June 5, 2026

AI for the Over 40 – Week 26: AI Research Strategy Starts Before You Research

I’ve been trying to decide when to use Microsoft 365 Copilot versus Power Automate versus Zapier.

Not a simple feature comparison. I wanted a decision framework that could help me determine which tool fit which scenario — and when a hybrid approach made the most sense.

A few years ago, that kind of research project would have required hiring a consultant, dedicating weeks to evaluation, and spending money I probably couldn’t justify.

Instead, I produced a 400-line strategic framework in about three hours using tools I already subscribe to.

But the most important lesson wasn’t about the AI platforms themselves.

It was about what I did before I asked AI to research anything.

The mistake most people make

When people discover “Deep Research” or “Research Mode” in AI tools, they usually do what feels natural:

They type a question and hit enter.

Fifteen minutes later, they get a polished report that sounds impressive but doesn’t actually help them make better decisions.

“Compare Power Automate and Zapier.”

AI will happily generate pages of comparisons: features, pricing, pros and cons, integrations, and limitations.

It will also be almost useless.

Because comprehensive isn’t the same thing as actionable.

The quality of research output is determined by the quality of the research question. Most people skip the question-design step entirely.

Think before you research

Before I touched Research Mode, I opened a conversation in Gemini’s Thinking Mode and asked:

“Can you help me design a deep research prompt to explain when to use M365 Copilot, Power Automate, Zapier, or a hybrid approach?”

What followed wasn’t an answer.

It was a collaborative design session.

The AI pushed me to clarify:

• Who is the audience?
• What decisions are they trying to make?
• What constraints matter?
• What would make this useful instead of merely comprehensive?

That conversation transformed a vague request into a structured 300-word research prompt with:

• Clear comparison dimensions
• Specific output requirements
• Audience framing
• Decision criteria
• Real-world context

That five-minute design conversation completely changed the quality of the research that followed.

The AI Olympics, upgraded

Back in Week 2, I introduced the idea of the “AI Olympics” — running the same prompt across multiple platforms to compare strengths and weaknesses.

I’m still doing that today. The difference is that now I’m testing frontier models with research-grade prompts instead of basic queries.

I ran the same prompt through Claude, ChatGPT, and Gemini in their deep research modes simultaneously.

The differences were substantial.

ChatGPT produced the most exhaustive output: highly detailed, technically thorough, and packed with specifications. But it buried practical guidance under hundreds of lines of information.

Claude produced the most usable framework. It generated memorable metaphors, decision heuristics, and language that would actually work in executive conversations.

Gemini landed somewhere in between, with a strong architectural structure but a more academic tone.

The lesson was important:

The most capable model is not automatically the most useful one.

The right output depends on the audience and the purpose.

The synthesis layer

Here’s where things became really interesting.

After reviewing all three outputs, I asked Claude a simple question:

“Can you improve your report using the strengths from Gemini and ChatGPT?”

That’s it.

What came back was better than any single model’s original output.

Claude combined its accessible framing with ChatGPT’s technical depth and Gemini’s structured organization. The result felt less like using one AI system and more like managing a collaborative research team.

That’s the evolved version of the AI Olympics.

Not just comparison. Synthesis.

The amplification effect

The math here is difficult to ignore.

Traditional approach:

• Hire consultants or analysts
• Spend weeks researching
• Build comparison frameworks
• Deliver recommendations

Estimated cost: thousands of dollars and multiple weeks.

What I actually did:

• 5 minutes designing the prompt
• 20 minutes running research across platforms
• 15 minutes evaluating outputs
• 20 minutes synthesizing the best pieces

Total time: roughly 90 minutes.

But the biggest shift wasn’t speed. It was accessibility.

I wouldn’t have done this research at all before AI because the effort required would have outweighed the perceived value.

AI didn’t just accelerate existing work. It made previously unrealistic work possible.

What the research actually produced

The final framework generated a genuinely useful decision architecture.

One example framed the three tools as a “three-speed gearbox”:

• Copilot handles unstructured intake and analysis
• Power Automate manages deterministic business logic
• Zapier acts as the connective bridge between systems

Another heuristic became instantly memorable:

“If the CFO would fire you for getting it wrong, use Power Automate.”

The framework also clarified the characteristic risk of each platform:

• Copilot risks hallucination
• Power Automate risks brittleness
• Zapier risks governance and data leakage

Those aren’t abstract insights.

They’re practical decision shortcuts I can use in client conversations immediately.

The technique, simplified

After 26 weeks of experimentation, the process has become surprisingly straightforward:

1. Think before you research

Use standard chat or thinking mode first. Collaboratively design the research question before launching deep research.

2. Run the AI Olympics

Use the same prompt across multiple frontier models. Different systems still excel in different areas.

3. Evaluate for usefulness, not completeness

The best research output is the one your audience can actually use.

4. Synthesize the strengths

Treat models like collaborators. Combine what each does best.

5. Iterate

The first output usually isn’t the final one. Refinement matters.

Your Week 26 challenge

This week, pick a research question you’ve been avoiding because it felt too large, expensive, or time-consuming.

Before using Research Mode, spend 10 minutes designing the question collaboratively with AI.

Clarify:

• Audience
• Decision needed
• Constraints
• Desired output
• What “useful” actually means

Then run the research.

If you have access to multiple platforms, compare the outputs. Notice which system best serves your actual purpose.

Because the real breakthrough isn’t that AI can research for you.

It’s that AI can help you think more clearly about what’s worth researching in the first place.

The bottom line

Twenty-six weeks ago, I thought AI might help with small productivity improvements.

Now I’m producing strategic research that previously would have required resources I simply didn’t have.

But the transformation isn’t the technology itself.

It’s the system that’s emerging around it:

• Thinking before researching
• Comparing across platforms
• Synthesizing strengths
• Designing intentionally instead of prompting casually

The techniques from earlier weeks are starting to compound into something larger.

Not just better prompts.

A better way of thinking.

This post is part of my “AI Over 40” series. It first appeared on LinkedIn: AI for the Over 40 [Week 26]: Think Before You Research

Read more AI and Copilot blogs.

Trending Posts