how many used AI words or phrase, and then let AI check the content against the list or add it a project context. So AI will flag them automatically and the revised copy will sound less AI generated. So I'll put the AI word list below if you need it.
And third as specific details. Ask AI to find at least three areas. You can add specific details like examples, methods, evidence, like this post, I asked it to propose three areas where I can add concrete details. So this will give you ideas on how to make your copy even better and distinctive.
Fourth, break the pattern. AI always default to a certain writing pattern or formatting because of the training data, so try to mix it up. For example, let it start with a question, not a statement. Turn the bullet list into one to two short sentences or any sentence structure that you personally use.
By now you should know how to communicate with ai and refine the output, but here's what's separate. Good from the top 1%, the ability to scale with ai. Most people use one AI for everything. But we just saw that different models, tools, excel at different things. The secret is AI orchestration, combining tools strategically. So each handles what it does best. And here are three of my favorite workflow patterns to get you started.
Pattern number one, research to polished deliverables. For example, I can just do focus search on perplexity, our specialized AI search engines, and grab a list of useful and quality source link. And then move to NotebookLM the AI note taking app by Google to synthesize and give you grounded insights. Then we can generate different punished content, deliverables like slide decks, infographics, strategy reports, even video overviews, flashcard mind maps. And the best thing is they are all backed by research.
Pattern number two, turn reports or documents into slide decks. For example, we can start with ChatGPT Agent Mode, which is a unique features of ChatGPT. Let it do an in-depth website audit by browsing, taking actions, and prepare a comprehensive audit report. And then let Gemini turn this audit report into beautiful presentations, which is ready to be export to Google slide. So you are leveraging the strength of both AI tools.
Pattern number three, turn reports or documents into beautiful dashboard. So like here we can do Perplexity deep research on certain topics, and then we can load this entire report into Gemini and one, click it into a beautiful dashboard using the built-in visualization creation feature. So within 10 minutes we have transformed a lengthy report of a complex process into a visual story. And this is what system thinking looks like with AI. You are not picking the best tool, the best model. You are building a system where each tool handles what it excels and they work together to compound your outputs.
Right now you can make AI work for you one task at a time, but what happens when you need the same workflows running 10 times a week, or a task that takes five steps across two different tools? That's why the next logical stage, AI automation and agents, basically there are three levels, starting with the easiest one.
Level one, scheduled task and custom AI assistant. This is using automation feature built-in from ChatGPT Gemini Perplexity. Just let it schedule a task like search for the latest news from the past week on specific topics. This can be done within 30 seconds and now you will start receiving email notification. Then set specialized AI assistant a custom GPT, Gemini Gem with predefined workflow details, like a GEM that generates presentation outlines automatically from long documents or reports. Even a simple AI automation like this will save you huge amount of time doing recurring tasks.
Level two simple AI automation. AI automation follows a fixed path to achieve your goal with a trigger. So everything is highly predictable. So when something happen, AI will do a sequence of actions automatically. So most no-code automation tools like Zapier, Make.com will get you started easily with the visual builder. Just define your trigger, like an email arrival, a form submit, and the following step that it should take on what tool. And even Google has its Gemini-powered workflow builder that is equally intuitive. So the technical barrier is lower than you think.
Level three AI agents. This is where AI works mostly on your behalf and decide the best path to make autonomous decision. So you may need an AI agent when your task is more complex and require more intelligent decision making. So popular no-code AI agent building platforms like n8n, Lindy AI, GumpLoop, let you design your AI agents or build agentic workflows with a visual builder. Just design the role for the agent and the autonomous step that it should take. Like a sales lead qualifier agent that is specialized to qualify leads based on incoming email inquiries and conditions. Analyze the details, prioritize with defined criteria, draft emails and send out the slack message.
So my two bonus tip here, first, ask yourself, is this task really necessary? Do not automate work that shouldn't exist. And second, at least do it once manually. If you can't explain the process, you can't verify the automation works correctly.
The real power of using AI is not treating it like a task executor, but using it as a thinking partner to challenge your blind spot and amplify what only you can bring. It requires experimentation and willingness to iterate. To make this actionable for you, I have compiled everything from this video into these AI data guide so you can grab it for free in the description. If you enjoy this video, please also gimme a thumbs up. I also encourage you to join our community. Every month we have specialized workshop and you will also unlock all the prompt resources. I hope to see you there. And before you go, also watch this video about AI prompting to level up your prompt results. I will see you next time.