You're Missing Out Big Time If You Don't Use This #ai #smartphone #workautomation
You're Missing Out Big Time If You Don't Use This #ai #smartphone #workautomation
#GoogleAIStudio #Gemini #FreeAI Save $30 a month. A real-world PM's guide to using Google's free AI Studio. 5 secret Google AI Studio features only working PMs know (Boost productivity 10x) Google released an insane AI tool for free, and you're still not using it?
Stop spending money on ChatGPT Plus or Gemini Advanced. With Google's free AI Studio, you can actually use Gemini even more powerfully.
As a PM with 6 years of experience at an AI startup and a background in design, I'll reveal exactly how I use AI Studio's hidden settings (Search Grounding, Thought Budget) and the expert's secret weapon, System Instructions, to skyrocket my work productivity. I'll even share my practical workflow using the Nanobanana model.
#AIproductivity #workautomation #PM #planner #marketer #GoogleSheets #Gemini #GoogleAIStudio #VibeCoding
[Timeline] 00:00 The True Value of AI Studio You Didn't Know 01:10 Secret 1 & 2: The Importance of Search Grounding and Thought Budget 03:25 Secret 3: 'System Instructions' to Perfectly Tame Your AI 05:40 Secret 4: Practical Use of 'Nanobanana', the Image-Specialized Model 08:10 Secret 5: 'Code Execution' Data Analysis for Non-Developers 09:20 Practical Workflow: This is How a PM Works 10:00 Real-World Q&A (Large Files, Security, Spreadsheets) 10:30 Next Step: Building Your Own Micro-SaaS
#planner #marketer #GoogleAIStudio #Gemini #FreeAI #GoogleSheets #Gemini #GoogleAIStudio #VibeCoding
'This feature' from Google that AI experts use in secret. Tired of giving the same instructions over and over? With Google AI Studio's 'System Instructions', you can train the AI to be your perfect assistant. Discover the expert's secret in the full video. #GoogleAIStudio #WorkHacks #AIUsage
Hello, this is Callit AI, your guide to surviving in the age of AI. Are you spending nearly $30 a month on ChatGPT Plus or Gemini Advanced? Today, I'll share, based on my experience and data, how you can save that money and actually run Gemini more powerfully for your real work. The main stage for this is Google AI Studio. It's the perfect space for experimenting with the latest Gemini models and designing workflows, and it's great for getting your hands dirty in a real startup environment. You can start right away in AI Studio, try out prompts, share results, and iterate quickly from a VibeCoding perspective. You can start for free, and even if you add a paid plan, there's no charge for using AI Studio itself.
Before we dive in, we need to establish the two pillars that experts always set up when handling Gemini. The first is 'Search Grounding', and the second is 'Thought Budget'.
First, 'Search Grounding'. This is a safety bridge that allows the model to access real-time information without making things up, or 'hallucinating'. When you're dealing with frequently changing information like the latest policies, prices, or competitor release notes, if you don't turn this on, the AI will confidently lie using last year's data. Search grounding reduces these critical errors and even provides source links in its answers, boosting credibility. From a PM's perspective, it's a massive efficiency tool that eliminates the cost and communication loops of having to ask a team member, "Is this really correct?" and re-verify everything.
Second, 'Thought Budget'. This gives the model the time and space to develop its own thoughts. Think about it—even humans need time to solve difficult problems, right? It's the same for AI. If you provide a sufficient thought budget, the model performs 'multi-step reasoning' by breaking down the problem into several stages, planning for each stage, and self-verifying. For tasks where accuracy is key, like summarization or error checking, you can set it low. But for complex tasks like designing a business strategy or brainstorming new ideas, you must set it high. As a PM with a design background, I can guarantee you this one dial completely changes the depth and quality of the output.
Now for the actual setup and operation.
In the execution environment, I first choose the model. For tasks that require zero error, like summarizing reports, identifying risks, or organizing feature definitions, I lower the creativity temperature to prioritize accuracy. On the other hand, for tasks that need new ideas, like planning brand experience tones, ad copy, or UX microcopy, I raise the creativity temperature to broaden the scope of thought. On top of this, I manage the risk of outdated information by turning on the Search Grounding we discussed earlier. And most importantly, I lock in the team's language with 'System Instructions'. It's a structure where I define the analysis criteria, tone, length, and prohibitions at the system level, make the AI act as my advisor at the role level, and then give it the actual task at the question level. From a PM's viewpoint, this isn't just a prompt template; it's a 'collaboration protocol'. It's the team's rulebook. This is precisely why the tone and depth of the output remain consistent even when team members change.
Okay, now I'll clear up the confusion about 'Nanobanana'. Nanobanana is the nickname for the 'Gemini 2.5 Flash' image model, which you can find in the AI Studio model list. In short, it's a model specialized for image generation and editing. This model's real strength is 'consistency'. It excels at storytelling tasks that require creating a series of scenes while maintaining the identity of people and objects. This makes it advantageous for establishing a concept with a prompt, refining details while looking at the results, and extending the same characters and products across multiple scenes. Ethics and transparency are also important here. Gemini-family image generations have an invisible 'SynthID' watermark applied. This is a technology designed to allow for later verification of whether an image was generated or edited, making it a reliable safety net when transparency is required for brand or media collaborations.
Alright, let's combine all this and get into a practical example. haha First, VibeCoding-style image exploration. Let's assume we're creating the 'hero section', the most important first screen of a startup's landing page. First, we lock in the desired user behavior change in a single sentence: "The user will understand our product's core value within thirty seconds and click the free trial button." Then, we describe the target persona, user context, desired emotional tone, color temperature, texture, lighting, scene composition, the size and position of the main subject, and the background density, in that order. We generate a draft with Nanobanana, then iterate three or four times, refining the details to improve quality. Since Nanobanana is good at maintaining the consistency of people and products, we can quickly generate an ad campaign set by extending the same person and product into multiple scenes. Creating a draft with AI and then polishing the final one percent with a designer's touch—this is the most time-efficient method.
I tried with Nanobanana, but I didn't get results I liked, so I took it to Claude to work on it.
Second, pair text and images. I first refine the slogan, core value, and a one-line narrative describing the user's behavior change using a text model. Then, I instruct Nanobanana to visualize that narrative as a scene. For example, if I've decided on the narrative "An intuitive dashboard that shows complex data at a glance," I give this exact sentence to Nanobanana and say, "Create an image in a minimal and sophisticated style that conveys the feeling of this sentence." I safely pull in the latest trends and case references using Search Grounding. Matching them up this way greatly improves the consistency between the copy and the visuals, reducing revision loops.
Nanobanana is fast and great, but unfortunately, it doesn't handle Korean text input well. So, I usually download the generated image, bring it into Figma, and manually edit the text portions. Or, I just create an image with no text at all and add it in Figma.
Third, quantify qualitative data with code execution. I gather customer survey and interview results, define categories, and extract representative sentences and their frequencies. Code execution can draw simple statistics and visualizations in a Python environment, so I can quickly get the numbers I need for a PM's decision-making. For example, if I say, "Draw the frequency of each category as a pie chart," it will immediately generate an image chart. Reusing that result by pasting the table and chart into a slide deck significantly reduces lead time. This is especially beneficial for non-developers and reduces the team's dependency on the data team.
Just like in Google AI Studio, Gemini isn't great at creating pie charts either. So I usually use Google AI Studio for text-based tasks like writing prompts and strategies, and then I take it over to Claude, GPT, Kimi, or GenSpark to do the image and visualization work.
Let's also dive deep into some common real-world concerns. When training on large documents gets stuck, 'chunking'—breaking the file into smaller pieces and feeding them sequentially—is still the most practical solution. The important thing is 'context preservation' between chunks. You need to re-inject a summary and key keywords from the previous chunk at the beginning of each new one, and at the end, explicitly ask for a comprehensive summary of the whole thing. From a PM's perspective, processing dozens of pages of meeting minutes and competitor research notes this way noticeably reduces the time it takes to write a first draft of a strategy document. The goal of chunking isn't just to reduce volume, but to build up snapshots while preserving context.
I also get a lot of questions about connecting company Google Workspace documents directly to a personal account model. This isn't a technical issue, but one of security and policy. Workarounds are a no-go; you must use official channels that leave permissions and logs. At an organizational level, instead of external search, you can design a system that uses 'internal search'—searching only company-internal documents—as a model tool to ensure up-to-date information. The principle is the same as Search Grounding, but you bring the data control in-house.
Watch the Video
This post is based on our YouTube video. Watch it for more details!
Originally published on YouTube: 10/14/2025