At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Hi, welcome back to the research session. In the previous episode Claude searched a few pages to answer a short question. This time we face a much more complicated one. We'll use deep research: Claude explores many sources for several minutes, then hands over a full report with citations. Our example comes from Dewi's office at Maju Bersama Co.
Step 1: Quick search or deep research?
Claude has two ways to search. A quick search reads a few pages. Deep research can explore dozens of pages, searching again with new keywords every time it finds a lead. That's why the time is different. A quick search finishes in seconds, while deep research can take a few minutes. You can leave it for a while and come back when the report is ready. The results are different too. A quick search gives a short answer, while deep research delivers a structured report with many citations. So use a quick search for a fact or a single number. Use deep research for questions with many sides, such as market opportunities, comparing options, or complicated rules. To give you a picture, here are questions worth deep research. Laras wants to know whether Sunset Coffee should open a second branch. She needs data on location, rent, competitors, and buying power. Mr. Hendra from Cerdas Tutoring wants to know which study methods are proven to work, complete with studies and examples of how to apply them. Nadia, a social media freelancer, wants to choose the platform that best suits her new client. Dewi wants to know whether an extra warehouse is needed. All of these questions have many sides, and the answers are spread across many sources. That's the mark of a question for deep research. Turning it on is similar to web search. First, open a new conversation. Keep one research topic per conversation so the results stay focused. Second, choose research mode in the tools menu near the chat box. It may be called Research. The screens in this video are simulations, so menu names may differ slightly. You can check availability and usage limits for each plan on Anthropic's official pages. Third, write a complete research brief. We'll cover what goes in it shortly. Fourth, answer any clarifying questions Claude asks, and then let it work.
Step 2: Breaking down a big question
Let's meet the case. Dewi, operations manager at Maju Bersama Co., has been asked by the director to assess one opportunity: should the company start selling stationery to schools? This question is too big to answer all at once. If we ask it directly, Claude will give a general answer that sounds smart but is hard to use for making a decision. So before turning on deep research, we spend two minutes breaking the question down. Those two minutes save a lot of time reading an unfocused report. Here's how: break it into small questions. First, buyers: who decides stationery purchases at a school? The principal, the treasurer, or the school cooperative? Second, timing: when do schools usually buy? Just before the new school year, or spread throughout the year? Third, competitors: which suppliers already serve schools, and what are their strengths? Fourth, requirements: what documents or process are needed to become a school supplier? These four small questions will become the outline of the report. Claude's research stays focused, and Dewi can easily see which parts still lack data.
Step 3: Writing a research brief
Now we write the research brief, the full instructions for deep research. Start with the goal: what decision is this research for, and who will read it? Then the context and scope: what company, which kind of schools, and which region. Without a region limit, Claude might research all of Indonesia. The task is the four small questions from before, written as they are. Limits matter a lot in deep research. Dewi asks for official sources published in the last three years, and asks Claude to flag sections where the data is weak. Finally, the report format: a five-point summary, one section per question, a competitor table, and a source list. Dewi chooses research mode and pastes in the brief. On screen we only see its opening sentence. Notice that Claude doesn't set off right away. It asks two clarifying questions: are uniforms and books included, and does Dewi want an estimate of spending? Dewi answers specifically: stationery only, and estimates are fine as long as they're marked. A short answer like this keeps the research from drifting into things nobody needs. Deep research uses more of your usage allowance than a normal chat, so make sure the brief is right before you start. You can check the limits for each plan on Anthropic's official pages.
Step 4: Running it and reading the report
The research begins. On screen we see the first search, about stationery procurement at private schools. Then Claude searches with new keywords about school purchasing calendars. This is the difference from a quick search: every finding triggers the next search. The third search is about the requirements for supplying a school cooperative. After a few minutes, Claude has read and compared dozens of sources. The number thirty-four is just an example; it differs with every research run. The report is ready. Claude honestly flags two sections where the data is weak. Flags like these are actually very valuable for Dewi. The research report appears in the right panel. Dewi asks for the summary and the competitor table so she can read it quickly. The summary answers the four small questions, each with source numbers. One point is honestly marked as an estimate because the data is limited. Below it is a competitor table with each competitor's strength and source number. Everything here is an example, not real market data. How to read a research report: the summary first, then the sections flagged as weak, then open the sources for numbers you'll use in a decision.
Step 5: Following up on the research
After reading the report, Dewi follows up on the weak section. She asks: how can we gather the spending data ourselves? Claude gives three practical ways, from short interviews with school treasurers to a small trial package, and offers a list of interview questions. Data that isn't online really does have to be gathered in the field. Dewi accepts the offer and sets limits: short, polite, ten minutes at most per school. Claude builds the interview guide in the right panel. The questions go from easiest to answer, and the budget question is placed near the end, when the person is more comfortable. The online research and the interview results will later be combined into one report. We'll cover how to write a tidy research report at the end of this session. Before using a research report, check five things. First, the summary answers all the small questions we asked. Second, sections flagged as weak or as estimates are noted, so they aren't read as facts. Third, the numbers used for the decision have been checked against the original source. Fourth, there's a next step for data that isn't available yet, like the interviews. And what you must not do: copy the whole report as the final decision. A research report is input for a decision. Dewi and her director still make the call. Use the report to see the big picture and choose the next step.
Done!
We've learned to tell a quick search from deep research, break down a big question, write a research brief, read the report, and follow up on the weak sections. A research report contains many claims and sources. In the next episode we learn how to check facts and judge sources quickly. See you there!
Practice exercises
Try these on your own after watching to lock in what you learned.
Write one big question from your work and break it into four smaller questions
Write a research brief with a goal, scope, preferred sources, and result format
Run deep research, then ask two follow-up questions based on the report
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,278 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
