WEBVTT 00:00:00.000 --> 00:00:03.520 You know, shopping online, it used to feel simple, 00:00:04.099 --> 00:00:07.040 almost effortless. But now, more often than not, 00:00:07.059 --> 00:00:10.439 it feels less like browsing and more like, I 00:00:10.439 --> 00:00:13.019 don't know, a really hard, unpaid job. It's the 00:00:13.019 --> 00:00:15.099 overwhelm, right? You want to buy something simple, 00:00:15.140 --> 00:00:17.199 maybe a new vacuum cleaner, and suddenly you've 00:00:17.199 --> 00:00:19.699 got 20 tabs open in your browser. You're trying 00:00:19.699 --> 00:00:22.480 to compare Amazon reviews with some YouTube durability 00:00:22.480 --> 00:00:25.239 test. And then you're down a rabbit hole on Reddit 00:00:25.239 --> 00:00:29.440 about the exact decibel level of the motor. Density 00:00:29.440 --> 00:00:32.219 of information it just brings on this weird fatigue 00:00:32.219 --> 00:00:35.200 the dizziness is real and it usually ends with 00:00:35.200 --> 00:00:37.820 the worst possible outcome You just turn off 00:00:37.820 --> 00:00:40.539 the computer you buy nothing. Oh, I have absolutely 00:00:40.539 --> 00:00:42.640 been there I think I spent three hours last week 00:00:42.640 --> 00:00:45.619 researching ergonomic mice pads I ended up buying 00:00:45.619 --> 00:00:49.630 a cheap flat one out of sheer spite that feeling 00:00:49.630 --> 00:00:51.570 that paralysis is what we're trying to avoid. 00:00:51.909 --> 00:00:54.810 And the good news is that's now possible. There's 00:00:54.810 --> 00:00:58.350 a new feature inside ChatGPT that acts like a 00:00:58.350 --> 00:01:00.310 personal assistant who does all that heavy lifting 00:01:00.310 --> 00:01:03.770 for you. Welcome to the Deep Dive. Today we're 00:01:03.770 --> 00:01:06.349 looking at the source material on this new shopping 00:01:06.349 --> 00:01:09.829 research feature in ChatGPT. Our mission is to 00:01:09.829 --> 00:01:12.670 walk you through exactly how to use it, why it 00:01:12.670 --> 00:01:15.569 works, and maybe most importantly where it still 00:01:15.569 --> 00:01:17.769 falls a little short. Yeah, we've got a really 00:01:17.769 --> 00:01:20.650 clear roadmap. We'll start with how this is fundamentally 00:01:20.650 --> 00:01:23.170 different from what we all think of as just Googling, 00:01:23.650 --> 00:01:27.030 then the three simple steps to actually use it, 00:01:27.370 --> 00:01:29.069 including the kinds of prompts that get you the 00:01:29.069 --> 00:01:31.329 best results. And, you know, we have to talk 00:01:31.329 --> 00:01:34.109 about the bad things, the limitations that the 00:01:34.109 --> 00:01:36.390 marketing materials don't really mention. OK, 00:01:36.489 --> 00:01:40.079 let's unpack that core idea first. What is shopping 00:01:40.079 --> 00:01:42.239 research? And how is it really fundamentally 00:01:42.239 --> 00:01:43.879 different from the search we've been using for, 00:01:43.879 --> 00:01:46.159 what, 20 years? The difference is all about labor. 00:01:46.340 --> 00:01:48.700 That's the best analogy. Think of a normal Google 00:01:48.700 --> 00:01:50.519 search. That's like walking into an enormous 00:01:50.519 --> 00:01:53.560 library. You ask the librarian, where are the 00:01:53.560 --> 00:01:55.900 books about running shoes? The librarian just 00:01:55.900 --> 00:01:57.900 points to a massive shelf and says, over there. 00:01:58.140 --> 00:02:00.540 Good luck. You tell us to pull out every single 00:02:00.540 --> 00:02:02.439 book, read the index, and piece it all together 00:02:02.439 --> 00:02:05.180 yourself. Exactly. That's traditional search. 00:02:05.480 --> 00:02:07.780 Shopping research, on the other hand, is like, 00:02:08.080 --> 00:02:10.099 you hired a personal assistant. You give them 00:02:10.099 --> 00:02:12.039 these incredibly specific needs. You don't just 00:02:12.039 --> 00:02:14.319 say running shoes. You say, I want running shoes, 00:02:14.620 --> 00:02:17.319 my feet hurt when I run downhill, and my budget 00:02:17.319 --> 00:02:21.180 is strictly under $100. Yes. And that assistant 00:02:21.180 --> 00:02:24.139 takes all those constraints, runs into the library, 00:02:24.280 --> 00:02:27.520 reads everything, and comes back with just the 00:02:27.520 --> 00:02:30.539 top three options, all summarized for you. That 00:02:30.539 --> 00:02:33.699 filtering, that is the game changer. Because 00:02:33.699 --> 00:02:36.889 it's trained. Specifically for this. Precisely. 00:02:37.090 --> 00:02:40.729 This AI is trained to understand products, prices, 00:02:40.889 --> 00:02:42.990 and features. It's designed for commerce, not 00:02:42.990 --> 00:02:45.889 just for general facts. So you don't have to 00:02:45.889 --> 00:02:47.990 filter out the bad information yourself. And 00:02:47.990 --> 00:02:49.810 the interface itself is pretty surprising, isn't 00:02:49.810 --> 00:02:51.889 it? It's not just a boring wall of text. No, 00:02:51.889 --> 00:02:54.590 not at all. The screen actually changes. It shifts 00:02:54.590 --> 00:02:56.949 to show what they call product cards. They have 00:02:56.949 --> 00:02:59.050 these nice pictures, the price. And even little 00:02:59.050 --> 00:03:02.150 buttons like more like this are not interested. 00:03:02.250 --> 00:03:04.490 It feels like a smart shopping app. Let's push 00:03:04.490 --> 00:03:07.349 on that a bit, though. Product cards sound nice, 00:03:07.849 --> 00:03:10.530 but what's really behind them? Is this just a 00:03:10.530 --> 00:03:13.009 prettier way of showing as sponsored links, or 00:03:13.009 --> 00:03:15.949 is the sourcing actually different? That's a 00:03:15.949 --> 00:03:20.139 vital question. Trust is everything. OpenAI's 00:03:20.139 --> 00:03:22.500 claim is that the results are organic. That means 00:03:22.500 --> 00:03:25.120 brands can't pay to get to the top. The results 00:03:25.120 --> 00:03:28.159 are based on the AI synthesizing reviews, specs, 00:03:28.520 --> 00:03:31.319 and your query. It's not just pulling from blogs. 00:03:31.580 --> 00:03:33.560 So it's filtering better because it has access 00:03:33.560 --> 00:03:36.599 to a specialized pool of commercial data. Exactly. 00:03:36.680 --> 00:03:38.560 It understands product trade -offs, and that 00:03:38.560 --> 00:03:40.919 helps it guide you beyond just a list of links. 00:03:41.039 --> 00:03:43.379 Okay, that helps. So let's get into the application. 00:03:43.500 --> 00:03:45.379 Walk us through it. What's the three -step process 00:03:45.379 --> 00:03:47.460 for actually using this thing? It's a very clean 00:03:47.460 --> 00:03:50.099 process. So step one is the trigger. There's 00:03:50.099 --> 00:03:52.680 no secret button. You just open up ChatGPT and 00:03:52.680 --> 00:03:54.360 you start chatting like you want to buy something. 00:03:54.580 --> 00:03:57.020 So you just type, help me find a new laptop for 00:03:57.020 --> 00:03:59.780 college. That's it. That's all it takes. The 00:03:59.780 --> 00:04:02.719 AI sees that shopping intent and it just switches 00:04:02.719 --> 00:04:04.699 modes. Sometimes a little button will pop up 00:04:04.699 --> 00:04:06.560 that says use shopping research and you just 00:04:06.560 --> 00:04:09.860 click it. Then step two, the interview. You said, 00:04:10.180 --> 00:04:12.659 this is the part you can't rush. This is the 00:04:12.659 --> 00:04:14.560 most critical part. It doesn't just spit out 00:04:14.560 --> 00:04:19.420 results. It pauses, and it asks you clarifying 00:04:19.420 --> 00:04:21.480 questions. Right. For the laptop, it would ask 00:04:21.480 --> 00:04:24.660 something like, is this for office work or gaming? 00:04:25.279 --> 00:04:27.560 Lightweight. What's your max budget? Exactly. 00:04:27.740 --> 00:04:30.600 And it needs to know those constraints. And our 00:04:30.600 --> 00:04:33.139 advice here is really important. Answer those 00:04:33.139 --> 00:04:35.560 questions carefully. The quality of your results 00:04:35.560 --> 00:04:37.800 is directly tied to the detail you give in this 00:04:37.800 --> 00:04:39.939 step. If you rush it, you're just going to get 00:04:39.939 --> 00:04:43.040 generic junk. Then you get to step three, the 00:04:43.040 --> 00:04:45.579 buyer's guide, the AI goes into a researching 00:04:45.579 --> 00:04:47.639 mode, and then the output is superstructured. 00:04:47.779 --> 00:04:50.000 You get a clear top pick, a budget pick, and 00:04:50.000 --> 00:04:52.480 really specific pros and cons for each. Like, 00:04:52.699 --> 00:04:55.160 this model has a 10 -hour battery, but reviews 00:04:55.160 --> 00:04:58.230 say the screen is a bit dark. That one detail 00:04:58.230 --> 00:05:00.569 that could save you hours of reading through 00:05:00.569 --> 00:05:03.230 forums. It saves brain energy. That's the real 00:05:03.230 --> 00:05:06.069 value here. So why is that second step, the interview, 00:05:06.610 --> 00:05:08.949 so critical to getting useful results? Because 00:05:08.949 --> 00:05:11.550 the AI needs detailed constraints to narrow its 00:05:11.550 --> 00:05:14.250 focus and understand your real needs. Let's clarify 00:05:14.250 --> 00:05:18.110 the nuance here. Inside Chat GPT, you've got 00:05:18.110 --> 00:05:20.230 this new shopping research, but you also have 00:05:20.230 --> 00:05:22.439 the normal web search. How do you know which 00:05:22.439 --> 00:05:24.660 one to use? Use the normal search, the little 00:05:24.660 --> 00:05:27.220 globe icon for quick facts. How much is the iPhone 00:05:27.220 --> 00:05:30.579 16? When's the next big sale? You'll get fast 00:05:30.579 --> 00:05:33.079 factual answers. And jobbing research is the 00:05:33.079 --> 00:05:34.860 decision maker. You use that when you need to 00:05:34.860 --> 00:05:37.560 compare things or analyze trade -offs. Exactly. 00:05:37.680 --> 00:05:40.480 Something like, should I buy the iPhone 15 or 00:05:40.480 --> 00:05:42.959 wait for the 16 if I mainly take photos at night? 00:05:43.129 --> 00:05:45.430 It helps you evaluate the alternatives. And this 00:05:45.430 --> 00:05:47.610 all works because shopping research is what's 00:05:47.610 --> 00:05:50.870 called an agent. Right. An AI agent is just a 00:05:50.870 --> 00:05:54.310 software robot trained to do a specific complex 00:05:54.310 --> 00:05:57.490 task. In this case, that task is shopping research. 00:05:57.769 --> 00:06:00.649 It has a specialized brain, probably based on 00:06:00.649 --> 00:06:03.110 something like GPT -5 Mini, that understands 00:06:03.110 --> 00:06:05.149 buyer trade -offs. It knows that if you want 00:06:05.149 --> 00:06:06.689 something super light, you're probably giving 00:06:06.689 --> 00:06:09.410 up some battery life. Yes, exactly. A generic 00:06:09.410 --> 00:06:11.790 AI can sometimes miss that nuance. It's something 00:06:11.790 --> 00:06:14.370 we call pr - prompt drift. To find prompt drift 00:06:14.370 --> 00:06:17.550 for us. Prompt drift is when you start a conversation 00:06:17.550 --> 00:06:20.389 about one thing, like a camera, and a few messages 00:06:20.389 --> 00:06:22.449 later the AI is talking about travel destinations 00:06:22.449 --> 00:06:25.069 because it got stuck on one word. It's basically 00:06:25.069 --> 00:06:26.889 forgetting the original point of the conversation. 00:06:27.170 --> 00:06:30.569 Losing focus. Yeah. And I'll be honest, I still 00:06:30.569 --> 00:06:33.170 wrestle with prompt -to -drift myself when using 00:06:33.170 --> 00:06:35.850 generic AI, so having an agent that's trained 00:06:35.850 --> 00:06:38.290 specifically on these trade -offs is a huge relief. 00:06:38.329 --> 00:06:41.189 It stays on topic. Which brings us to what might 00:06:41.189 --> 00:06:43.129 be the most valuable lesson for you listening. 00:06:44.250 --> 00:06:47.110 Specificity is power. You have to tell the AI 00:06:47.110 --> 00:06:49.810 a story about your life, not just throw keywords 00:06:49.810 --> 00:06:51.509 at it. Yeah, let's run through a few examples. 00:06:51.629 --> 00:06:54.189 The hardest thing to shop for is a gift. A bad 00:06:54.189 --> 00:06:56.610 prompt is just, find a gift for my dad, you know 00:06:56.610 --> 00:06:58.819 what you're going to get. a tie or a coffee mug. 00:06:59.000 --> 00:07:02.139 But the powerful prompt tells a story. It connects 00:07:02.139 --> 00:07:05.240 things that seem unrelated. Like try this, I 00:07:05.240 --> 00:07:07.839 need a birthday gift from my dad. He's 60, just 00:07:07.839 --> 00:07:10.240 retired and started gardening, but he complains 00:07:10.240 --> 00:07:12.959 his back hurts when he bends down. He also loves 00:07:12.959 --> 00:07:16.439 old rock music. My budget is under $100. You've 00:07:16.439 --> 00:07:18.720 given it everything. Context, physical limits, 00:07:18.939 --> 00:07:22.379 hobbies, a price. And the result is so smart 00:07:22.379 --> 00:07:25.120 because the AI makes these logical leaps. It 00:07:25.120 --> 00:07:27.759 connects back pain and gardening and suggests 00:07:27.759 --> 00:07:29.839 a garden kneeler seat. Or it links gardening 00:07:29.839 --> 00:07:32.740 and music and suggests a waterproof Bluetooth 00:07:32.740 --> 00:07:35.600 speaker for the garden. Yes. Wait, that's incredible. 00:07:36.079 --> 00:07:38.139 That ability to connect human constraints is 00:07:38.139 --> 00:07:40.439 the real shortcut. That saves so much stress. 00:07:40.579 --> 00:07:43.019 It is. And it works for tech stuff, too. Say 00:07:43.019 --> 00:07:44.959 you need a professional monitor. You can be super 00:07:44.959 --> 00:07:48.300 specific. SRGB needs to be almost 100 % for color 00:07:48.300 --> 00:07:51.040 accuracy. It has to charge my MacBook Pro over 00:07:51.040 --> 00:07:55.259 USB -C. Max 27 inches, $400 budget. So it filters 00:07:55.259 --> 00:07:57.459 out all the gaming monitors and looks for professional 00:07:57.459 --> 00:07:59.800 models, and it'll even compare the charging wattage 00:07:59.800 --> 00:08:03.160 of the USB -C port. And one more fashion. I need 00:08:03.160 --> 00:08:05.899 a dress for a beach wedding. It has to be breathable, 00:08:06.139 --> 00:08:08.399 like linen or cotton, elegant, kind of like the 00:08:08.399 --> 00:08:11.899 $300 brand X, but my budget is under $100. And 00:08:11.899 --> 00:08:13.480 instead of just showing you expensive stuff, 00:08:13.899 --> 00:08:16.279 it figures out the aesthetic and then finds cheaper 00:08:16.279 --> 00:08:18.480 brands with the right fabric. What the success 00:08:18.480 --> 00:08:20.920 of these prompts really teaches us is that the 00:08:20.920 --> 00:08:23.740 system is great at linking multiple, unrelated 00:08:23.740 --> 00:08:26.180 human needs into one solution. Okay, we've talked 00:08:26.180 --> 00:08:27.980 about the power of this. Now we have to shift 00:08:27.980 --> 00:08:31.139 to the critical analysis. We aren't selling this, 00:08:31.139 --> 00:08:33.519 we're analyzing it. So let's be honest about 00:08:33.519 --> 00:08:36.480 the pros and the cons. First, the good things. 00:08:36.879 --> 00:08:39.159 The first, and maybe the biggest, is just saving 00:08:39.159 --> 00:08:42.639 your brain energy. Seriously. You read one summary 00:08:42.639 --> 00:08:44.960 instead of 10 conflicting articles, the stress 00:08:44.960 --> 00:08:47.840 reduction is immediate. The second thing is contextual 00:08:47.840 --> 00:08:50.460 memory. This is really powerful. If you told 00:08:50.460 --> 00:08:53.940 chat GPT yesterday that you adopted a dog, and 00:08:53.940 --> 00:08:56.480 then today you ask it for a sofa, it'll suggest 00:08:56.480 --> 00:08:59.019 scratch -resistant fabric without you even asking. 00:08:59.159 --> 00:09:01.120 Google can't do that. That's a permanent layer 00:09:01.120 --> 00:09:04.120 of understanding about you. Whoa. Imagine when 00:09:04.120 --> 00:09:06.899 that scales. When it retains context across a 00:09:06.899 --> 00:09:09.960 billion users, that level of personalization 00:09:09.960 --> 00:09:12.580 based on actual history, not just tracking cookies, 00:09:13.120 --> 00:09:15.799 that's a total game changer. And the third pro 00:09:15.799 --> 00:09:19.080 is spotting fake or junk products. Because it's 00:09:19.080 --> 00:09:21.059 synthesizing reviews from all over, including 00:09:21.059 --> 00:09:23.440 places like Reddit, it can warn you. It might 00:09:23.440 --> 00:09:26.299 say, users report this specific model breaks 00:09:26.299 --> 00:09:29.090 after one month. Okay, now for the limitations. 00:09:29.909 --> 00:09:32.029 The bad things. This is what you really need 00:09:32.029 --> 00:09:34.710 to know before you start relying on it. Price 00:09:34.710 --> 00:09:36.789 errors. This seems to be the biggest problem. 00:09:37.009 --> 00:09:39.690 The AI says a camera is $500. You click the link 00:09:39.690 --> 00:09:42.889 and it's actually $550 on the website. Why does 00:09:42.889 --> 00:09:45.039 that happen with such an advanced system? Well, 00:09:45.200 --> 00:09:47.919 it's probably a mix of things. Prices on Amazon 00:09:47.919 --> 00:09:50.620 and other sites change hourly, sometimes by the 00:09:50.620 --> 00:09:53.419 minute. Constantly checking millions of prices 00:09:53.419 --> 00:09:56.899 in real time is a massive, expensive computation. 00:09:57.440 --> 00:09:59.440 So the data it's using might be a few minutes 00:09:59.440 --> 00:10:02.159 or even a few hours old. So the lesson is always, 00:10:02.159 --> 00:10:04.379 always check the final price on the seller's 00:10:04.379 --> 00:10:06.360 site. Absolutely. And then there's the dreaded 00:10:06.360 --> 00:10:07.940 sold out. You get the perfect recommendation. 00:10:08.019 --> 00:10:10.960 You click, and it's gone. Real -time stock checking 00:10:10.960 --> 00:10:13.500 is another clear limitation right now. And it 00:10:13.500 --> 00:10:15.639 struggles with rare items, right? It's great 00:10:15.639 --> 00:10:19.779 for iPhones, but not for, say, spare parts for 00:10:19.779 --> 00:10:24.299 a 1980 sewing machine. Yeah, it'll struggle with 00:10:24.299 --> 00:10:26.460 that because that specialized data just isn't 00:10:26.460 --> 00:10:28.480 in its training set. OK, we have to address the 00:10:28.480 --> 00:10:32.149 anxiety, the safety and trust questions. Is this 00:10:32.149 --> 00:10:35.049 just a fancy ad system? It's a fair question. 00:10:35.169 --> 00:10:37.850 But as we said, OpenAI claims the results are 00:10:37.850 --> 00:10:40.889 organic. Brands can't pay to be the top pick. 00:10:41.490 --> 00:10:44.169 The ranking is based purely on reviews and how 00:10:44.169 --> 00:10:46.470 well a product matches your prompt. And will 00:10:46.470 --> 00:10:48.690 it steal my credit card info? Absolutely not. 00:10:48.809 --> 00:10:50.850 Let's be crystal clear on this. It's a language 00:10:50.850 --> 00:10:53.070 model. It just generates words and links. It 00:10:53.070 --> 00:10:55.330 never sees a payment portal or stores your card 00:10:55.330 --> 00:10:57.909 number. All transactions happen on the actual 00:10:57.909 --> 00:11:00.870 store's website. Chat GPT never touches your... 00:11:00.809 --> 00:11:05.629 Okay, last one. Tracking. Does it track me? Yes, 00:11:05.870 --> 00:11:07.909 but transparently, through its memory feature, 00:11:08.090 --> 00:11:10.110 it might give you proactive suggestions like 00:11:10.110 --> 00:11:11.990 recommending coffee beans a week after you bought 00:11:11.990 --> 00:11:14.210 a coffee machine. If that feels creepy, you can 00:11:14.210 --> 00:11:15.990 just go into your settings and turn the memory 00:11:15.990 --> 00:11:18.549 feature off. It's fully in your control. So let's 00:11:18.549 --> 00:11:20.990 bring it all together. Our conclusion, based 00:11:20.990 --> 00:11:23.789 on the material, should people actually try this? 00:11:24.049 --> 00:11:27.710 Yes. 100 % yes, try it. It's not perfect, especially 00:11:27.710 --> 00:11:30.330 with the price and stock issues. But it fundamentally 00:11:30.330 --> 00:11:33.330 changes how we approach finding things online. 00:11:33.970 --> 00:11:36.470 It turns shopping from that tiring, frustrating 00:11:36.470 --> 00:11:39.529 hunt into a focused conversation with a helpful 00:11:39.529 --> 00:11:42.029 guide. It seems most useful for people who are 00:11:42.029 --> 00:11:44.649 really busy, or maybe non -tech savvy people 00:11:44.649 --> 00:11:48.129 who need a complex gadget, or honestly, anyone 00:11:48.129 --> 00:11:50.450 who struggles to buy a thoughtful gift. It's 00:11:50.450 --> 00:11:52.789 a filter for the noise. You still make the final 00:11:52.789 --> 00:11:55.029 call, but the noise is turned way down. So don't 00:11:55.029 --> 00:11:57.669 just listen to this. Put it to work. Open ChatGPT 00:11:57.669 --> 00:11:59.590 and use this template right now. Just fill in 00:11:59.590 --> 00:12:02.070 the blanks. I want to buy item name for a size 00:12:02.070 --> 00:12:04.470 or application. The most important features for 00:12:04.470 --> 00:12:06.769 me are feature one and feature two, and my hard 00:12:06.769 --> 00:12:09.250 budget is around. Please find the three best 00:12:09.250 --> 00:12:11.809 options and compare them for me. And that really 00:12:11.809 --> 00:12:13.690 raises our final provocative thought for you 00:12:13.690 --> 00:12:16.399 to consider. What does it mean when our consumption 00:12:16.399 --> 00:12:19.159 is entirely mediated by these super personalized, 00:12:19.240 --> 00:12:22.360 constantly learning AI agents? What do we lose 00:12:22.360 --> 00:12:25.019 in terms of serendipity or maybe even just accidental 00:12:25.019 --> 00:12:27.799 discovery when that whole tiring hunt is completely 00:12:27.799 --> 00:12:30.100 outsourced? Something to think about as you click 00:12:30.100 --> 00:12:32.360 that researching button. Thank you for joining 00:12:32.360 --> 00:12:34.159 us for this deep dive. We'll see you next time.