Doubao's newly released Deep Thinking makes AI search more like human thinking

Written by
Iris Vance
Updated on:July-08th-2025
Recommendation

Doubao's "deep thinking" function makes AI search closer to human thinking.

Core content:
1. The evolution of AI search tools, from "search first and think later" to "think while searching"
2. Actual case analysis of Doubao's "deep thinking" function
3. The advantages and value of "deep thinking" compared with traditional AI search

Yang Fangxian
Founder of 53AI/Most Valuable Expert of Tencent Cloud (TVP)

 

Doubao’s newly launched Deep Thinking seems... not so “artificially stupid” when used?

When AI search tools such as ChatGPT, Perplexity, and Mita first came out, they all said they would overturn traditional search. What was the result? I excitedly threw a complex question to the AI, and it searched hundreds of web pages, but the answer it gave me was still a little lacking.

Either the information is too fragmented, like piling up data; or the AI ​​fails to grasp the key points and gives irrelevant answers. I always feel that AI is just mechanically transferring information and does not really understand my questions, let alone think like a human.

Just like... just like those perfunctory workers in the workplace, the reports they submit seem to have a lot of content, but in fact they do not solve the core problem.

However, this cannot be entirely blamed on AI. After all, when we search for information online, we are easily confused by all kinds of information and lose sight of the key points.

AI developers are also aware of this problem. Recently, many AI search tools have begun to upgrade, striving to make AI smarter and think and search more like humans.

Doubao’s new “Deep Thinking” feature claims to allow AI to learn to “think and search at the same time.”

From “search first and think later” to “think while searching”, is AI search really going to become smarter?

Previous AI search tools, although using AI technology, are essentially still the old way of "search first, think later". Just like a diligent "porter", they first search for a lot of information from the Internet, then simply organize it and hand it over to you.

Doubao's new version of "Deep Thinking" is different. It wants AI to think deeply and reason while searching. In simple terms, it allows AI to think with questions like humans, and constantly adjust search strategies according to the progress of thinking to find more accurate and in-depth answers.

Sounds a bit abstract? No problem, let’s put this into practice with a few examples.


Case 1: Stock Market Analysis

I asked Doubao: "How do you evaluate the A-share market on March 28, 2025?"

Let’s first take a look at Doubao’s deep thinking chain.


The "thinking process" that Doubao showed when answering this question really surprised me.

It does not simply search for "A-share market on March 28, 2025" and then give the data. Instead:

  1. 1.  First, divide the characters according to the goal,  determine the questions to be answered, and divide the tasks according to the goal.
  2. 2.  Perform a preliminary search  to obtain basic information on the A-share market on March 28, 2025.
  3. 3.  Dig deeper  to further search for missing information.
  4. 4.  Make inferences based on the available information and  finally draw conclusions.

This "think while searching" model makes the AI ​​search process more like the thinking process of the human brain, more flexible and in-depth.

Let’s take a look at the answer given by Doubao after deep thinking.

First of all, because the data comes from real-time search network information, it is accurate.

Then, it provides real-time data based on different dimensions, from the overall to the local, from market conditions to capital flows, from external markets to internal policies, from technical aspects to market sentiment.

Finally, it also gives market outlook and operation suggestions like an investment consultant, not only telling the results but also giving guidance.

For comparison, I turned off Deep Thinking and asked the same question.

It can be seen that after turning off "Deep Thinking", Doubao still searches for real-time data through the Internet to ensure the accuracy of the data, but the answers are obviously simplified, lacking logic and depth, and are more like a simple summary of information.

With this comparison, we can clearly feel the importance of "thinking like a human".  The blessing of "deep thinking" makes AI's answers more refined, more intimate and more valuable.


Let’s look at another question, which is about suggestions for traveling in Harbin.

I asked Doubao: What suggestions do you have for traveling to Harbin? My kids want to see snow. Find a plan with the best value for money. The thinking process and answer are as follows.


As you can see, Doubao is like a caring travel assistant. It first analyzes my travel goals, then gradually conducts in-depth analysis and gives a complete travel plan, including time, transportation, accommodation, attractions, activities, food, budget, etc. It also gives caring suggestions on warming measures and itinerary arrangements.

This "thinking-driven" search method makes me feel that AI is no longer a cold tool, but more like a warm assistant that can understand my needs.

Doubao's "deep search" experience website:

https://www.doubao.com/chat/

Remember to turn on the "Deep Search" option.


So, how is this search-based deep thinking model different from openAI’s previous DeepResearch?

DeepResearch also processes and analyzes various types of information from the Internet, including text, images, PDF documents, etc., breaking a complex task into small pieces, and then solving them step by step, and finally generating a research report based on the analysis results. However, DeepResearch requires more computing power and takes a long time to generate reports. It is more suitable for rigorous academic research, fast and efficient market analysis, and complex data processing.

In my opinion, Doubao's deep thinking mode is more suitable for medium and low intensity needs. For example, the above travel decisions, stock market analysis, mobile phone purchase suggestions, etc. If these problems are solved using conventional AI big models, the answers will often be not real-time enough due to the lag of data, and even produce a lot of AI hallucinations. Through the AI ​​thinking mode based on real-time data, the accuracy and real-time nature of the answers can be guaranteed while improving efficiency and liberating manpower.

Technology visionary Ray Kurzweil once predicted that future searches would think like people, not index like machines. Now it seems that this prediction is coming true.

The emergence of “think while searching” has made AI search no longer a simple keyword matching and information piling up, but it has truly begun to understand human language and needs, and help us solve problems like a smart assistant.

In the future, AI search may evolve into a more powerful "intelligent agent" that, like a real human assistant, automatically completes things that we want to do in our minds but have not expressed clearly.