Background: AI-Based Research
OpenAI’s latest AI model is called “Deep Research.” At the same time, DeepSeek has introduced “Reasoner.” Google’s Gemini 2.0 features “Flash Thinking,” an enhanced reasoning model capable of articulating its thought process to improve performance and explainability.
A typical research interaction with, for example, OpenAI Deep Research includes:
- After your initial question, AI will follow up with several concrete, relevant questions to fine-tune the research question.
- AI will search in 20-30 external resources.
- You can follow its internal argumentation and method of thinking.
- If need be, AI will use computer code, OCR and image recognition, and specific methods (like reading MRI outcomes) to augment the research.
- Then AI will arrange the presentation, including relevant images, graphics, and tables.
- AI allows further discussion to explore details as you continue your research.
- TIP: Try this “Save the above in a PPT format”
Recent AI research developments have shown impressive results. I quote my friend Dori Stein:
“I tested the 3 AI research tools for competitive analysis, and one blew the others away. The results shocked me — especially ChatGPT Pro.“
After accessing ChatGPT Pro, I conducted head-to-head tests with Google’s AI and Perplexity on deep competitive research. All three tools were tasked with gathering information for decks on PR, blog posts, pricing, partners, hiring, management, features, and competitors.
Here’s what he discovered:
- Google AI impressed with unique sources and competitors the others missed. This advantage stems from its Google Search integration versus Bing. I didn’t spot hallucinations, though I didn’t do exhaustive verification.
- Perplexity works well for focused, specific research tasks (like calculating a bottom-up TAM) but dropped the ball on multi-faceted deep research. It didn’t demonstrate hallucinations in my testing.
- ChatGPT Deep Research was mind-boggling. It is far more comprehensive than the others, with exceptional depth and thoughtfulness on each subject. I did spot some hallucinations (I hope for fixes soon), and the quality directly corresponded to prompt detail—vague prompts yielded vague research, while detailed prompts produced gold. I was impressed by the 20-minute per query approach with follow-up questions and visible reasoning.
Day to Day Re(search)
A widely accepted definition of scientific research involves systematically exploring and analyzing materials and sources to ascertain facts and draw new conclusions (Oxford Dictionary). The scientific method encompasses essential components such as the research question, hypothesis, method, and plan.
When you look deeper, you will find many more ideas related to research, such as bias, double-blind studies, reproducibility, peer review, statistical significance, falsifiability, control groups, placebo effects, and meta-analysis.
AI for scientific research is a topic for another post (or a book). For this post, let me focus on day-to-day research: what car to buy, what country to visit, what “checks and balances” across nations are, what my career will look like… etc.
Humans Doing Research: Changes
Research has become a staple outside academia in recent years as knowledge has become a key factor in our doings. In our work life, we do user, market, and competitive research. In our private life, we research careers, health, and family — all inviting research areas for burning questions.
Over time, people will transition from simply searching for answers to conducting structured research projects.
Ultimately, actual research will be done with AI (in the same way that digital search engines like Google replaced going to the library). Yet with great power comes great responsibility. Key points to remember:
- Research is not search.It is much more time-consuming to define the questions, evaluate the answers, and implement the outcomes.
- Deep understanding that comes from details. When we have a 20-page report in 20 min, a few things stay in our brain. When things go wrong — and they will go wrong — we have no idea how to bypass or overcome changes.
- Distinguishing Facts from Fiction. Issues like misinformation, biases in AI-generated research, and transparency in sources must be addressed.
- Critically evaluate the logic—does it align with your understanding? The ability to critically analyze AI-generated research will be a key skill, ensuring that human judgment remains central in drawing meaningful conclusions.
The positive side is that we will have much more time to think about the questions, and perhaps at the same time, it took us in the past, we will be able to do twice as much research.
