Navigating AI Models: Finding the Right Fit
Navigating the landscape of Generative AI can feel like walking into an art studio blindfolded. You know you need to create something, but you aren’t sure if you’re holding a paintbrush, a marker, or a pastel.
In a recent episode of the Tech for Business podcast, CIT’s Power Platform Developer, Cole, and Brand Strategist, Kelsey, sat down to demystify the “Big Three” models—Microsoft Copilot, ChatGPT, and Google Gemini. They moved beyond technical architecture to discuss practical outputs, prompting strategies, and how to balance innovation with data governance.
The Big Three: Matching the Model to the Method
Choose Microsoft Copilot for secure, internal business data retrieval; ChatGPT for reasoning, memory retention, and “sounding board” conversations; and Google Gemini for creative visual output and live canvas editing.
The “best” AI model is rarely a static answer; it depends entirely on the context of your task. Cole and Kelsey break down the distinct strengths of the market leaders:
Microsoft Copilot: The powerhouse for business integration. Because it connects directly to your Microsoft 365 tenant, Copilot excels at retrieving specific information from your SharePoint, Teams chats, and Outlook emails. It is the go-to tool for drafting content based on internal documents without risking data leakage.
ChatGPT: The conversationalist. Kelsey notes that ChatGPT’s memory features are superior for retaining user preferences (e.g., remembering specific formatting styles or constraints like “gluten-free” for recipes). It serves as an excellent “devil’s advocate” for refining ideas.
Google Gemini: The creative canvas. For users embedded in the Google Workspace, Gemini offers seamless integration with Docs and Drive. It is also highlighted for its superior image generation and “canvas mode,” allowing for a more visual, interactive drafting process.
The “Art Tool” Analogy: Why Trial and Error Is Essential
Treat AI models like different artistic mediums. Just as a paintbrush and a marker both apply color but yield different textures, different AI models produce different nuances in tone and format.
“I look at each of the different models like an art tool,” Kelsey explains. “All of them can put red on a canvas… but I know in my gut what I want it to look like and which one is gonna help me get there.”
To find the right fit for your workflow, the team recommends:
Run A/B Tests: Input the exact same prompt into Microsoft Copilot, ChatGPT, and Gemini. Compare the outputs side-by-side.
Consult Leaderboards: Tools like “AI Arena” allow users to see which models currently rank highest for specific tasks like coding, creative writing, or logic.
Check the “Vibe”: Sometimes, a model’s default tone matches your needs better. Cole notes that Copilot can sometimes be “overeager to be right,” whereas ChatGPT might offer more critical reasoning when prompted correctly.
Prompt Engineering: Turning “Crap In” to Quality Out
The quality of AI output is directly proportional to the context provided. To avoid hallucinations or generic answers, instruct the AI to ask you clarifying questions before it generates a response.
Regardless of the model you choose, a poor prompt will yield poor results. “If you put in something ridiculous, you’re gonna get ridiculous out,” says Kelsey.
To elevate your prompting game, adopt these strategies:
The “Intern” Mindset: Treat the AI like a new intern. Give it clear goals, context, format requirements, and constraints. If you don’t tell it to “go North,” don’t be surprised if it runs South.
The “Don’t Answer Yet” Technique: Cole suggests ending your initial prompt with a specific instruction:
“Don’t answer this yet. Ask me clarifying questions first to ensure the best possible output.”
This forces the model to pause and gather the necessary context it needs to succeed.Iterative Refinement: If the output isn’t right, tell the AI why. Ask it, “What did I miss in my prompt that caused this error?” Use the tool to teach you how to prompt it better next time.
Navigating Governance and “Shadow AI”
Employees often bypass approved tools like Microsoft Copilot for ChatGPT due to ease of use. To bridge this gap, organizations should run pilot programs that define safe use cases rather than issuing blanket bans.
A common challenge in modern businesses is the gap between the sanctioned tool (often Microsoft Copilot due to its enterprise-grade security) and the tool employees prefer (often ChatGPT).
This isn’t necessarily rebellion; it’s often a workflow preference. However, it introduces risk. “You can make almost any one of these platforms have the same context knowledge,” Kelsey notes. “It’s just how much work you wanna put in and keep up with updating that versus how much do you wanna connect to live data.”
Best Practices for Secure AI Adoption:
Sanitize Data: If using a public model, strip all Personal Identifiable Information (PII) and proprietary data before prompting.
Define the Lane: Establish clear SOPs. For example, Marketing may need ChatGPT for creative brainstorming, while HR must use Microsoft Copilot for handling internal policy documents.
Pilot Programs: Identify specific wins—like sales follow-up emails—and train teams on how to execute them safely within the approved tech stack.
Listen to the Full Conversation
Ready to dive deeper into building AI agents and refining your workflow? Listen to the full episode below.
