Beyond Google: Harnessing the Power of ChatGPT for In-Depth Research

If you want to make ChatGPT sing, you've got to learn to dance with it

Apr 22, 2023

This is a long-ish post about how to effectively use ChatGPT for conducting introductory research into a given topic of your specification. There’s a bunch of background knowledge required for you to effectively use these techniques, so the first parts of the post may not appear to have anything to do with research. Patience is a virtue, as they say.

Introduction

A lot of people approach ChatGPT as they do Google: they enter something they’re searching for, and they sit back, expecting the computer to do all the work. And you can use ChatGPT in that way. Sometimes, you’ll even get a good response! But the majority of the time, you’ll end up disappointed if you use ChatGPT as you use Google.

Here’s a different way to think about using ChatGPT: think of it as a colleague, or even an intern, with whom you are interacting. When you start to think about interacting with ChatGPT in the same way as you think about interacting with a colleague, a whole new world of prompt possibilities opens up. You can now consider iterative, interactive prompts, in which you provide a seed prompt, ChatGPT responds by asking for additional information, you provide additional information, ChatGPT iterates upon your seed prompt, it seeks additional information, etc.

You and ChatGPT can repeat this iterative cycle indefinitely, improving your prompt—and therefore ChatGPT’s ultimate output—over time. When you provide ChatGPT with a simple, unbounded, prompt, it responds with simple, generic words. When you provide ChatGPT with a complex, tightly constrained prompt, it responds with comprehensive sophistication. It’s really quite magical once you learn this skill.

Enter the iterative prompt

I previously discussed how to engineer iterative prompts, here. When using this strategy, the initial prompt sets the stage for your interaction with ChatGPT: it provides ChatGPT a set of overarching instructions to which ChatGPT can repeatedly refer as it iterates through prompt optimization.

So here’s the prompt:

I want you to help me create a great prompt so that you can provide the best output for me. I’d like you to execute the following steps:
Ask me what the topic of the prompt is.
After I submit my prompt, generate 3 sections: (a) your revised prompt; (b) suggestions on how to improve the prompt; (c ) clarifying questions.
Continue this iterative process, by asking me for additional input, until I am satisfied with the prompt.
After three iterations, ask me whether you should execute the prompt or whether we should continue iterating. Do this for every iteration after the third iteration.

Cool, so what does this prompt do?

We can observe several things about this prompt:

It’s complex and tightly constrained.
However, the prompt allows for a nearly unbounded set of outputs to flow from it!
It’s logically structured in the way that computer code is logically structured.
It contains an escape hatch: after the third iteration, ChatGPT will ask you whether it ought to execute the latest prompt, or whether it should iterate through another cycle of prompt optimization. Critically, this escape hatch persists after the 3rd iteration. This means you don’t get stuck in an infinite loop of prompt optimization.

Great. What does this have to do with research? Gimme the research deets!

OK, so let’s go back to the initial prompt which I provide above. I’ve entered that into ChatGPT, and it replies: Sure, I’d be happy to help you create a great prompt. To begin, please tell me what the topic of the prompt is.

Me: I need to write a market research report for the mobile communications market. I want you to generate an outline for me and provide links to relevant research.

ChatGPT:

(a) Revised Prompt: Please provide an outline for a market research report on the mobile communications market, along with links to relevant research sources.

(b) Suggestions to improve the prompt:

Specify the region or country you’re focusing on for the mobile communications market.
Indicate the time period for which the market data should be relevant.
Include any particular aspects of the mobile communications market you’d like the report to cover, such as technology, competitors, trends, or regulations.

Is there a specific region or country you’d like the report to focus on?
Do you have a preferred time period for the data and research included in the report?
Are there any specific aspects of the mobile communications market you’d like to prioritize in the report?

[Interlude]

Rather than provide you with the rest of this iterative process, I will provide you the final prompt that ChatGPT and I collaboratively designed:

Please provide an outline for a market research report on the mobile communications market in the United States over the past five years, focusing on the deployment of 5G and successor technologies, a comparison of Verizon and its competitors, and an analysis of the impact of regulations on the market, along with links to relevant research sources.1

As you can see, this is a much more complicated and constrained prompt than the original one.

You might be wondering at this point: “This is easy. Why do I need this complicated iterative prompt optimization process when I can just construct a complex and constrained prompt to begin with?”

And, if you’ve got that skillset, you’re right: you don’t need iterative collaboration with ChatGPT. You can just jump into robust prompt creation and have ChatGPT generate good output for you. But if you’re new to prompting a chat bot, this provides a robust structure for you to improve your prompt engineering skills.

The final output

What does the output look like? Well, check this out:

Introduction
1. Overview of the mobile communications market in the United States
2. Scope of the report: technology, competitors, and regulatory analysis
5G Deployment and Successor Technologies
1. Network coverage and infrastructure investment
  1. Progress of 5G deployment in the United States
    1. Research source: https://www.statista.com/statistics/1106538/5g-coverage-share-usa/
  2. Infrastructure investment and key players
    1. Research source: https://www2.deloitte.com/us/en/insights/industry/technology/5g-network-infrastructure.html
2. Consumer adoption and market penetration
  1. 5G device sales and subscriptions
    1. Research source: https://www.counterpointresearch.com/us-5g-devices-sales-to-grow-50-in-2021/
  2. Successor technologies and their potential impact
    1. Research source: https://spectrum.ieee.org/6g-and-beyond
Verizon and Competitors Analysis
1. Market share and financial performance
  1. Verizon
    1. Research source: https://www.verizon.com/about/investors/annual-reports
  2. Major competitors (AT&T, T-Mobile, and others)
    1. Research source: https://www.statista.com/statistics/199359/market-share-of-wireless-carriers-in-the-us-by-subscriptions/
2. Strategies and competitive advantages
  1. Network quality and coverage
    1. Research source: https://www.rootmetrics.com/en-US/content/mobile-performance-in-the-us
  2. Pricing, plans, and promotions
    1. Research source: https://www.whistleout.com/CellPhones/Guides/best-cell-phone-plans
Impact of Regulations on the Mobile Communications Market
1. Key regulations affecting the market
  1. Federal Communications Commission (FCC) spectrum allocation
    1. Research source: https://www.fcc.gov/auctions
  2. Net neutrality and its implications
    1. Research source: https://www.brookings.edu/research/net-neutrality-facts-on-the-open-internet-order/
2. Analysis of the impact of regulations on market competition and innovation
  1. Effects of spectrum allocation on market players
    1. Research source: https://www.ctia.org/the-wireless-industry/positions/spectrum-policy-and-the-future-of-5g
  2. The role of net neutrality in shaping the market landscape
    1. Research source: https://itif.org/publications/2021/04/05/impact-net-neutrality

Conclusion

This is a powerful way to get the most out of ChatGPT.

If market research doesn’t interest you, you can put literally any other research project into the prompt, and have ChatGPT generate a robust outline for initial research.

A final note of caution: it’s on you to determine the relevancy and accuracy of any of the information provided at the links that ChatGPT surfaces.

I note, for the record, that it is ChatGPT, and not me, who added “please” to the prompt. Try as I might, I can’t bring myself to say “please” to a bot.

Buy the Rumor; Sell the News

Discussion about this post