EP 607: How To Use Gems, GPTs & Projects for Real Business Results
🎯 Summary
Podcast Episode Summary: EP 607: How To Use Gems, GPTs & Projects for Real Business Results
This episode of the Everyday AI Show focuses on a critical shift in how professionals should be utilizing generative AI tools—moving beyond the default “New Chat” button to leverage persistent, customized environments like Google Gems, OpenAI’s Custom GPTs, and the Projects feature in both ChatGPT and Claude. The host argues that failing to use these organizational tools leads to wasted time due to repeated context setting and poor organization.
1. Focus Area
The primary focus is on Practical AI Implementation and Workflow Optimization across major LLM platforms (OpenAI, Google, Anthropic). The discussion centers on the strategic advantages of using Customized AI Agents (Gems/GPTs) and Organized Workspaces (Projects) to maintain context, leverage proprietary data, and achieve superior, repeatable business results.
2. Key Technical Insights
- Model Selection in Custom Tools: A significant recent update is that OpenAI now allows users to select the most powerful models (like GPT-4 Turbo) for their Custom GPTs, a feature previously restricted, making them far more capable than before.
- Platform Feature Parity and Differentiation: While Gems/GPTs and Projects share the core function of custom instructions and file access, Projects (in ChatGPT and Claude) often retain advanced features like Canvas Mode (ChatGPT) or Artifacts (Claude), and specialized research tools (like ChatGPT Projects’ Deep Research), which are often disabled in the simpler Gem/GPT format.
- File Handling Limitations: Claude has a notable limitation where uploaded files cannot exceed 30MB, which was a constraint encountered when trying to upload large datasets (like email history), whereas Google Gems and ChatGPT GPTs handled larger files.
3. Business/Investment Angle
- ROI Through Context Persistence: The central business argument is that setting up these persistent environments eliminates the time wasted repeatedly feeding context, brand voice, or proprietary data to the AI, leading to significant time savings and better output quality—a direct path to Gen AI ROI.
- Democratizing Consultancy: By loading extensive proprietary data (analytics, search console data, internal documents) into a custom agent, users can potentially generate high-level strategic insights that might otherwise require expensive, time-consuming external consultancy.
- Competitive Advantage via Organization: The ability to organize work within Projects (folder structure) prevents the chaos of unsearchable chats, ensuring that valuable AI interactions and derived knowledge are retained and accessible.
4. Notable Companies/People
- OpenAI (ChatGPT/GPTs/Projects): Highlighted for the recent, under-the-radar update allowing users to select advanced models for GPTs and the unique capabilities of ChatGPT Projects (Canvas, Deep Research).
- Google (Gems/Gemini): Noted for Gems utilizing the powerful Gemini 2.5 Pro model and deep integration with Google Workspace apps (Drive, Gmail, Calendar).
- Anthropic (Claude Projects): Mentioned for its Project feature supporting Artifacts and the Model Context Protocol (MCP), which acts as a web-based API layer within projects.
- Jordan Wilson (Host): The host uses the podcast to provide a transparent, live demonstration of how the Everyday AI team uses these tools internally, sharing their own data and prompting strategy as a real-world case study.
5. Future Implications
The conversation strongly suggests that the future of professional LLM usage is agent-centric and persistent. The industry is moving away from one-off queries toward building specialized, data-rich AI assistants that function as embedded team members. Staying current with feature rollouts (like model selection in GPTs) is crucial, as these updates fundamentally change the efficiency ceiling of existing workflows.
6. Target Audience
This episode is most valuable for AI Practitioners, Business Leaders, and Tech Professionals who are already using LLMs daily but are struggling with efficiency, organization, or maximizing the depth of their AI outputs. It is specifically targeted at those ready to move beyond basic prompting to building scalable, customized AI solutions.
Comprehensive Summary Narrative
The episode addresses the common pitfall where users, despite using AI daily, are actually wasting time by defaulting to the “New Chat” function across platforms like ChatGPT, Gemini, and Claude. Host Jordan Wilson argues that the rapid evolution of AI tools necessitates adopting persistent, customized environments—Gems, Custom GPTs, and Projects—to save time and elevate output quality by pre-loading context and data.
Wilson frames the discussion around a live, unedited demonstration where he inputs the exact same complex prompt into four identically configured agents (a Google Gem, a Custom GPT, a ChatGPT Project, and a Claude Project), all utilizing their respective platforms’ most powerful models (Gemini 2.5 Pro, GPT-4 Turbo, Claude Opus 4). The goal is to analyze 50 specific strategic insights derived from a massive dataset comprising Google Analytics, Search Console, podcast stats, and email history.
The core of the episode is a comparison matrix detailing the pros and cons of these four persistent tools. Gems are praised for their Gemini 2.5 Pro power and Workspace integration. Custom GPTs are now significantly more powerful following the update allowing selection of the latest models and offering unique API/Action connectivity. Projects (from both OpenAI and Anthropic) are highlighted as superior organizational structures that retain the full feature set of the underlying models (e.g., Canvas, Artifacts, Deep Research), features often stripped from the simpler GPT/Gem format. A practical challenge noted was Claude’s restrictive 30MB
🏢 Companies Mentioned
đź’¬ Key Insights
"Generative AI is generative. Just because it doesn't work once doesn't mean it's broken. Right? This is why I encourage companies that I work with that hire us to, you know, don't just do something once and write it off. You need to be revisiting these things... you should at least be testing it minimum five times minimum, and you need something to measure."
"But like I tell you all every time, you need to be reading this chain of thought. This tells you because we're using these either reasoning models or hybrid models... the quality of my prompt was very low. Right? The custom instructions were not very great. So when you don't spend a lot of time before you go and set these very powerful models in motion, you don't really know how they're going to respond."
"Now think instead of go clicking new chat, now all of a sudden, if you just unload your brain, unload everything about your position, put your job description in there, who you are, who you work with, meeting transcripts, right? You can fit so much information in these Gems, GPTs, and projects that all that time that you would normally spend trying to get the most and explain things and feed all this data to a large language model, if you just set it up correctly and then use it every single time, think of not just how much time you're saving, but how much better the outputs are going to be."
"Unfortunately, Claude would only accept nine because it has a file limit. All right. So anything over 30 megabytes you cannot upload. So I had a spreadsheet that was like 60 megabits or no, it was like 42 megabytes or something like that, and Claude couldn't take it, whereas Google Gems and GPTs could."
"Generative AI is generative. It's going to be different every single time. So by reading this chain of thought, it's going to tell me how I should improve my inputs to get a better output."
"I can go through here and see the chain of thought, see how it thought, see how it thought."