Combining AI Tools for Better Results

Summary

AI Explored: Combining AI Tools for Better Results

Podcast Information:

• Title: AI Explored
• Host: Michael Stelzner, Social Media Examiner—AI Marketing
• Episode: Combining AI Tools for Better Results
• Release Date: July 22, 2025
• Description: This podcast is tailored for marketers, creators, and business owners eager to harness the power of AI. Join Michael Stelzner as he shares practical and actionable strategies to effectively implement AI in your endeavors.

---

1. Introduction

Michael Stelzner opens the episode by emphasizing the evolving landscape of AI in marketing and invites listeners to explore beyond familiar tools like ChatGPT. The focus of this episode is on leveraging a combination of AI platforms to achieve superior results.

2. Guest Introduction and Background

[02:23] Michael Stelzner:
"Grace is a digital growth consultant who helps marketers grow organically with AI strategies."

Grace Leung shares her journey into the AI realm, starting her career in digital marketing in Asia over a decade ago. Her pivotal moment came in late 2022 when she first interacted with ChatGPT at her SaaS company in Canada. She recalls:

[03:01] Grace Leung:
"ChatGPT was a game changer. It made AI accessible to non-technical marketers like me, transforming how we conduct research, prepare copies, and brainstorm ideas."

Transitioning from corporate to independent consulting, Grace has expanded her expertise, embracing a variety of AI tools to enhance her business and client services.

3. Expanding Beyond ChatGPT: Importance and Benefits

Michael addresses the common reliance on ChatGPT among marketers and entrepreneurs, prompting Grace to illustrate the necessity of diversifying AI tool usage.

[06:15] Grace Leung:
"Different AI models have unique strengths. For instance, ChatGPT excels in language-based tasks, while Google’s Gemini is multimodal, handling text, images, audio, and video seamlessly."

By integrating multiple AI tools, users can tap into diverse functionalities, ensuring they do not miss out on opportunities that a single tool might overlook.

4. Understanding Different AI Models and Their Strengths

Grace elaborates on the distinctions between various AI models:

[06:35] Grace Leung:
"Gemini has been trained multimodally from the start, enabling it to handle diverse input types. In contrast, ChatGPT, primarily text-based, offers logical and structured responses."

She highlights that newer models like Gemini offer broader capabilities, such as processing multimedia inputs, which can provide a more comprehensive understanding and analysis compared to text-only models.

5. Structured Framework for Testing AI Tools

To effectively utilize multiple AI tools, Grace recommends a strategic approach:

[08:29] Grace Leung:
"Develop a structured testing framework. Identify recurring tasks, design prompts, and test these prompts across different models to evaluate speed, accuracy, bias, and tone."

Michael reiterates the importance of this framework, sharing his personal experience of using both Claude and ChatGPT to enhance survey question quality:

[10:18] Michael Stelzner:
"Using both Claude and ChatGPT allowed me to uncover unique insights from each, resulting in a more robust survey than using a single tool."

6. Practical Use Cases: Analyzing Large Data Sets

Grace delves into one of her favorite use cases: analyzing large data sets with AI.

[13:35] Grace Leung:
"AI excels at identifying patterns and structuring unstructured data. For example, you can input YouTube comments or business reviews into Gemini to structure the data into tables for further analysis."

She emphasizes the transformative impact of AI in handling vast amounts of information efficiently, enabling deeper insights that would be time-consuming manually.

7. Deep Dive into Specific Tools

Google Gemini vs. Claude

[14:18] Grace Leung:
"In audience research, I use Gemini for initial data analysis and then pass the insights to Claude for strategic refinement. Gemini handles the heavy data intake, while Claude excels in presenting the data story."

Michael adds:

[18:00] Michael Stelzner:
"Gemini Pro is ideal for data-heavy tasks, whereas Claude is better suited for strategic and detailed work."

Perplexity for Quick Research

Grace outlines the use of Perplexity for swift information retrieval:

[28:55] Grace Leung:
"For quick searches, like identifying top competitors, I use Perplexity. It provides immediate, concise answers that I can then expand upon using other tools."

She contrasts this with more in-depth research requirements, where tools like Gemini and NotebookLM come into play.

NotebookLM for Internal Research

[37:43] Michael Stelzner:
"Can you elaborate on NotebookLM and its role in internal research?"

[38:01] Grace Leung:
"NotebookLM, powered by Gemini, allows bulk imports of sources, minimizing hallucination while ensuring responses are based strictly on the imported data. It's a powerful tool for reverse engineering Google’s ranking strategies for SEO."

8. Research Applications and Workflow Integration

Grace shares a comprehensive workflow for conducting audience and market research:

1. Initial Data Collection: Use Perplexity for quick competitor identification.
2. Data Structuring: Utilize Gemini to analyze and summarize large data sets.
3. Strategic Refinement: Pass the structured data to Claude for persona development and strategic insights.
4. Content Planning: Employ NotebookLM to integrate Google-ranked sources, ensuring content aligns with high-quality standards.

[32:36] Michael Stelzner:
"Using ChatGPT to develop a research plan and then executing it with Perplexity enhances the quality and structure of the research output."

9. Pricing and Tool Selection Considerations

Grace discusses the cost-effectiveness and features of various AI tools:

[34:05] Grace Leung:
"The paid version of Perplexity costs around $20 per month, offering unlimited quick and deep searches compared to the free version’s limited daily queries."

She advises selecting tools based on specific needs, such as using Claude Sonnet over Opus for better token management during data-heavy tasks.

10. Closing Remarks

Michael and Grace wrap up the discussion by summarizing the benefits of a multi-tool AI approach. They emphasize the enhanced insights and efficiency gained by integrating tools like Gemini, Claude, Perplexity, and NotebookLM into marketing and research workflows.

[41:33] Michael Stelzner:
"Grace, thank you for sharing your invaluable insights. This multi-tool approach undoubtedly equips marketers and entrepreneurs to stay ahead in the AI-driven landscape."

Grace provides her contact information for listeners interested in further collaboration or learning more about her AI strategies.

---

Key Takeaways:

• Diversify AI Tool Usage: Relying solely on ChatGPT limits potential insights. Combining multiple AI models like Gemini and Claude can unlock a broader range of functionalities and deeper analyses.
• Structured Testing Framework: Establish a systematic approach to evaluate and integrate different AI tools based on specific tasks, ensuring optimal performance and minimal bias.
• Practical Workflows: Implement workflows that leverage the unique strengths of each AI tool for tasks such as data analysis, persona development, and content planning.
• Cost vs. Features: Consider the pricing and specific features of AI tools to ensure they align with your business needs and budget.

By embracing a multi-tool AI strategy, marketers and business owners can achieve more comprehensive, accurate, and strategic outcomes, positioning themselves as AI-savvy leaders in their respective fields.

Transcript

[Michael Stelzner]

Hey there, it's Michael Stelzner. As a loyal podcast listener, you've been with us through major marketing shifts. That's why we want to make sure we're covering what matters most to you when it comes to AI. Your answers to our brand new AI survey will influence the AI topics we feature in future podcast episodes. So would you do me a favor and pause this podcast right now and visit social media examiner.com AI survey? The survey closes in just a few days. Thanks so much for helping us create AI content that serves you best. Now onto the show.

[Grace Leung]

Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner.

[Michael Stelzner]

Hello, Hello, Hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzer, and this is the podcast for marketers, creators, and business owners who want to know how to put AI to work. Are you mostly a ChatGPT user? Are you thinking there's got to be better tools or maybe there's a better way for me to do my work if I was to employ different AI tools creatively to accomplish very specific use cases? Well, in today's podcast interview, we're going to explore some really fascinating use cases that I bet you have not explored. So if you've always wanted to kind of take it to the next level, today's podcast is for you. And I'm going to be joined by Grace Learn. By the way, if you're new to this podcast, be sure to follow this show so you don't miss any of our future content. Let's now transition over to my interview today with Grace.

[Grace Leung]

Helping you simplify your AI journey. Here is this week's expert guide.

[Michael Stelzner]

Today, I'm very excited, excited to be joined by Grace Learn spelled L e u n g and Grace is a digital growth consultant who helps marketers grow organically with AI strategies. Her community and newsletter is for marketers and entrepreneurs who want to strategically grow their businesses with AI. And you can find her on YouTube at. Grace L E U N G Y L Grace, welcome to the show. How you doing today?

[Grace Leung]

I'm good. Thanks for having me, Michael.

[Michael Stelzner]

I'm super excited that you're here today. Grace and I are going to explore how to combine different AI plots, platforms and tools to get better results. Now, before we go there, I want to hear your story. How did you get into AI? Start wherever you want to start.

[Grace Leung]

So I first started my career in this field back when I was in Asia. I've been in the digital marketing field for over 10 years until three years ago I moved to Canada with my family and then since I moved here I joined a SaaS company and then leading the growth marketing programs for the new product. And that time I still remember, you know, 2022, late 2022 when ChatGPT first came out. And then I still remember a day I my colleague shared chatgpt through Slack. Okay, try this, please try this. It's so amazing and okay, let's try it. Oh my mind was blown because it's so powerful, right? Because you can just ask anything to ChatGPT and then it will give you response. So that time I already see, oh there's a huge potential how this thing we change how we do marketing. And so I started experimenting more and more. Maybe like drafting social media poses like creating ideas, brainstorming ideas or even maybe comes up some Persona details for our campaign copies. And then I just tried more and more. So I think this is a real game changer and I think is AI has been in the industry, in the marketing industry for a long time, right? Like Google, Facebook has been using AI to do ad targeting, recommending the ads. But however I think ChatGPT was so powerful is that it's become so accessible, it makes AI so accessible. It's not just about certain group of people, not just data scientists but also non technicals like me as marketers to everyone. And so I think although now looking back it may feel very basic but that time imagine as marketers we need to do research ourselves, we need to prepare the copies, ideas. So I think that the transformation impact was so huge.

[Michael Stelzner]

So bring us up to the present with your story. So what are you doing now? You're still working for that company or have you gone out on your own? What are you doing today?

[Grace Leung]

Yeah, so to continue that story, I think at the same time that time I started my YouTube channel because I think okay, I just started as a passion project. I just think okay maybe in this new chapter of my life I to try something new. So I just start to share whatever I know about digital marketing. And then I also start using AI because I found this is super powerful. And that time more AI tools came out. Like I still remember those days, maybe content creation tools like copy AI, Jasper Writesonic and then I started doing more reviews and then Perplexity came out and then German High and then Claw. So you just go down to the rabbit hole, what's next, what's next? And then you keep sharing on YouTube and then until perfect. From the corporate to now, as an independent growth consultant, I use AI more and more for my own business and also for my clients. Because I think two years from now has changed a lot. It's not just, no longer just content creation. You can do a lot with AI strategy, charts, visualization, image generation, video generation. So the impact is huge.

[Michael Stelzner]

Awesome. Well, thank you Grace for sharing that story. So many people listening to this podcast, whether they be marketers or entrepreneurs or creators, are predominantly focused on just ChatGPT. Right? Because maybe that was the first tool they started using or it's the one that everybody uses. But why is it important for people to expand their horizons beyond just ChatGPT?

[Grace Leung]

Yes, that's a good question. I think it's very natural. Most people start with ChatGPT like myself and then like Steph taught and then try to experiment with ChatGPT. But however, as time goes on, as AI become more advanced, you can see that the more AI models coming out, like for example Gemini, maybe the first version bark is wasn't super impressive. But however, like Gemini has been catching up. And then I would say different AI models have their different uniqueness and strength because the difference they are trained so differently. So for example ChatGPT is a test based model. But however Gemini has been trained at multimodally since the very beginning. So that's why Gemini can handle tags, images, audio, videos, all together at the same time as input and it's not a problem. Or maybe cloud is trained more focused on linguistic and also having more consideration in human tone, ethical consideration. So you can see different models will start to develop their own capabilities. I'm not saying that as marketers you need to use different models, but however, you need to know that the strain so that you are not missing any opportunities. Maybe certain AI models is better fit for your certain tasks. So you need to be more open to test more and not just rely on those benchmarks.

[Michael Stelzner]

Yeah, and I'm going to double down on what you said. You know, Gemini by Google because it's multimodal as you said, it's got eyes, it's got ears and it can not just read, it can hear and it can see. And that's really important. And ChatGPT is evolving in that general direction. But there is something to understanding that a model that was just trained on text is not seeing the whole world and all the context that's out there where models that are trained to see things and understand things and listen are going to have a different perspective because they are Able to do things the other models cannot. So that's kind of setting a little bit of a groundwork stage. So for those who want to expand their Horizons beyond just ChatGPT, what do they need to be thinking about before we get into kind of some use cases here, what are some of the things that maybe they need to consider?

[Grace Leung]

I think so first is have a strategic or structured testing framework. You need to identify what are some recurring tasks for your day to day work. And do not start with different so many different use case. So for example, maybe your use case is about strategy work and then you pick the recurring tasks and then you design a prompt and then you try the same prom on different models at the same time because that would give you the most objective way to test the model ability. So you want to test the speed, like how fast is it responding, the accuracy, is there any bias or inaccurate information and also the tone. Right. How they present output. Because I would say for some model, maybe Claude, I use Claude a lot and is always have that kind of human touch, is always comfortable to chat with. While maybe for ChatGPT it's more logical, it's more subjective. Yeah, this is the fact. So you need to test it out and understand if this is the right tool to fit the task. And then the second step is to map out the workflow. Okay, so if I decided maybe perplexity is better for research, then I will use perplexity as my go to tool to do all research tasks. Maybe after perhaps, maybe in my case I always use it for strategy work and then I always love claw visual storytelling more strategic. So I will use that as my go to tool for doing this kind of tasks. So you need to first understand what is your recurring task and then map out your workflow and understand if any bias. And one tip I have is perhaps maybe you can just ask it some factual question. For example, let's say I just mix this up. Maybe I'm the marketing manager of notion. And then I would ask like give me what you know about this brand. Like give me the brand overview, product messaging, everything. And so you can compare it if there any bias, if any gap between different models. And then you can also study their reasoning, the chain of thought thinking. Right, because you can see there's some chain of thought thinking and how different models will break the complex tasks into multiple steps. Because that will also tell you a lot how different models will approach different tasks. And you can also spot okay if this is making sense or not. So you can decide which one you Want to use it more?

[Michael Stelzner]

I love this. I'm going to reiterate what I heard you say. Number one, have some sort of a structured framework. Identify some actual applications that you're already using internally. And once you've got a prompt, take that prompt and try it out with another tool like Gemini or Claude, and then just notice whether or not you feel like it's giving you response that's a better response. Maybe it's faster, maybe it's the way it's communicating is more preferable to you. And then specifically, you said look for biases. Right. And to look for biases, ask it things that you know the answer to, maybe about your brand or your business, and then see which one is more accurate. That might give you insights that might be more useful. And then of course, map it into some sort of a workflow. Right. So once you begin to identify, okay, this one here seems to be better for X. Like I'll say for my own case, I use Claude nearly exclusively for writing persuasive content, and I use ChatGPT to come up with very simple ideas. And I know that ChatGPT is good for coming up with ideas. Also, I'm working on a survey and I use CLAUDE to help critically analyze the survey. And then I took it over to ChatGPT and it actually found things that Claude didn't find. So I actually used both models. I used CLAUDE for OPUS and I used chat GPT03, which were the two most advanced models at the time of this recording, for each of these. And I found that each of them found things the other did not. And it had preference differences. And I was able to go back and forth and just kind of like use both of them. And each of them had a little value that they added to the equation. But it, but I. And it resulted in a better list of survey questions that I would have never had if I was only using one tool.

[Grace Leung]

Yes, I love it. Yes. So that's exactly why we need to use multiple tools. And especially when it comes to important work, strategy work, where you just want to have more diverse perspective, you just don't want to get the same response from one model. So I think this is. Yeah, the exact.

[Michael Stelzner]

Okay, cool. All right, so let's we come up with a couple of different use cases that we're going to talk through in detail here. Let's start with Animal analyzing large sets of data. Now, before we explain how to do it, why don't you discuss, like, what do we mean by analyzing large sets of data? Let's kind of define a little bit about what we need and what we mean and why we might do this because some people may have never done something like this before with AI and then we can get into the how. So let's describe what we mean by analyzing data sets and then let's talk about how.

[Grace Leung]

So data sets means it will consume a lot of tokens. Maybe it just is large reports have over a few hundred pages is like consume a lot of tokens. And to give you idea, for example Gemini can take up around 2 million tokens, is around 1.5 million words. So it's a lot of so many words and is in terms of tokens it can be as data sets or it can be a reports with so many pages. So here's how we define it as a large data set or files.

[Michael Stelzner]

And what might we do with those large data sets and files. Like let's just talk about the applications a little bit and then we can talk about how can AI assist? What would it be doing in this particular situation?

[Grace Leung]

Yeah, so actually this is one of my favorite use case because AI is so good in analyzing patterns, it's so good in analyzing unstructured data into structured format. So for example, you can just copy and paste maybe YouTube comments or maybe Photon reveals my business profile reviews and then just paste it to Gemini and then I'll say okay, help it to make it more structured, maybe present in a table so I can download it for further analysis. This is one of the ways you can do it. Another way is you can just directly ask it to give you the insights from these structured data and then help me to grip by themes. What are some common themes you can see from these large data sets? So I would say this is a game changer because most people just jump directly asking AI for recommendations. But however, AI is just so useful in finding linkage and insights from those unstructured data sets.

[Michael Stelzner]

Okay, so we're going to talk about how to do this in just a second. But the two kind of examples we were talking about is possibly lots of files, for example PDF files, right? You could have a whole bunch of different files or you could have one really huge file, right? Or you could have unstructured data like a bunch of comments, right? So you could have a whole bunch of comments that you could take off of a YouTube video, for example. Maybe there's hundreds of comments in a YouTube video, right? And. And that's an example of unstructured data. And you get that data somehow into Google Doc or a sheet or something like that. And then when you have all these files, large files or multiple files or large amounts of data, what you said, the beauty of AI is it can help you discern insights and stuff from that. So talk to me about how we would use the different AI tools together to kind of create something powerful.

[Grace Leung]

So one of the use case is for example audience research. I believe most marketers need to do audience research. And so let's say I have so many data sets, large maybe reviews, comments or anything, maybe reports. You just put it to Gemini 2.5 because Gemini the beauty is it can intake a large Data set with 1 million tokens Window size and then you can ask it to give you the data summary. So it's not necessarily, not necessarily the insights but however the data summary which means that what are some patterns you can see, what are some words or phrases that is used a lot. And then you can ask it to do some statistical summary analysis, just an initial analysis and then you can fit this analysis back to cloud. So why I use I prefer cloud in this use case is because I always find that cloud is a bit more strategic and also when it build a dashboard is it's present a story better. So I can just put this data to cloud and then ask it to generate the dashboard. Maybe like based on these analysis come up with two to three key Persona details. So what are the things you identify, what are the common pain points and then what are the motivations? Perhaps maybe you are starting a new SaaS product. Maybe is an email marketing product. So you have, you have Gemini to do the analysis and then you have Claw to define the Persona for you. So that or maybe even you can take it even further. Help me to recommend some messaging that would engage with these three target Personas. Based on the reviews, you analyze the theme, you analyze what matter the most to these target Persona. This is one of the most useful way in using them together.

[Michael Stelzner]

Okay, a couple of quick things folks. Gemini is if you have a paid Google account, then you have gemini.google.com and it's important. Why don't you distinguish between 2.5 flash and 2.5 pro. And this could change in the future. But explain which one is the better one for doing this analysis because it might be 2.6 or 3.0 by the time you listen to this in the future. But. But when you go into Gemini you have typically options. Which one should they be choosing when they're doing this initial analysis inside of Gemini, The Flash or the Pro?

[Grace Leung]

What's your thoughts I would prefer Pro for all kinds of data heavy analysis. So in the case we just discussed like audience research, I'll definitely prefer Pro. While Fresh is more about quick response, it's more lightweight so you just want to get ideas quickly, get response, maybe drafts use some content is good. However for 2.5 Pro is more about data heavy, maybe coding, maybe some other tasks that you need to do more through analysis. This is how I would like pick between the two.

[Michael Stelzner]

Okay, now when we take the output that comes from Gemini 2.5 Pro, are we just taking the output over to Claude or are we taking the output and the original data over to Claude? I'm trying to understand like that step from going from Gemini over to Claude.

[Grace Leung]

I will just take the output generator from Gemini to Claude. Or maybe if you prompt Gemini you can just let Gemini know that okay these analysis the output, I'm going to pass it to another AI to do more in depth analysis. So it also helped Gemini to generate more useful output. So I would just use the output generated and then pass it to Claw and not the original data set. Because the thing is, Claw has strict usage limits and also can't handle super large data set. So that's why in this case, that's the reason why we want to use Gemini to do that like dirty work initial nonsense before we pass to Claw for more attention to detail strategic work.

[Michael Stelzner]

What if I told you while your marketing peers are still fumbling with basic AI prompts, you could become the AI expert your entire company depends on. That's exactly what's happening inside the AI business society. Our members aren't just using AI, they're mastering it. They're automating repetitive tasks, creating professional videos in mere minutes, and making data driven decisions that leave everyone on their team completely amazed. Most marketers are barely scratching the surface of what AI can do for them. While they're stuck creating mediocre content, you could be leaping so far ahead that people look at you as if you're from the future. Here's how we make that happen. First, you get monthly live training from leading AI experts who actually use these tools in real marketing situations. Secondly, there is a incredible community of innovative marketers who are pushing the AI boundaries every day that you have access to. Third, we have multiple meetups per month where you get a chance to interact with your peers and ask questions from our experts. As member Lisa Kanda said, quote, the quality of training offered is the best I've experienced and from an organization that I can trust. Here's the truth the marketing landscape is changing very quickly. Don't get left behind while others race ahead. Visit social mediaexaminer.com AI and become the AI Enhanced Marketer your company or your clients need future you will. Thank you. Visit social mediaexaminer.com AI to learn more. I'm a heavy user of Claude, just like you are, and they have Sonnet and Opus, and as of now, I believe it's 4.0. Which one do you recommend when we're doing this kind of work? Specifically inside of Claude, Opus or Sonnet or do you find it doesn't matter?

[Grace Leung]

I use Sonic more, to be honest. I've also tried Opus, but I would say just from my personal experience, the token was just burns so fast with.

[Michael Stelzner]

Opus in particular, right?

[Grace Leung]

Yes, Opus. Yeah. So that's why I believe. And also from the official guide from Claude, the recommendation is if you are doing heavy coding, website development, landing page building stuff, maybe you want to prefer using Opus, but however, maybe this kind of analysis, you can just use Sonic. It's smart enough.

[Michael Stelzner]

Claude for folks that don't use Claude, has this really cool thing called Artifacts. Do you recommend artifacts at all when you're using Claude? And if so, describe a little bit of how that works because I think that's really cool. Right?

[Grace Leung]

Yeah. So artifact is just like a prototype that you can iterate in different versions. So you can just, okay, generate an artifact and it's working. You can maybe build tools. So there are so many ways you can use Artifact. Maybe you can build Dashboard, maybe you can just build a lightweight tool, maybe an SEO tool or a calculator. So this is just so useful. Like, and I would say this is one of the best way to use claude.

[Michael Stelzner]

Yeah, well, and Claude also Artifacts will format it in a beautiful format for you too. So if you want it to be having different headers and bullets and you want it to be nicely formatted, Claude will almost look like a Google Doc. And then you can just copy and paste it. So it allows you kind of to see what it looks like before you actually, you know, like, normally with these models it's not formatted beautifully, but when you put an artifact, it can make it look much nicer, which is I really love. Okay, so we've talked about this first application. Any other tips on moving from large sets of data, doing the initial research with Gemini and then moving over to Claude. Is there any other little things that you've discovered? Like, for example, do we accept Gemini's first output as the truth, or do we want to modify it and tweak it a little bit. You know when you're doing that initial analysis with Gemini.

[Grace Leung]

Yeah. So I think Gemini is good for doing all those rapid prototyping, initial, just all those dirty works and then you can just pause it to maybe do the fine tuning. So for example, maybe you are building some landing pages or building tools. You can just ask Gemini to build a prototype and then you can just pass it to Claude to refine it. Because the pain point of using cloud has always been the usage limit. So you want to always have Gemini to build the groundwork, the framework for you before you pass it to Claude. Another thing and use case I think of is because Gemini is multimodal. So you can ask it to analyze maybe videos, maybe audio and then you can analyze it, pass it to Claude to refine it. Because Claw is so good at writing, doing the style guide and so you can ask it to generate the style guide and. And then you pass it back to Gemini to maybe generate the audio. Because the multimodal capability of Gemini is real. Yeah.

[Michael Stelzner]

Okay. On that note, do you recommend uploading actual audio and video files into Gemini or do you recommend linking to videos? Which is. If you want to get audio and video into Gemini, what's the best way to get it in?

[Grace Leung]

From my experience, uploading is always the best. You will ensure the context doesn't lose the context. But for Google AI Studio there is native features offered that you can import the YouTube videos. So for example is the video is already on YouTube, you can just import it so easily and it will do all those analysis. It's so crazy because Gemini can detect the pause, the pacing, the structure, if you have even smile throughout the videos. So you will get a lot of really good analysis. Perhaps maybe you're doing some presentations, you want to improve that or any other on camera presence, you want to improve that so you can use Gemini in this way. And then before pausing the call.

[Michael Stelzner]

So we should just mention because you mentioned AI Studio. For those that aren't familiar, aistudio.google.com is more of their like engineering technical interface. It's a little more complicated to use. I can speak from my own experience. I've had mixed results with YouTube videos. I think it depends which model you pick. I've given it videos that I published and it couldn't even read it. So I don't know if you've had success recently. I think part of it has to do with which model you you pick because in Gemini there's a lot of different models. Is there not?

[Grace Leung]

Yes, yes. A lot. Like in June and then in May and so But I always, my preference always pick the latest. The latest version. Yeah. So far I haven't encountered the same issue as.

[Michael Stelzner]

Okay, well that maybe they've changed that. Okay, perfect. Okay, so now let's pivot to research. I know we've talked about research, but this is a different kind of research. Let's assume we're doing Internet research. Let's explore the kind of tools that we might use to help us do that and maybe talk about some specific kinds of research that we're talking about here.

[Grace Leung]

Yeah, so for research I would say there are few types of research, like everyday research, just get quick answers or maybe a more in depth research and then also maybe researching from your close knowledge base. So I think nowadays I use different types of tools for everyday searches, quick searches, I use Perplexity the most. For deep research, I use Gemini more and for maybe I just want to deep dive into the sources that are chosen, just this knowledge base. And then I use Noble LM because I think it's such a fantastic tool in so many ways. And I would say. And it all comes back to the recurring tasks that we talk about, like what are the most research needs for you? So for example, Perplexity, I always think it has a good research source, a source variety. But however, Google may be prioritized more bigger brands or maybe sources that have high authority. So you need to try that yourself. So sometimes maybe I just want to gather more sources. Maybe I'll prefer Perplexity and then I would just pick the sources that I got, I select them and I will import them to Nople lm. And what makes noblelm so useful these days is because it also just recently launched a bulk upload feature. So you can just ask Perplexity okay to export the list of sources for me and then I can just paste back to NotebookLM and do the bulk import. So it's very handy and so you can just combine different tools and then ask no problem question.

[Michael Stelzner]

Okay, so let's dig into this a little bit more. Let's start with Perplexity. Perplexity has been around for a while, but I think what I heard you say is use Perplexity for quick answers. Now give us some examples of how you would use Perplexity and what kinds of quick answers. Because many people listening right now might be just using chatgpt for quick answers instead of Perplexity. So kind of explain in more detail, kind of the advantage that Perplexity Brings to the table.

[Grace Leung]

Okay, so for example I am planning the go to market for certain brands or maybe categories, maybe an E commerce brand. And then I want to know the top competitors or top market players in this view. So I can just. This is an example of quick search. What are some top players in this field? Give the details for me so I can dig deeper. And then I would just select them, pick them to Noble lm. So for example, I want to enhance the messaging and then I'll just ask Perplexity, give me the top 20, maybe 10 to 20 like top players and then I will bulk import all the URLs to nople alm and then ask nope LLM question. Okay, so what are the content gap? What are the messaging in all these top market players you have seen? And I asked them in case I also want to build an odd new product. So how would you recommend for me to do the enhanced messaging and the beauty of Noople LM is the minimal level of hallucination. So it will always based on the sources that you import before it gives you the answer and nothing more. Yeah.

[Michael Stelzner]

Okay. Couple questions on Perplexity for those who've been listening to the show for a while. Understand that we're not supposed to use AI like a search engine, right? And the prompt matters. But what about Perplexity? Should we use very simple queries with Perplexity? You know, just give me the top 20 competitors or is it makes sense to give it a much larger prompt?

[Grace Leung]

I would say if, if you're just doing some quick analysis, if you just wanted quick answers, you don't mind the prompt engineering.

[Michael Stelzner]

Keep it simple.

[Grace Leung]

Yeah, you get your objectives so clear and you communicate in a way, so clearly. But however, maybe you are doing deep research, you just want to have some focus, research focus, prioritize what kind of sources, then you want to have some more detailed prompt. Or maybe this day Perplexity has just launched Perplexity Labs, which is research plus building. And also there is a quota of how many credits you can use for Perplexity Labs.

[Michael Stelzner]

So is it labs Labs. Is that what Labs. Okay.

[Grace Leung]

Labs.

[Michael Stelzner]

Okay.

[Grace Leung]

Yes. So that's another story. So because you have limited usage and also because it's a building, it's an agent agentic thing. So you want to have more detailed prompt. Maybe you give it more directions. What are the things you wanted to build? Or maybe you're using the deep research mode. What are the sources want to prioritize the research focus. And a tip for me is if you are doing Deep research with Perplexity, not just getting everyday quick searches, you can actually use a reasoning model, maybe ChatGPT03. You can just ask it to give you the research plan first before and then you pass it to Perplexity to execute it. So this is also my favorite way.

[Michael Stelzner]

Of like wait, okay, so hold on a minute, I want to make sure I got that. So I think what you said Grace, is that you could go into ChatGPT and use one of the reasoning models like O3, which right now is one of the more advanced ones, and ask it to do some of the preliminary research and then put that into Perplexity. Did I hear that right?

[Grace Leung]

Yes, yes.

[Michael Stelzner]

What would that look like? I mean, how would that result in a different prompt, if you will, in Perplexity? Because I thought you said keep it simple when you're putting prompts into Perplexity.

[Grace Leung]

Because for Perplexity you can do the deep research. But however, if you want to make sure it is high quality. So what I prefer what you can do is if you also don't want to craft the prom yourself, you can actually ask AI, an intelligent AI like ChatGPT O3 to craft the research.

[Michael Stelzner]

Ah, but not do the research, but just do the prompt is what you're saying is that.

[Grace Leung]

Yes, the prompt. So output the research plan in a prompt in a prompt format. So I can parse it back to Perplexity because what I found is Perplexity. Of course you can give it more detail prompt, but nowadays AI become more advanced, it is better in crafting the prompts. I can definitely see this, this trend. So you can just ask it to give you the version and then you fine tune it so it will give you more structured prompt like the research focus sources, how you should approach this research. Besides you just let Perplexity to do them themselves. So I would say this is a way to improve the quality from the research.

[Michael Stelzner]

Okay, couple questions about Perplexity then I want to come back to this Deep research stuff. Are you using a free version of Perplexity or a paid version of Perplexity in this research?

[Grace Leung]

Paid version.

[Michael Stelzner]

What's the advantage to the paid version?

[Grace Leung]

So for the paid version you have unlimited quick searches and then also for Deep Research you also have unlimited Deep research and then for Perplexity Labs it says capped at around 50 usage per month. Okay, but however for the free version is much more limited. Maybe I remember it's like five to ten free searches per day. I'm not sure, but there is a usage limit basically and how much does it cost?

[Michael Stelzner]

Because a lot of people are using the free version of Perplexity. Do you know approximately what you pay for perplexity?

[Grace Leung]

$20 per month.

[Michael Stelzner]

Okay, and does the paid version allow you to select which model? Because doesn't Perplexity offer different models? Like, doesn't it allow you to run through the API, ChatGPT, Claude, Gemini? Or is that not true with Perplexity?

[Grace Leung]

So Perplexity, it allows you to change the model. It has the features allow you to generate your own API key. But however, it doesn't give you options to use your own maybe API key from Claw. It doesn't work this way. But however it has. Just think of it like an AI wrapper. They just have like a middleman and then it can connect to different models. Okay, which model is your favorite? And then you can just pick it through the interface.

[Michael Stelzner]

When you're doing research in Perplexity and you want to grab those 20 recommendations, is there an easy way to get those other than opening 20 tabs and copying them all? Are you saying when you're doing Perplexity, quick answers, right? And you want to get like, who are the top competitors in this industry? How do you get all that over to Gemini for the Deep Research? Do you have to have each URL copied into Gemini? Do you understand what I'm asking?

[Grace Leung]

Okay, so what I usually do is I will export it to a markdown file and then I can open it and then I can just so from the exporter file there is a citation format because Perplexity, we include that in the exported file. So whatever format you choose, maybe markdown, maybe Google Doc, maybe DB PDF. And then what I would do is I will do a little reformatting. I would just ask maybe ChatGPT or Gemini. Okay, help me to remove the number so I can just directly and paste it back to Noble lm.

[Michael Stelzner]

Okay, interesting. So Deep Research, as of today, pretty much everybody offers Deep Research. Claude has it, Grok has it, they call it deep search, ChatGPT has it, Gemini has it. Why do you like Gemini over the others? Even ChatGPT has it. So what's your thoughts on why Gemini for Deep Research?

[Grace Leung]

First thing is Google is the best of doing search, right? Although it's not pivoting like from the traditional search to search. But like this is what Google do. The best is always search. It has a large database and also it has not just website but also from YouTube. So like when I mentioned you can try the same prompt using both ChatGPT research or Gemini and I always find Gemini can find much more sources, right? On average maybe few hundred, sometimes even over 500 websites.

[Michael Stelzner]

But it is also it generates a massively big file though. I mean right? This is the difference, right? Like I found with ChatGPT's deep research. The answer is a little bit more precise where with Google it's like a monster document that I have to page through, right? Is that good or is that bad? I mean, I guess it depends on the application, right?

[Grace Leung]

Yes, it depends on the application. And other thing is, I would say Gemini will give you a more overwhelms for some, maybe you are not familiar with some new topics in AI or new subjects. So Gemini Deep Research is so good and I would say it's more generous in terms of the usage limits. So actually it has a usage limits but however it doesn't officially say how many degree research you can run per month. But however on ChatGPT because I'm on a page the pro plan, I'm not a. I'm not, I'm on the plus plan. I'm not a pro plan. Yeah, so. So that's why you have to very cautious about the usage. But at Germana there's no such thing. And also I, I would say also yeah, it depends on your tasks. Yeah.

[Michael Stelzner]

Okay, you mentioned for internal research Notebook lm. So explain that application so people can process that a little bit. Because I think we mentioned it, but maybe very briefly, what do we mean by internal research and what is it about NotebookLM in this case that makes it different than the other stuff we've been talking about?

[Grace Leung]

So first of all, NotebookLM is a Google product is powered by Gemini. So that's why it also has a big context window. You can upload 300 sources. If you are using the Nobel LM plus version, you can just upload so many sources to it. And the beauty is it will give you the response based on the sources you import and nothing more. And so the hallucination can be minimized. And so one way is you can. So for example, nowadays Google is doing emphasizing more AI search. So one way you can just ask it to discover the source. This is also a new features from NobleLM because in the past NobleLM you cannot use it to find sources, you have to import them one by one yourself. But nowadays it has to discover source. So that's why you can just describe the top pick and then it will find a source for you. And the beauty is because it's powered by Google so it somehow also leverage Google's ranking or how Google prioritize the websites. So you can kind of reverse engineer how Google thinks what are the quality websites? So for example, if I want to reverse engineer the ranking for a project management tool like what is the best management project management tool and then I submit the query in nople lm and then it will feedback the 10 top sources for you. So these why these 10 top 10 sources is because Google think that they are high quality, right?

[Michael Stelzner]

Oh, so those sources are articles written by somebody else presumably, right? Is that what you're saying?

[Grace Leung]

Yes, yes. So it will come up the sources. So it's just like the ranking in the Google traditional search result page but however now translated to Noble LM and AI tool and then I can just import it. Okay, These are the 10 sources I upload to NobleLM, import it and then I can reverse engineer. So tell me what are all those subcuries I should write content about so that I have a higher chance to rank higher to make sure I cover all the search intent for user. Because nowadays we all know that it's no longer just matching keywords, right? Is more about matching the potential topics a user asks about in that subject. So you want to cover that as much as possible so that you can cover the whole user journey and not just matching keywords. And this is exactly noblelm can help you do and always think nobodylm is a secret SEO tool because it will give you a lot of insight how you can plan the content by just import those. What is working in the eyes of Google?

[Michael Stelzner]

Wow, Grace, this has been really interesting and hopefully people have a lot of new perspectives on how they can use different tools to accomplish different things as part of a workflow or process that they need to do for their work. Now, Chris, if people want to connect with you on the socials, where do you want to send them? And if they maybe want to work with you, where should they go?

[Grace Leung]

So so they can follow me on my YouTube channel, I share AI and about digital marketing and if they want to work with me, they can submit a form. There's a contact form on my website gracelearn.com they can also just find me on LinkedIn, Instagram, send me a DM that might be even faster. That's how people can find me. And also on X. Yeah, awesome.

[Michael Stelzner]

And folks, Grace spells her last name L E U N G. Grace, thank you so much for coming on the show and sharing your insights with us today.

[Grace Leung]

Thank you so much for having me today. That's a great job. Thanks, Michael.

[Michael Stelzner]

Hey if you missed anything, we took all the notes for you [e-mail] a63 also, be sure to follow this show on whatever app you're listening to us on. And if you've been a listener for a little while, I would love a review or maybe share it with your friends. Tag me on the socials and also do check out the other shows we produce, the Social Media Marketing Podcast and the Social Media Marketing Talk Show. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful.

[Grace Leung]

The AI Explored Podcast is a production of Social Media Examiner.

[Michael Stelzner]

Just a quick reminder before you go. If you're ready to become indispensable in the age of AI, the AI Business Society is your solution. Join now and secure your discounted membership by visiting social mediaexaminer.com AI I can't wait to see you inside the AI Business Society.