AI as a Research Tool: Optimizing Prompting Techniques for Research

AI is widely recognized as one of the most advanced developments in technology, as it can be used to meet a wide range of users’ needs and requirements.

AI can help with a wide range of tasks, from answering questions and brainstorming ideas to developing topics, proofreading content, generating images, and supporting research.

Research is a systematic and structured process of gathering, finding, and analyzing specific information or data to obtain accurate answers or solve the problem being investigated.

Unlike a typical search for information, research needs to be conducted carefully, with reliable and credible sources, and based on data rather than one-sided opinions.

If you want to conduct deep research with AI, it is highly recommended to refine your prompting techniques and make use of built-in deep research features, such as the one available in Gemini.

Here’s an overview of the methods and examples of AI prompting techniques you can use for in-depth research (deep research).

How to Use AI for Research

Research is commonly done by academics, professionals, content writers, marketing teams, and others to find information backed by facts and evidence. This helps ensure that the information is reliable and can be properly supported, whether it is intended for public use or personal purposes.

When researching a topic and gathering information and data from the internet, people typically follow several steps, such as defining the scope of the topic, collecting references, analyzing and comparing information, fact-checking, creating an outline, and more.

The same process applies when using AI as a research tool. The main difference lies in the method. Traditional research can take more time because you have to search for data one piece at a time using different keywords. With the right prompting techniques, however, AI can make the research process much more efficient and effective.

Here are some prompting techniques you can use to do in-depth research with AI:

1. Use Layered Prompting

If you have previously learned what a prompt is and how to create detailed prompts, this section will introduce you to layered prompting, a technique that can be particularly useful for research.

What is layered prompting? It is a method for creating prompts that gradually and systematically start with general context and narrow the discussion to more specific details.

The goal of using layered prompts is to get information from AI that is accurate, detailed, clear, and relevant to your needs. However, just as you would when discussing a topic with a teacher, lecturer, or scientist, getting more in-depth answers from AI often requires you to ask questions in a step-by-step manner. When necessary, you may also need to ask AI to verify the information repeatedly.

For example, you can start with a detailed process like this:

1. Define the research topic and objective
        ↓
2. Create an initial prompt to broadly explore the topic
        ↓
3. Divide the topic into subtopics and explore them in greater depth
        ↓
4. Analyze and synthesize the information
        ↓
5. Fact-check the information
        ↓
6. Organize the research findings based on your needs
        ↓
7. Draw conclusions from the research findings

2. Check the Sources Provided by AI

One important tip for using AI tools for research is to avoid simply copying and pasting the information without checking whether the sources provided are genuine and reliable. This is especially important when using tools such as ChatGPT, Perplexity, Gemini, or DeepSeek.

Before moving on to the next step, make sure to check the sources provided by AI, whether they appear as links, anchor text linked to specific sources, or a reference list displayed below the AI-generated response.

If the links are accessible or can even be downloaded and the content matches the context of your research question, you can move on to the third step: summarizing the information you have gathered with the help of AI.

3. Upload Files and Ask AI to Summarize Them

Sometimes, AI may recommend more than one reference source. You could end up with 5-10 references, including journals, research papers, academic publications, and other sources in formats such as PDFs, website URLs, or Word documents.

Reading each source one by one can take a lot of time and effort. To make the process more efficient, you can upload these reference files using the file attachment feature available in AI tools. You can then ask AI to summarize the key points, compare the sources, or analyze them from the perspective of the AI persona you previously instructed it to adopt.

Once AI has finished summarizing the sources, extract and save the key points or main findings you need from each reference. You can then review and analyze them further.

4. Create Prompts for Evaluation

Most in-depth research requires additional analysis to determine whether the data or information is valid. By creating an evaluative prompt, you are instructing AI to critically examine, analyze, and assess the journal or research paper being discussed.

The purpose of an evaluative prompt is to help expand and challenge the researcher’s thinking. It can also be used to identify potential biases in AI’s previous analysis.

Why can bias occur? AI may sometimes provide inconsistent answers or simply generate responses that are more likely to please the user. In other words, AI-generated results can change when users repeatedly ask for verification. This can happen because AI learns from data created by humans, which may not always be neutral, lacks sufficient representative data, and does not understand ethics or emotions in the same way humans do.

Now that we’ve covered these four steps, it’s time to put AI research tools to the test and see how effective they really are for conducting research.

Example of Using AI for Research

AI tools such as ChatGPT can be used to find specific data or information based on the instructions or questions you provide.

However, it is important to remember that even though AI tools can be quite powerful, you should still exercise caution and verify the accuracy of the data and information they provide.

Here’s an example of using ChatGPT for research and what the results look like:

Here’s an example of using ChatGPT for research and what the results look like:

Based on the prompt above, ChatGPT provided information on more than 10 brands offering pure matcha without added sugar or creamer. However, as an additional reference, you may also want information about ceremonial matcha, which is generally considered one of the highest-quality types of matcha.

Don’t worry! You don’t need to revise the previous prompt. Instead, you can simply continue the conversation by sending a follow-up prompt such as “Please add some ceremonial matcha for me.” This allows AI to provide a wider range of matcha product recommendations.

Looking closely at the information generated by AI above, there are several things you should pay attention to, including:

• AI repeatedly mentions only a few brands, such as Seduh and GEON.
• Some of the recommended links lead to pages that do not actually match the brands mentioned.
• The recommended prices for chasen do not match the actual prices listed in the stores, so you need to be careful and double-check the information.

If these three points still leave you with some questions, you can take the research a step further by creating a more detailed and well-structured prompt and narrowing the research down to a more specific subtopic, as shown below.

And here are the results generated from the prompt above:

After reviewing the AI-generated results, it’s time to validate the information and check whether the data provided by ChatGPT is accurate. It’s worth noting that ChatGPT can now provide up-to-date information and data from 2026 by crawling various websites across the internet, particularly online marketplaces.

So, let’s verify the information and data presented by ChatGPT. Is everything accurate?

1. In the first recommendation, ChatGPT provided information that generally matched the user’s request in the prompt, including Seduh Permata, Javaland, and HAYS. However, while some of the links were accessible, others either redirected to general search results rather than the specific matcha products associated with the listed brands. Therefore, the AI-generated list of pure matcha products cannot be considered 100% accurate
2. The explanation of the differences between culinary-grade and ceremonial-grade matcha was accurate. However, the information about matcha tools, particularly the chasen (bamboo whisk), was not entirely accurate. ChatGPT listed chasen starting at around IDR 15,000, while standard chasen typically cost around IDR 60,000 or more. Chasen priced at IDR 15,000-40,000 are generally plastic whisks rather than bamboo ones. This shows that AI-generated information is not always fully accurate and needs to be repeatedly verified to ensure its reliability.

After analyzing and checking the reference sources (links) provided by AI for the matcha research, it turns out that the information still needs to be verified. You should also compare the information using a research agent, which can provide a more thorough approach to research.

Now, it’s time to make the most of prompting techniques and apply them to research the same topic.

Fact-Checking Matcha Using a Research Agent

Since the information generated by ChatGPT was not entirely accurate, it’s time to use a prompt that is more or less similar to the previous one. If you previously relied on ChatGPT, now it’s time to try Perplexity AI.

The prompt is as follows:

As an Indonesian, I’m a new matcha buyer and enthusiast. Since buying matcha drinks regularly can get expensive, I’d like to start making them at home. So, please help me research pure matcha and provide the following information:

• A list of affordable matcha powders priced under IDR 100,000 from Indonesian marketplaces (Shopee, Tokopedia, and Blibli), along with links to the products.
• 10 brands that sell pure matcha, including ceremonial-grade options.
• The prices of the most affordable matcha-making tools.
• Make sure all the links you provide are accessible.
• Provide the answer in English.

Here is Perplexity AI’s response:

Despite the information provided by Perplexity AI above, some details are still not entirely accurate. For example, Perplexity recommended ceremonial-grade matcha brands from overseas, meaning buyers in Indonesia cannot order them directly in the same way they can from marketplaces such as Shopee or Tokopedia.

Since this information is not entirely accurate, you can validate it again by providing a screenshot to challenge the AI’s claim, like this.

Once you’ve gathered all the research you need, the next step is to ask Perplexity to summarize the findings so they can serve as supporting information or be used for further research. It’s not too difficult, right?

Updated on August 19, 2026 by Anisa K.