Today we're talking about the wonder weapon Deep Research. I think it's one of the best features in almost all AIs out there today. And I'd also like to explain to you, now that it's been on the market for a few weeks, I made a video about it a few weeks ago when it was released by OpenAI, to compare the different providers. And I'll show you a method that you can use to ensure that you always that you get a significantly better result in Deep Research, how it works, what you can do with it. That's what today is all about. So maybe very briefly, but to put it in context. Deep Research is not something new, but every provider, We've compared ChatGPT, Perplexity, Gemini and Grok, each of them has Deep Research as an additional tool, as a so-called agent. Only Grok calls it Deep Search and Deeper Search, but that's just Grok. Otherwise, you'll find it at the bottom of each tool. When you open the chat, you have to check this little box with Deep Research. That's what it's all about. But first, what did we do differently? Because usually you just start by typing in a prompt and triggering the research. And the simple, easy trick, which should always be the first step, is this. It's as simple and basic as it gets, but I don't see anyone doing it that way. You're using a completely new chat, ideally a reasoning chat, so with ChatGPT you go to o1 or you do the reasoning function in other chatbots, that you use, it doesn't matter. And put it in there, and I just did this here, for example, hey, I want to do some research on the introduction of GenAI in companies in Germany. Where are these companies? Helps me to formulate this search prompt. In that style, you're my search agent, that's your job. And I don't know what I should consider. Help me think about what aspects need to be and so on, send it off. And the reasoning model will help you make a really long, comprehensive search prompt. It can be one of those boards. And that's exactly what we've taken and put into the various tools. So, why is that so important, why do so few people do it? You can see from the very first tool, ChatGPT, that the assumption is that ChatGPT is designed as an agent, as a deep research agent, that it asks you a question, many people think that I don't have to write a good prompt, because ChatGPT will ask me anyway. But it makes a huge difference. Because if it's already a board like this, you'll get much more specific queries. In this case, ChatGPT asked whether the roof space is relevant, B2B, B2C and two or three other questions that were really good. Also about the quality of the sources, which I then narrowed down again, because I hadn't thought about doing that. Perplexity, for example, doesn't do that. That means that with Perplexity, the research starts straight away. In other words, the better you are at the prompt, the better your research will be afterwards. Google in turn creates a research plan for you. You can also edit it and go in. I just don't find it quite as intuitive as answering a query on ChatGPT. And with Grok, I was honestly a bit surprised, whether I should use deep search or even deeper search. Then I just turned on deeper search. Without SuperGrok. But let's see what happens. What I didn't want to forget is that we have new dates and places for our AI Office training. And for now, cities Berlin, Stuttgart, Munich and Zurich in the store at ai-officer.io. With YOUTUBE150 you also get a 150 euro discount on the AI Office training. Why should you do all this? On the one hand, it makes sense in light of the EU-AI Act, because having expertise in the company on this topic is definitely a requirement. We've had really good feedback on it and that's why we have more things in our store, like Crash Course for Copilot and also ChatGPT. But above all, this is about ai-officer.io. If you say you want to have the expertise in your company, you'll get on-site refueling from our team together with HÄRTING. And the whole thing on ai-officer.io. So, pragmatically speaking, in order, who finished first? Because the exciting thing about deep research as an agent is that the model gets more latency to do something. That's the special thing. That means you also want as much time as possible. It doesn't have to be fast. That's the exciting thing. You want it to go as deep as possible and research everything. I'm happy every time I make it, crack the maximum 30 minutes it has on ChatGPT. It's similar with Perplexity's Deep Research. In fact, in this case ChatGPT was the first to finish. I can't remember how many sources. Shortly after that, Gemini was finished. Then it took quite a while until Perplexity was finished. And Grok is still looking. But let's take a look at the quality of the results first. Because that's what it's all about in the end when I do something like this. And it has to be said, as with all AI topics, I have to look again to see what sources were tapped into. What were the questions that the model used. And I would always treat it as more of an 80% draft that I see. That is, with ChatGPT, it's like, it's very nicely readable. It has directly, you could see that in the introduction, structured the report in such a way that what was important to me what I had already put into the chat, so to speak, in the prompt, that this was also emphasized directly at the top. That was the biggest difference for me compared to Perplexity, for example. Perplexity started very strongly, working with the sources it found. And also evaluated them differently. I don't know exactly why Bitkom was rated better now than the McKinsey study. The models only know that themselves. I said in both cases that they should be recognized sources. I would now consider both to be recognized sources. That fits perfectly. The structure of Gemini was a little different again. A bit unstructured at the top. But then came back with very good ideas at the bottom. You could also see the reading flow, there was a table built into the topic, how do I build an AI Center of Excellence. Grok is still searching. All three were in the results so I would say, okay, I would run at least two of these deep research. So, for example, if you guys say you're Gemini users, because you have the Google account, you're using Gemini and Perplexity. If you say you are a heavy user in Chat-GPT, you use Chat-GPT and Perplexity. Or Chat-GPT and Gemini, either way. I would combine the whole thing. And I'm not doing Grok any favors right now. But in this case today, Grok didn't have any results for us. In that respect, it can happen. We've run it a few times now. Normally it works and delivers good results. But still quite buggy here and there. Particularly when the prompt is such a slab, like the one I put in at the beginning. But that's my method now. Because this prompt simply provides a super solid, good foundation. It's like sending a person through and saying, hey, you've got a week to write a report on this topic. And you get this one within minutes. Get an overview and think about it, what do I do with it now? Remember, what we're discussing here is really more complex work. It's not just a matter of somehow finding me I don't know, pick out 10 restaurants or comparing the best travel destinations for us as a family. Which are also super exciting deep research tasks. I'm now interested in creating reports, that I can continue to work with in a business context. And I think it's completely legitimate, if you also initiate several searches and get into a different depth. And one observation that I have made not just once, but again and again, that because AI can do so much, often comes the assumption, hey, I've got an old annual report here of market data from 2023 or 2024. And then just throwing that into ChatGPT and saying, redo that for me or do it in Copilot. And that doesn't work. It doesn't work with the search function either. Theoretically, but if you do it agentically, you can say, hey, this is my old report, I do it once a year and I need to update these figures every year. Then you can use ChatGPT you can have a prompt formulated from this old report. And then you do the same and enter it into Deep Research. And it will then write you the new annual report, so to speak. That's the mode. And you just have to think about that, One is an agent, Deep Research, that can do that if it has the right prompt. And the other is basically simply a large language model. It can't just take your old report and rework it. And you have to keep that apart a bit. That's why this agentic thinking so incredibly important. And that's basically what we did briefly here in the video. Because if you break it down like this, you build the prompt first and then the report, then you're already in the middle of agentic thinking. Welcome to the future. This is my super simple deep research workflow and it has now become my most important tool tool in my working day. The agent that I say I wouldn't want to do without. And it's important to know, it is an agent. It's different from Operator or Manus or anything you see on YouTube now, that can do other things. Clicking on websites and so on. This is about special agents, that are designed to understand websites, read through PDFs and categorize them categorize them accordingly. And that's why I'm particularly pleased that you're here too, because in the YouTube comments, we are still writing ourselves and not the AI. And that's why please comment diligently. Share the video, because together AI is more fun together than alone. And of course I'd be delighted if you subscribed to the channel, because because I'll see you again in a few days.