🔎 Grok DeepSearch vs ChatGPT DeepSearch vs Gemini DeepSearch
15:44 Share

Welcome everyone to new special deep dive here on AI Unraveled. This show, as always, is created and produced by Etienne Newman. He's a senior engineer based up in Canada and also a very passionate soccer dad. And hey, if you're getting value from these explorations into AI, we'd really appreciate it if you could take just a second to like and subscribe to AI Unraveled over on Apple Podcasts. It genuinely helps other folks find the show.

Okay, so today we're tackling something I think many of you are grappling with. When you need to really dig deep into a topic using AI, how do the different tools stack up? We're zoning in on the deep search capabilities or lack thereof, sometimes of Grok, ChatGPT, and Gemini.

And we're basing this on a really interesting discussion we found over on the artificial subreddit, real users talking about their experiences. Yeah, it's such a relevant question now, isn't it? I mean, we're all just flooded with information constantly. And the big promise of AI is helping us

cut through that noise, efficiently pull out the important stuff. So whether you're doing research for work, maybe a big presentation, or just pursuing a personal interest, or even just trying to understand something complex, knowing the strengths and frankly the weaknesses of these AI search tools is super useful. It's really about getting the knowledge you need without feeling totally overwhelmed. Exactly. Getting those key nuggets, like you said. And it's interesting because Grok specifically calls its feature deep search.

But, you know, ChatGPT and Gemini, they offer pretty comparable ways to do that kind of in-depth web research, even if they don't use the exact same name. Right. Different paths to hopefully the same goal. Yeah. Thorough AI driven research. Yeah.

So, yeah, our mission today is really to unpack what real users are actually finding when they put these tools to the test. Let's start with Grok. It has that explicit deep search mode. What did the Reddit users notice about its approach? Well, Grok really leans into real-time data. It pulls heavily from X, you know, what used to be Twitter along with the wider web. And the idea behind its deep search is what they call an agentic process.

Basically, it tries to act like a little research agent going out, checking multiple sources and then trying to compile a detailed report for you. OK, an agentic process. Yeah. Sounds sophisticated. It does. And one thing that really stood out from the Reddit thread was users found Grot particularly good for topics that weren't super technical or, you know, didn't involve dense legal jargon or weren't.

as one user put it, overly intellectual. So maybe better for like understanding general opinions or trends rather than complex scientific papers? That seems to be the implication, yeah. It suggests Grok might be quite good at capturing sort of the pulse of more informal discussions or current events happening on platforms like X. That makes sense with the X integration. That real-time angle could be its edge for, you know, breaking news or public sentiment. Maybe faster, if not always as deeply vetted as other sources.

And speaking of Grok's unique aspects, this user highlighted something as they really liked or maybe like the lack of. Ah, yes. The cookie cutter guardrails, as they called them. Exactly. They mentioned that except for obviously explicit stuff, Grok seemed less restrictive. They actually used the phrase truly unbridled.

and even said they'd be willing to pay for that less filtered kind of access. Yeah, that's a strong statement. It points to a desire some users have for AI tools that feel less curated, less sanitized, perhaps. They appreciated fewer built-in limitations on what they could ask or what grok might surface. It's an interesting trade-off, though, isn't it? That unbridled access. Absolutely. It definitely raises questions about reliability. If it's less filtered, you might get unique insights, but you also run a higher risk of encountering

well, misinformation or just less credible sources. Navigating that is key. Good point. Yeah. And the user did also mention a potential drawback to Grok's approach, even beyond the content itself. Right. The speed that came up. Grok's deep search was noted as being potentially slower compared to the others. Okay. So maybe that agentic process takes a bit more time to run? Possibly. Yeah. And another point was depth.

For certain tasks, especially ones needing, say, serious academic rigor or really detailed analysis, some users felt Grok's output wasn't quite as comprehensive. So maybe strong on breadth in real time,

but potentially less deep for specialized research. That seems to be the takeaway from those comments. Yeah. Which actually leads us nicely into ChatGPT. Right. Now, ChatGPT doesn't have a big deep search button plastered on it. So how does it handle this kind of deeper research task? Good question. Yeah. No dedicated button, but it definitely has robust web searching built in. It uses the Bing index, plus OpenAI has its own web crawlers gathering data.

And a really significant improvement lately, I think, is its citation capability. Oh, yeah. How so? Well, it can now cite multiple sources for claims it makes, and often it'll even highlight the specific text snippets it pulled the information from. That's a big step for transparency and letting you verify things yourself. That is huge, being able to track back the source easily.

So what did the Reddit folks say about using ChatGPT for this kind of deep research? Well, interestingly, one key user initially considered ChatGPT the gold standard, the best overall tool for deep research tasks. The gold standard. OK, high praise. Why was that? The killer feature, according to them, was its ability to let you upload a pretty significant amount of your own material, your own documents, and then you can instruct ChatGPT

to work with that material, analyze it, synthesize it based on your specific instructions. They described it almost like a normal prompt on steroids with real heavy follow through. - Wow, okay, so you can feed it your own context, your own data, and have it researched based on that combined with the web. That sounds incredibly powerful for specific projects.

Exactly. It opens up possibilities for really tailored, specific analysis that just a general web search can't really do. You're combining your knowledge base with the Internet's vastness. But there's always a catch, right? Did they mention any limitations? There was one, yeah. The credits?

Apparently, this upload and analyze functionality uses up credits, and the user mentioned the amount was somewhat limited. Ah, okay. So maybe not something you can rely on for constant heavy use without paying attention to usage limit. Precisely. And there was another interesting point of friction mentioned. Oh. Yeah. Some frustration about a perceived dip in the quality of ChatGPT's research results more recently. It seemed to coincide with what the user called reticence.

rather dismissively, some two-type research auto-select nonsense. Huh. Two-type research auto-select nonsense.

Doesn't sound like they were a fan. Any idea what that actually means? Not entirely clear from the comment, honestly, but it suggests maybe some behind the scenes change in how ChatGPT chooses its search strategy. And for this user, at least, it felt like a step backward in quality or control. Interesting. It shows how these tools are constantly being tweaked and not always in ways users prefer. So overall, where did ChatGPT land in the comparison? It seemed to sit kind of in the middle ground, you know, seen as clickbait.

Closer perhaps to Gemini in some ways, but crucially without what was described as Gemini's, let's say, tendency towards infinite and admittedly a bit annoying verbosity. OK, that's a perfect segue. Let's talk Gemini, Google's offering. How does it approach deep search?

Gemini's big advantage is its tight integration with Google search itself and also Google's AI overviews feature. This gives it arguably incredibly fast access to a massive, constantly updated pool of information, real time indexed web data.

Makes sense. Leveraging their core strength. And obviously it probably plays nicely within the whole Google ecosystem workspace and all that. Exactly. So for the Reddit user we've been following, what was their experience with Gemini's deep research? Well, this is where it gets really interesting. For programming related questions, Gemini was apparently the clear winner for this user by a long shot. Yeah, they said it, and I quote, "Bi-R had the best answers to all my questions, from designs to library searches to anything else."

That's a very strong endorsement, especially for technical fields. Wow. Okay. So for coding, development, technical documentation, Gemini really shone for them. That suggests it's maybe very good at tapping into that vast amount of technical info Google indexes. It seems so. But interestingly, they also mentioned that their initial experience with Gemini's deep research capability wasn't actually that great. Oh, it got better. Apparently, yes.

They felt there was a really significant improvement after Google rolled out newer, more capable underlying models. They mentioned the 1.5 model specifically. That just highlights how fast things are moving, doesn't it? A tool might be mediocre one month and significantly better the next just due to an underlying model update, constant evolution. Absolutely. Now, were there any downsides mentioned for Gemini? We heard about the potential verbosity. Yes, that was the main one. The output could be just bad.

overly wordy, using more language than strictly necessary, which might mean more time spent digging up the key points.

OK, so potentially thorough, especially on technical stuff, but you might have to wade through a bit more text. Seems like it. And another point echoing what we heard about ChatGPT was the desire to upload their own documents. Ah, that feature again. Seems like a common request. Very much so. The ability to ground the AI's research in your own specific data or context seems highly valued. They wish Gemini had that capability like ChatGPT does. Makes sense.

Anything else on Gemini? Yeah, a couple of other interesting nuggets. The user was actually surprised that Gemini didn't seem to have any obvious limits on the number of deep research queries they could run. Oh, unlike the credit system mentioned for ChatGPT's upload feature. Exactly. That could be a big plus for people doing a lot of research. But on the flip side, they pointed out a rather baffling omission. Which was? You apparently can't easily search through your past conversation history within Gemini. Wait, really? From Google?

The search company. I know, right? Seems odd. You'd expect robust history search to be standard, especially from them. A definite point of friction for that user. Okay, so we've covered the big three, Grok, ChatGPT, Gemini, in some detail based on these user experiences.

Now, before we synthesize, I know you mentioned wanting to talk about something important for listeners looking to deepen their own tech skills. Ah, yes. Perfect timing. Because understanding these tools is one thing, but actually mastering the underlying technologies, the skills needed in AI and related tech fields, that's crucial too. And for anyone listening who wants to really take that next step, maybe get certified in high demand areas, Etienne Newman, our producer, has actually created an incredible resource. That's right.

We wanted to mention the GM Cat Tech app.

It's AI-powered and specifically designed to help you master and, importantly, ace over 50 different in-demand tech certifications. It's really comprehensive. Etienne's put his engineering expertise into making it effective. It's not just about reading material. Jenga Tech includes things like PBQs, performance-based questions. Which simulate real-world tasks, right? Exactly. Plus quizzes, flashcards for quick review, hands-on labs for practice, and even full simulations to really test your readiness. It covers a lot of ground.

So if you're thinking about boosting your skills, getting certified, definitely check out the Jenga Tech app. We'll mention it again later, but it's a great tool for Metzian. Okay, back to our AI comparison. The Reddit thread also touched on a few other tools, didn't it? Briefly. It did, yeah. Just quick mentions mostly. One user brought up a tool called O3, said they found it better and faster than the main three, and notably not overly verbose. So maybe one to watch. Okay, O3. Good to know. What else?

Perplexity and You.com were mentioned, but the original poster seemed generally unimpressed with their deep research features specifically. So less positive feedback there, at least from that user. Right. Claude also came up. One user reported getting consistently great answers from Claude, specifically mentioning the Sonnet 3.7 model for deep research and web searches. Sounds promising. But another user chimed in with a less positive experience with Claude.

Mentioned some inconsistencies, maybe concerns about business practices or pricing changes. So a bit of a mixed view there. Okay, so Claude is in the mix, but maybe experiences vary. Anyone else? Deep Seek was also mentioned, mostly in the titles of related discussions on the subreddit comparing it to ChatGPT and Grok.

sometimes focusing on cost-effectiveness for developers so it's definitely another player in this space it really shows how dynamic this field is lots of options constantly shifting absolutely it's not just the big three anymore so if we try to pull this all together yeah synthesize the key differences and takeaways from this reddit field exploration what stands out okay well based on these user accounts grok seems to be finding its niche with that real-time X insight and the

the less filtered approach, maybe best for current events or general sentiment.

but potentially slower and less deep for rigorous stuff. Right. And chat GPT. Still a very strong all-rounder, it seems. That ability to upload your own documents is a major plus for customized research, but you have those credit limits and maybe some recent wobbles in quality or user control that people noticed. Gotcha. And Gemini. Gemini appears to really excel on the technical side, particularly programming questions, leveraging that Google search power. Fast access to info.

The main downside seems to be that potential verbosity and the lack of document uploads. So it really sounds like the best tool is, well, it depends. Exactly. It totally depends on what you, the user, actually need. What kind of information are you after? How deep does the analysis need to go? Is speed the absolute priority? And is integrating your own documents a must-have feature? Yeah. All those factors play in. Precisely.

The big picture, though, is that all these major AI players are moving beyond just spitting out static knowledge. They're all integrating real-time search, trying to synthesize information from the live web. And they're doing it differently, right? Using different data sources, different methods. Yeah, which leads to these distinct strengths and weaknesses we've been discussing. It's a fascinating divergence watching how each platform evolves. It really is. And it underscores the need for skills, not just in using these tools, but understanding the tech behind them.

Which brings us back. Yes, to the Jamga Tech app we mentioned. If this discussion has you thinking about leveling up your own technical skills, maybe in AI, cloud, cybersecurity, or networking, seriously consider checking out Jamga Tech. With its focus on practical learning through those PBQs, labs, simulations, quizzes, flashcards,

It's designed to help you truly master the material for over 50 certifications. It's a way to move beyond just understanding AI concepts to actually gaining certifiable in-demand skills. A really valuable resource created by Etienne Newman, so do look up the Jamka Tech app. Well, this has been a great deep dive, unpacking how these AI search tools are performing in the real world based on user experiences. And as these tools keep getting better, faster, maybe weirder, it really makes you think, doesn't it?

How is the whole nature of research, of how we find and absorb knowledge going to change in the next few years? That's the big question, isn't it? How does this transform learning, discovery, staying informed? And which of these AI assistants, or maybe one we haven't even seen yet, will end up being the best fit for your own needs as things evolve? Something to ponder.

Definitely food for thought. Well, thank you so much for joining us for this deep dive on AI Unraveled. If you found this valuable, please, one last reminder, do like and subscribe on Apple Podcasts. It helps us keep these conversations going. We'll be back soon with more explorations into this ever-changing world of AI. Thanks for listening.