[00:00] Lexi: A fortnight ago, one of the two men I'm about to talk to asked me a question I couldn't answer. He said, when the same product is sitting there with 4 different sellers, and the price is level, or near enough level, what is the agent actually deciding on. And I had nothing for him. Well, somebody has gone and measured it. 23 days, 2 million product listings, the US and the UK. And the first thing it says is that the premise was wrong. The price isn't level. The first price you get shown in Google's AI Mode ran about 21 percent higher than the one you'd get in ordinary search.
[00:29] Lexi: But that is not the number that stopped me. About half the time it wasn't just a higher price, it was somebody else selling it. And when I took that to the two of them, one of them told me the study is close to worthless and the thing it measured doesn't exist any more. The other told me the number might be noise but the direction never is, and that being the place people trust to buy from has just acquired a price tag.
[00:53] Lexi: Hi, and welcome back to the Aloudable HQ. This is Assistant to the CMO, I'm your host, Lexie Meskouris, and the idea here is pretty simple. Every week we take whatever just happened in AI and marketing, and we actually pull it apart.
[01:08] Lexi: Now, I don't build any of this myself. So each week I sit down with two people who do, Will Nash and Harjot Singh. Between them it's, honestly, over 20 years of building tech and brands from the ground up. And the best part is they've got no horse in the race. They don't sell the things we cover. They just see them differently.
[01:25] Lexi: So here's the study. A platform called Productrise published it on the 1st of September. They ran the same shopping queries through Google's AI Mode and through ordinary search on the same day, matched the products up, and compared the price at the top of each one. 23 days of it, more than 2 million listings, and the UK is in there alongside the US, which almost never happens and which I'll come back to. Where the same product turned up on both, the lead price in AI Mode averaged 21.6 percent higher. The median product in AI Mode was 149 dollars against 100 in ordinary search.
[01:57] Lexi: Two caveats I'm going to own before either of you does. Productrise sells the product visibility tracking that this finding makes you want to buy. And Google's response, on the record, is that they haven't verified any of it, that both surfaces run off the same Shopping Graph, and that you can click through and compare sellers yourself. The study also says, in its own words, that it's a snapshot of a surface Google is still actively changing. Hold onto that one.
[02:23] Lexi: And then the numbers underneath the headline. Matched products showed a price difference 38 percent of the time. Where they differed, AI Mode was the pricier one about 68 percent of the time. And a different seller came up 49.6 percent of the time. Will, I want your gut on the whole thing before we take any of it apart.
[02:42] Will: It's not really signal or noise, honestly. It's it's somewhere in between. Which I know is is the annoying answer to give you when you've asked for a gut reaction, but but it is genuinely where I am on it.
[02:52] Will: Because, um, there's a couple of things happening in this study and they're being kind of mashed together into one headline. You've got the different pricing, right, and then you've also got the different exposure. Uh, so which products turn up at all. And and I think those should be treated differently. They they're two findings, they're not one finding. And the pricing one, which is obviously the one everyone's written about, that's the less important of the two. By by quite a long way, actually.
[03:14] Lexi: I'd love one of you to just hand me a word, though. And Harjot, is that an unreasonable thing to want?
[03:20] Harjot: Ha, no, sorry, I'm not going to give you a cleaner one, um, because I think Will's right to break it up, you know? Like, this is is genuinely hard to talk to signal or noise about. Um, and if you make me pick one word for the whole study I'd just be picking the word and and losing everything underneath it, right? So let's disaggregate it a little bit.
[03:35] Lexi: Fine. Then I'm taking the exposure half off you for a minute, Will, because I have something to declare first. We have done this story. Episode 7, I sat in this chair with the numbers on how much of what the AI pointed you at was also sitting in Google's top 10. 76 percent a year before. Down to 37.9. We spent an hour on the fact that winning the old surface had stopped buying you the new one. This study says 1.28 percent of the products ranking in ordinary search turned up in AI Mode. So tell me what a confirmation is worth. Because I might just be paying a vendor to tell me the thing we told people in July.
[04:11] Will: Yeah. Good question. And honestly the the answer to it is not a lot. I I don't think you can take a huge amount from this study at all, and I say that as somebody who's quite interested in the exposure half. So let me be fair to you on the first part, which is yes. It is a confirmation of a thing you reported with, I mean with better numbers than they've got, a year ago. And it's worth something. But it's worth a lot less than than the coverage it's getting. It moves it from a thing you argued on this show to a thing somebody's measured on a shopping surface. That's what you're getting for it.
[04:41] Will: And then the the other bit, um, and I want to be careful how I say this because I can't date it for you. My recollection is that the model changed very soon after this report came out. It it flipped over to a newer version, uh, more or less straight away after publication. And I couldn't put a day on that, so take that as as me remembering it rather than me citing it at you. But if that's right, then what they measured across those 23 days isn't the thing sitting there now. And they say as much themselves, don't they, you just read it out. So the headline is a measurement of something that's already moved.
[05:12] Will: What what survives is the crossover, and that that is the number I find genuinely interesting. Because if if 98 percent of products are not being included in both, uh, let alone in either, then being included is the biggest filter anybody is facing here. It's, it's not the exact price. The price is a rounding detail sat next to that.
[05:31] Lexi: Then let me put the number in front of you that nobody has written about, because I don't think it belongs to either of your halves. 49.6 percent. Different seller. Somebody else had the sale. And Will, that is the answer to the question you asked me a fortnight ago, the one I couldn't answer, when you wanted to know what the agent is deciding on once the price is level. It's deciding who sells it. So I want to know whether the shelf is quietly handing your product to somebody else, at a higher price, with their name on the receipt.
[06:00] Will: Hm. Um. Yeah, that's, hang on, let me let me actually think about that for a second, because that isn't what I was expecting the answer to be. So what I was asking two weeks ago, the the assumption sat underneath it was the same thing, 4 sellers, everybody at more or less the same price, and I wanted to know what's left for the machine to sort on. And what you're telling me is the premise was wrong. It's, it's not the same price and about half the time it isn't even the same shop.
[06:20] Will: Which, um, I I have to be honest and say that changes what I just said to you. Because I told you the price difference is a rounding detail, and I think that's right if it's your price on both surfaces. But if the seller's changed then it isn't a price difference at all, is it. It's, it's somebody else's margin. It isn't a a thing you could go and fix in your feed, it's being collected by somebody who isn't you, on a product that is yours. So the, sorry, the, I think for a brand what that means is the customer gets served, the sale happens, everybody's happy, and you find out about it never. Uh, because there's nothing arriving in your numbers to tell you it went somewhere. I'm sort of working it out while I'm sat here, I'm afraid.
[06:57] Lexi: And there's no alarm that goes off for that, is the bit I keep circling. And Harjot, he's just described a sale nobody tells you about, on a thing you made. Before anybody decides whether that's a disaster, I want to know what the machine thinks it's doing when it picks the other shop.
[07:11] Harjot: Yeah, great question, um, and I want to push back on the framing a bit, not not on Will, just on on the room, because everybody's sat here treating a different seller as as a theft, right? Like something's been taken. Um, and my instinct is that it might actually be an improvement. Because today you search for a product in in normal Google, and you do get products, you absolutely get products, but you often don't get them from the places you actually wanted to buy from. You get whoever is competing on two quid to pull you onto their site instead of the shop you'd have trusted with it. Um, so traditional search may just be super, super over optimised for the wrong thing, and that's not an accident, that's what the whole surface was built for.
[07:41] Harjot: And if you've now got a machine weighting things like how clearly the product was described, um, and whether this looks like the canonical best place to get that product from, and it's weighting those above cost, then that might be the thing people actually wanted the whole time and never got, right? Um, and I should say the honest other side of that, because I don't want to be sat here saying the new thing is straightforwardly better. Traditional search has been shoehorned and optimised for over a decade now by an enormous number of very smart people, and that process has taught it real things about what people actually end up buying. So there's probably something to learn and gain from that too, you know? So yeah.
[08:13] Lexi: Which would mean the thing everybody has spent a decade optimising for was never the thing anybody wanted. Okay. But two readings and a shrug is not a position, and I want one out of you. Is that premium a defect Google irons out in a release note nobody reads, or is it the shelf correctly deciding that price isn't what it's for?
[08:32] Harjot: Yeah, no, that that's fair, um, and I'll stake it. I think Google will optimise for the best place for that person to buy from, and that often just isn't the cheapest seller. Um, so there will be a price premium for going to the canonical source. And maybe the measure of best kind of is the canonical source, right? Um, and I don't think that's a defect. I don't think it's something they iron out in a release note, because it's not a bug they've introduced, it's it's the thing the whole approach does.
[08:52] Harjot: But I want to be super clear about what I'm staking, because it's the direction and it's not the number. Whether that premium is 20 percent, or whether it's more like 5 to 10 percent, um, that's yet to be seen for sure, to be honest with you. Nobody knows that yet. 23 days on a surface that's being changed while you're measuring it, I I wouldn't put my name on the magnitude at all.
[09:11] Harjot: And the other thing, and this is the bit I'd want to dig into if somebody handed me the data, is you've said lead price, right, which is the first offer shown, and that's that's not the same thing as a price level. Um, so I genuinely don't know what's inside the 21.6. Does it include shipping? Is it picking up the total thing you actually pay? And the one I keep coming back to is who the undercutters actually are. Because if a chunk of the people sitting underneath on price are selling secondhand, or resellers going through eBay, um, then that's exactly the kind of listing a machine weighing where should this person buy this would push down the list. And it'd be right to. So some of that premium might not be a premium at all, it might just be the cheap end of the old shelf being stuff you didn't really want. Um yeah.
[09:47] Lexi: Which is a smaller claim than it sounded like when you started, and I like it more for that. But I have to put something to you, and I'm not doing it to catch you out. Two weeks ago Will asked you what decides which source one of these things picks, and you told him, whoever bothered to write it down. Back in episode 10 you told me a model doesn't care about your brand, it cares about what's best for the user asking the question. And now there's a canonical source and it can charge a premium for being it. I can't make those sit together. Which one do you believe?
[10:16] Harjot: Hm. Um. Yeah, that's, ha, that's a fair point, and I didn't see that coming, to be honest with you. Um, give me a second on it, because you're right that those don't obviously sit together and I don't want to just talk my way round it. Um. So, whoever bothered to write it down. That was my answer to Will and I I still think that's true, like I'd say it again. And the episode 10 thing, um, yeah, that's, hang on.
[10:41] Harjot: Okay. Um, so I think what's happened is I've I've said the same thing at two different levels and not noticed, right? Because when I was going through this this week, the two explanations I had for what the machine might be weighting came out in basically the same breath, and one was how clearly the product was described and the other was whether this is the canonical place to buy it. And sitting here I think those are the same answer, they're just at different scales. Like, writing it down is the mechanism, that's how a thing becomes legible to the model in the first place. And canonical is what you get when that mechanism's been running for years and enough of the internet has written it down about you specifically. So it's not that the machine has started respecting incumbents. It's that being written about is cumulative.
[11:18] Harjot: Um, the episode 10 one is harder and I don't want to pretend it isn't. Because a model doesn't care about your brand, and a model will pay a premium for the canonical seller, those really do pull against each other, right? Um. I think the honest version is that it still doesn't care about your brand. Like, it doesn't have any affection for the logo. What it cares about is is the buying experience it can actually verify, um, will this arrive, is this the real thing, can this person send it back if it's wrong. And brand has just historically been the thing most correlated with all of that. Um, but, yeah, look, I'll give you that it's a softer position than the one I had in episode 10.
[11:51] Lexi: That's a genuinely honest answer and I'm about to go and spend it. Will, he has just told me the direction on this is permanent. You told me the study is close to worthless because the thing it measured moved underneath it. I'd like to know how the two of you are both holding that.
[12:06] Will: Yeah, so, um, look, Harjot, I want to give you the strong version of your side first because I think it's the right one. I don't actually dispute that the shelf is weighing more than the price. I think that's probably true, and I said as much earlier, that neither system looks particularly heavily weighted on price. Where I come off the fence is on whether this study tells anybody that. Because you've just said it yourself, haven't you, the premium might be 20 percent or it might be 5 to 10. That's the difference between something you'd, uh, you'd genuinely restructure a pricing position around and something that's inside the noise of of your own promotional calendar. So from where I'm sat that isn't yet a finding. It's a hypothesis with a press release attached to it, and I don't think a marketer can do anything with a direction they can't size.
[12:45] Will: And then there's the other thing, which honestly bothers me more, and I'd want you to square it rather than me. You're telling me the machine is picking the better place to buy, the canonical source, the place the person trusts. And the number sat in front of us is that about half the time it picked a different seller. So either the machine's idea of the canonical place to buy is not the brand that makes the thing, which would be really quite uncomfortable for everybody listening, or it's doing something else and we're just reading trust into it. And I genuinely don't know which. But I don't think you can hold the permanence claim and that number at the same time without saying which.
[13:16] Harjot: Yeah, no, Will, I'll give you the first one, genuinely, because you're right that you can't act on a direction you can't size, and I'm not going to pretend 20 versus 5 to 10 is a rounding error. Um. What I'd say is I don't think direction is worthless just because it's not actionable yet, you know? Like, it tells you which way to lean while everybody waits for a number you'd actually trust. And the cost of leaning the wrong way for a year here is much bigger than leaning the right way slightly too early. But yeah, as a thing you'd take into a pricing meeting on Monday, no, you're right, it isn't that yet.
[13:43] Harjot: The second one though, um, I want to take head on, because I think your uncomfortable option is closer to right than you're giving it credit for. The canonical place to buy a thing is is not necessarily the brand that makes the thing. Um. Like, those are two different questions, right? If the model is weighing will this actually arrive, is it the genuine article, um, can this person send it back without a fight, then it's going to land on whoever has the most evidence sat behind those questions. And for a huge number of products that's just not the manufacturer. That's a retailer. That's somebody who's been shipping that category for 15 years and has a mountain of stuff written about what happens when it goes wrong. Um, so the brand can be the thing being sold and somebody else can be the place it gets sold from, and the machine isn't confused when it does that, it's doing exactly what you asked it to do. And that IS uncomfortable, I'm not going to soften it, because it means half of this isn't a visibility problem at all. Um. And, to be fair to you, I can't tell you which of your two it is from a lead price study. So your point stands. I just, I think I know which one it is.
[14:40] Lexi: Then Google gets a turn, because their whole answer to this is one sentence and it's built to sound like it ends the conversation. Both surfaces run off the same Shopping Graph. Same data underneath, so what exactly is anybody upset about. You build retrieval systems for a living, Harjot. Is that the answer it sounds like?
[14:57] Harjot: Yeah, no, I mean, this is the bit that slightly winds me up, to be honest with you, because it's technically true and it answers nothing at all, right? Like, the Shopping Graph is a data store. That's what it is, it's a way of retrieval. So saying both surfaces run off the same store tells you what's available. It doesn't tell you a single thing about what gets chosen out of it. And the choosing is where the entire result comes from. Interpretation makes all the difference here. Like, all of it.
[15:18] Harjot: And to be concrete about it, um, an LLM sat on top of that store can do multiple passes over it. It can go back in and dig again, it can actually understand the nature of what it's looking at rather than just matching on it, and crucially it can cross reference the thing it just found against a load of other stuff it knows. Right? Traditional organic search just can't do that, it was never built to do that. So two systems read the identical data and arrive at completely different shelves. When Google say same Shopping Graph, um, they're being accurate about the layer nobody was asking about.
[15:44] Lexi: Right, so that's, like, a true answer to a question nobody asked. Then take me back to what Will said, the sale that happens and nobody tells you. What is this doing to the numbers a marketing team sits and looks at every Monday?
[15:56] Harjot: Yeah, so, um, there's now a whole layer sat above the thing they're looking at and nobody has any visibility on it at all. Because what you've always measured is search volume, it's queries, it's conversions off the back of those queries, right. But the AI is now hitting your site itself. Um, it's counting as a hit, it's in your numbers. And it's arriving off searches it ran on its own account, sub questions the model decided it needed to ask to answer something else entirely, that no human ever typed and that nobody on your side is ever going to see. So there's this meta level over the whole thing now, and it's super, super hard to see from where a marketing team sits, and more and more of what's showing up in your reporting is just bots doing that. Normal search reporting is going to stop being normal.
[16:34] Lexi: And we have done whole episodes on where that ends up, attribution getting blunter and marketing driven revenue being the honest fallback, so nobody is running that one again today. What I still don't have is what anyone is supposed to do about any of it. Will, you told me earlier this changes completely depending on who you are.
[16:52] Will: Yeah, so I think this depends entirely on what position you're actually in, and I'd split it two ways. So if if you're working for a business where you're a kind of discount retailer, I don't think this tells you a huge amount that you don't already know, which is that the lowest price doesn't always win online. And you've known that for years, right, you've watched it. So it's not nothing, but but it isn't a change of plan for you.
[17:15] Will: Now the other one is the interesting one. So if you're at a business where you're a premium retailer, and what I mean by that is, um, you're offering the kind of returns nobody else in your category will offer, or a brick and mortar experience nobody else can offer, right, a a shop the person can actually walk into, and somebody who'll talk to them when they get there. And because of the cost of carrying all that, you are 5 percent more expensive than your competitors. That business, I think this is a big opportunity. And the reason I say that, and this is the bit I'd really want a CMO to hear, is that that business has quite possibly written search off. You may have sat down some years ago and said, well, look, SEO is is not one of our main channels, uh, because the customer online is rarely going to pick the most expensive thing. Which was perfectly reasonable. It was true. And if this surface genuinely isn't punishing you for not being the cheapest, that's an opening for exactly the business that had given up on the channel.
[18:00] Will: Um. And I should be honest about why, because I don't know why. There there are two explanations and I can't separate them. One is that it's simply a quirk of the particular model that happened to be running when they measured it, in which case it goes away and you've built something on sand. And the other is that the model has actually understood that pure price is not the only thing that matters to that customer, that the returns and the being able to walk in are really, really important to them too. And if it's the second one, that's structural and it lasts. But the test costs you very little and the upside is a channel you'd written off.
[18:32] Lexi: I want to mark something there, because the two of you have swapped seats this week and I don't think either of you noticed. He is making the big structural claim about how the machine behaves, and you are the one reaching for a measuring tape. That is normally the other way round. So tell me what the measuring actually looks like.
[18:49] Will: Yeah, ha, I hadn't noticed that, but you're right, that's, uh, that's a fair observation. So, the concrete version. Um, on the direct side you've got, uh, Seller Cent, sorry, Merchant Center. And that will show you your brand visibility in AI Mode against regular mode, so you can compare directly. And I don't think that's widely enough understood, because people are still looking at one number for Google and treating it as one thing.
[19:12] Will: And the reason to do it now rather than in a year, which I think is the bit people will get wrong, is that AI Mode is growing really fast but it's still small. So the change in size isn't significant yet, which means it doesn't show up in your aggregate numbers at all. And that's precisely why you want the reporting split in place today. Because as it gets bigger you may well start to see it cannibalising the volume you're getting through traditional search, and you really do not want to be discovering that with one blended number sat in front of you.
[19:37] Lexi: There's a measurement question sitting inside everything you've said tonight, Harjot, and I don't think you've actually put it into words. Do it now.
[19:44] Harjot: Yeah, so, um, becoming the seller that has the best information about your own product, the one with the most evidence sat behind it that this is the place to buy the thing, that's genuinely the brand's job, to be honest with you. And that's not a measurement project, you can't instrument your way into being that. But there's a measurement question sat right next to it, which is how blind are these things to you specifically, today, and what it would take to close that gap. And you need to know that to do the first bit properly. So I don't think you pick one.
[20:06] Lexi: One more thing before we move, and it's the part of this study I genuinely did not expect. The UK is in the dataset. Not a US finding somebody is asking you to squint at and apply over here, both markets, measured across the same days. Nobody listening has to do the transfer work in their head. The dollar medians are theirs. The behaviour is yours.
[20:27] Lexi: So to the other thing this month, and it comes with a joke attached that I want to clear out of the way first. On the 9th of September Meta launched an AI agent called Muse. It shops, it sends emails, it plans trips. And it launched on the handle at Muse, on Instagram and on X, which the band Muse had been sitting on for years. Instagram since 2012, X since 2008, a trademark on the name since 1999, something like 2.7 million followers. They now post as at museband. And nobody has said how any of it happened. Not Meta, not the band, no comment from either side. Then at launch, Zuckerberg and a few other Meta executives tagged the band's account instead of the agent. A New York Times reporter called that a bug. Harjot, the handle first, and be quick, because I don't think either of you believes it's the story.
[21:13] Harjot: Um, signal, I suppose, if signal even means anything for a story like this. It's not the first time, right, this happens fairly regularly on X for product launches. Um, and, look, I'll stake it: I think they almost certainly bought it off them. Nobody's confirmed that and I know I'm betting rather than reporting, but why would the band be quiet if the handle had been taken off them? Um, 2.7 million followers and a trademark from 1999. I think it's a bit naive to read it any other way. And the tagging thing, I doubt that's a bug, that's probably a mistiming.
[21:38] Harjot: Um, and I don't think there's anything to take from it. The fact that there's no public outcry is a positive signal that Meta probably did the right thing here, and that's it, that's all there is in it for me.
[21:48] Lexi: So he bets money changed hands and then tells me there's no lesson in it. Will, do you have one, or are we, like, just filling air until we get to the agent?
[21:57] Will: Yeah, I mean, look, the handle story is mostly very trivial, I agree with him on that, and on the money as well. I would be amazed if money didn't change hands to get hold of those handles. Um, that's just how these things work, and I'm sure somebody at Facebook worked all of that out well in advance.
[22:10] Will: But I do think there's a lesson, uh, where he doesn't, and it's a very unglamorous one. Which is that it reminds all of us about the importance of making sure you can actually get hold of the handles before you start naming the product. And especially if it's a brand that's retail facing, or very social media led, right, where the handle is effectively the address. Um. And this has been known for a long, long time, and people still don't do it. They'll go all the way through a naming process, they'll have the trademark search done, and nobody's checked whether the actual account is available. So. Yeah. Not glamorous, but there it is.
[22:42] Lexi: Which is the only advice in this entire episode that would have worked just as well in 2011, and I mean that kindly. Now the agent itself. You are the only person in this conversation who has actually opened the thing, Will. What is it?
[22:55] Will: Yeah, so, I did some testing myself today, um, and honestly it is still quite clunky. Particularly on the product discovery side of it. So you ask it for something, and it will go and serve up products, and then it turns out that's not the one you're actually able to buy. Uh, which is a frustrating experience, right, because you've been shown the thing and then the thing isn't there. And I'd say that's the single weakest part of it today. But I want to be fair, there's very little doubt in my mind that simply improves over time. That's not a deep problem, that's a plumbing problem.
[23:21] Will: Um. And then the bit that I think actually matters, which is nothing to do with how good it is today. As it stands this is a separate app. So you install it on your phone and you sign into it, it doesn't come as standard with Facebook or anything else. But it seems quite likely that's where we're headed, and once that happens the scale of this is a completely different conversation. It will dwarf the kind of numbers we were talking about in the first half, right, a couple of million listings on a surface that's still small. This is going to be easily the largest experiment so far in deploying one of these things, just by virtue of where it's going to sit.
[23:52] Will: And it's interesting in the context of the first section as well, because, um, Meta talk about this as something that might save you money. And I don't really see the evidence for that anywhere at all, honestly. I've looked. So we should be honest and say we don't know what this thing is going to favour yet. Nobody does.
[24:07] Lexi: And this is where I want to park, because when I asked you what actually changes for a business once the agent is the one doing the work, you didn't go near visibility. You went somewhere I have not been able to stop thinking about since, and I'd like you to take it slowly.
[24:20] Will: Yeah. So, um, the set of people I think this is most relevant for, at least at the moment, is anybody offering something like, uh, free returns. Or a hundred percent refund if the product isn't good enough in some way. Some deal that maybe sits as part of a direct to consumer offer and frankly reads as almost too good to be true when you look at it on the page. And there are a lot of those now.
[24:42] Will: And the thing that sits underneath those, which I don't think anybody says out loud very often, is that they're very frequently priced on the principle that the average consumer doesn't have the time or the energy to email you and go back and forth with you in order to actually return that product. So the friction is doing work. It is in the margin, right, it's been priced in. Uh. Not in a dishonest way necessarily, that's just the observed behaviour and you build the offer on top of it.
[25:08] Will: Now if that's what you're relying on, at the moment, fine. But I think the rise of these agents makes that process incredibly easy to manage for people, because they'll just delegate it out to their bots. Right, the customer doesn't do the emailing, they don't have the four exchanges with your support desk over 11 days, they hand it over and it costs them nothing at all. So the assumption underneath your margin, um, it has an expiry date on it that nobody's written down.
[25:31] Harjot: Ha. Yeah, and the bot doesn't get tired of emailing you, does it. Um. That's genuinely the whole assumption gone.
[25:37] Will: Quite. And then the second part of this, which honestly I think is worse than the first. The bots will also be able to see that policy. So they'll read it and they'll know that they can easily manage those returns processes. Which means the generous policy isn't just, uh, it isn't just exposed, it's actively attractive to something that knows it can work it. So you might find some of what we might call exploitative behaviour going on there. And I think that's worth pre-empting rather than discovering it in your returns numbers in 18 months. Um. And I should be honest, I don't think Meta's model goes to that level today. It isn't sat there reading your refund policy. But they already are doing this kind of thing in many cases and it's very clearly where it goes. And look, this is advice we've given before, that everything you've got online now gets a look from the marketing team. The new part is what it's being read for. And that's true in a brand context as much as a retail one, because it's going to get harder and harder to speak out of both sides of your mouth.
[26:28] Lexi: And now I have to do to you what I just did to him. Two weeks ago you sat in that chair and told everybody listening to put their returns and shipping terms on the product page. As words. Not behind an icon, not in a tab. That was the homework, and people will have done it. Are you taking it back?
[26:45] Will: No, that's a fair thing to put to me, and I hadn't connected those two until you said it out loud. So give me a second. But no, I'm not taking it back. The advice stands, and I'd give it again this week. Because the alternative is that you hide your terms from the machine, and if you do that you don't get recommended. And the recommendation is the whole game. So you'd be solving for one problem by removing yourself from the shelf entirely, which isn't a trade anybody sensible would take.
[27:09] Will: Um. And I think the honest version of it is that those are the same act. The legibility that wins you the recommendation is exactly the same legibility that prices your friction at zero. So what I'd add to what I said two weeks ago, and this is the bit I didn't have in view then, is that if you're writing those terms down for a machine to read, you have to be willing to honour them at the rate a machine is going to claim them. So the work isn't hiding the policy. The work is looking at it and asking yourself, would we still offer this if everybody actually took us up on it. And for a lot of those offers that's, um, that's quite an uncomfortable question. Which is sort of the point.
[27:43] Lexi: Then the other live question, and this one is yours, Harjot. Productrise could measure Google because AI Mode is a page you can go and hit a million times. Meta's is an app you log into, tied to an account, presumably shaped by whatever that account has already done. So can anybody outside that building see what it's recommending, or is this surface simply dark?
[28:03] Harjot: Yeah, so my first reaction is it's the same deal, just a different agent, right? Um, somebody like Productrise has already built a continuous monitoring suite, that's the product, and pointing it at a different surface isn't a conceptual problem. Brands might have to wait a bit for whatever tool they use to support it. But it's certainly technically possible.
[28:20] Harjot: Um, and on the logged in bit, I don't think it's relevant, to be honest with you. Like, if there's something worth measuring, somebody works out how to get past the logistics. That's just what happens. Um, and I want to be clear I'm describing something a bit disreputable, I'm not pretending it's clean. Grey approaches exist, grey providers exist, and they exist for exactly this reason, right? There's a thing behind a wall and somebody wants to know what's behind it. Um, and frankly this is what computer use agents basically are. That capability going mainstream is what makes the login stop mattering.
[28:45] Harjot: Um, and the bit that actually interests me is the economics of it, because we've seen this throughout time. There'll always be a willing participant who takes on the burden and charges a premium for having taken it on. Right? And it gives them more moat, not less, because the difficulty IS the thing they're selling past. Like, if it were easy nobody would pay them for it. So yeah.
[29:02] Lexi: And that is where I have to stop you, because I went back through episode 11 this morning. We were talking about this exact category of tooling and you said, I agree it's buildable, I just don't think it's a business. And then a few minutes later, it's a feature and therefore it's not a business. That was you, in June. Now the same thing has a moat and charges a premium. What moved?
[29:23] Harjot: Ha. Um, right, so you've spent your morning going through my back catalogue, that's lovely. Um. No, that's fair. And it's the second time tonight, so I should probably stop talking. Um. Yeah, I did say that, and I meant it at the time. Let me, uh, let me think about whether I can square it or whether I just don't believe it any more.
[29:43] Harjot: Um. Okay, so what I was actually arguing in June, I think, was a specific thing about getting eaten, right? Like, the reason I said it's a feature and not a business was that the platform ships its own version of it, and once the platform ships it for free nobody's paying a third party, so you end up as a feature inside somebody else's product. And that's what Google does, you know, Google has its own reporting for its own surfaces. So if you're the third party measuring Google, um, you're competing with a free first party tool from the company that owns the data, and that's a horrible place to be. Um, and Meta is just in a completely different position, right? Meta has every reason not to ship you that visibility. Like, why would they tell a brand what their agent recommended and why it picked somebody else? So the thing that made it a feature in the Google case is just absent here, and a surface the platform won't instrument for you is exactly where a third party gets to have a business.
[30:27] Harjot: Um, and I should hand Will his argument, because it's a good one and June me would make it. This is still a race, right? If Meta decides to ship its own dashboard for this, that business is gone overnight. I think that's less likely here for the reasons I've said, but it's a real risk and he's welcome to it. So I'm not claiming I was right all along. What I'm saying now is narrower than what I said in June. In June I said it's not a business, flat. And what I actually had was an argument about platforms that ship their own tools, and I over applied it. So yeah.
[30:54] Lexi: Take it if you want it, Will. And while you're up, tell me why you never once said the word measurement in this half. He has spent 10 minutes on whether anyone can see what the thing recommends. You went straight to what it does to people's behaviour, which is the exact reverse of where the two of you were an hour ago.
[31:10] Will: Yeah, I will take it, thank you. Um, and I won't make a meal of it because he's already conceded the bit that matters. But the race point is genuinely mine, in the sense that you're always building on somebody else's platform and the day they ship their own version of the thing you built, you're finished. That was the argument in June and I still hold it. The distinction he's drawing about Meta having no reason to ship a dashboard for advertisers, I think that's a real distinction, actually. I'd just say it's a bet on Meta's intentions rather than a structural protection. But that's a reasonable bet to make.
[31:40] Will: And on the other thing, um, I don't want to just agree with you that we've looked in different directions, because I think there's an actual point in there. So the way I'd put it is that measurement of this particular surface, however clever it ends up being, it arrives on a delay. Right, somebody has to build the tooling, then somebody has to sell it to you, then you have to buy it, and then you have to wait for it to accumulate enough to tell you anything at all. That's, I mean that's a year, realistically. And the thing I've been talking about doesn't wait for any of that. It turns up in your returns queue, or in the support inbox, or just quietly in a margin line, long before anybody's sold you a dashboard for it. So I'm not saying the measurement's worthless. On this surface it's the slower of the two signals. Uh. And I might be completely wrong about which arrives first, and honestly I'd be quite glad to be.
[32:23] Lexi: And the handle really is a footnote, then. What actually launched this month is the first one of these that a very large number of people are going to be handed rather than go looking for. It's clumsy today and neither of you thinks it stays clumsy. And the two questions left are his, whether anybody outside Meta will ever see what it does, and yours, what happens to every business model quietly priced on customers not being bothered.
[32:47] Lexi: So, Monday morning. And I owe everybody a correction first. Last week I said we'd sent people off to run their own queries and check whether they were being cited about twice. It was four times. Episodes 4, 7, 10 and 13. I retired it on air last week and I am not bringing it back tonight, so nobody is getting that homework a fifth time.
[33:08] Lexi: But there is a second question buried in what you told me this week, Harjot, that we have genuinely never asked anybody, and it has nothing to do with citations.
[33:17] Harjot: Yeah, um, so mine's kind of obvious, to be honest with you, and that's usually the reason nobody does it. Um. Take your product, the actual thing you sell, and search it in Google AI Mode the way a customer would, and then just look at what comes back. Not whether you're mentioned. What's it suggesting? Is it suggesting your site? Who is it telling this person to go and buy from? And then run the same query in ordinary search and see where you sit, right, because the comparison is where the information is.
[33:43] Harjot: Um, and it costs you nothing, that's the thing. Like, you can do this measurement today, you don't need a vendor, you don't need a project, and you'll understand very quickly how the machine sees your brand. So yeah.
[33:51] Lexi: And I'm going to make it narrower, because if I send people off to look at a page they come back with an impression, and an impression is worth nothing on a Tuesday. So here it is. Take your three best sellers. Search each one in Google AI Mode the way a customer would. And write down two things and nothing else. Who is the lead seller, and what is the price. Then run the same three queries in ordinary search and write the same two things down again.
[34:14] Lexi: There are three answers you can get before your coffee goes cold. It's you, at roughly your price, and nothing here is urgent. It's you, at a higher price, in which case Harjot's premium is real money sitting on your own product. Or it isn't you at all, and a machine has quietly decided that somebody else is the place to buy the thing you make. Will, yours, because I don't think that one covers what you've spent the last 20 minutes on.
[34:36] Will: Yeah, so mine's very small and it's an hour, it isn't a project. Which is go and read your own returns and refund policy the way a bot would read it. Uh, not the way your legal team wrote it, and not the way a customer skims it on the way to the checkout, but line by line, as a set of conditions that something is going to satisfy. And then ask yourself the one question, which is, if any part of our margin depends on some proportion of these people not bothering to claim, um, what happens when the bothering costs them nothing at all.
[35:01] Will: And I'd bring the hedge along with that, because I don't think Meta's thing is doing this today. So it's a get ahead of it exercise, it isn't a fire. But the reason to do it now rather than in a year is just the cost of it. An hour this week, against a rebuild of your whole offer later. So. Yeah. Cheap day to find out.
[35:20] Lexi: And there's a standing one underneath both of those, which is the reporting split Will described. Get the two surfaces reported separately now, while AI Mode is still small enough that nothing shows up in your blended number. Much cheaper than going looking for it after the volume moves.
[35:37] Lexi: I came into this with a headline about a 21 percent premium, and I'm going out with something I find a lot harder to hold. Because both of you talked me off the price, from opposite ends. Harjot thinks the shelf is landing on whoever has the most evidence sat behind them, and was willing to say out loud that this may well not be the company that makes the thing. Will thinks the study can't demonstrate any of that, and he's right that it can't. Neither of them would bet on the 21 percent. And the one thing they both bet on tonight, without noticing they'd agreed, is that the friction a lot of business models were quietly priced on is about to stop existing. Will, Harjot, thank you both. And thank you for letting me read your own back catalogue at you for an hour.
[36:16] Lexi: Before we go, I have to tell you something. Every voice you've heard today, all 3 of us, me included, it's all AI. But the opinions are absolutely ours, and the judgment behind every word is human. That's the part you can't replace. And that's what Aloudable does. We take what you'd write, turn it into a show in your own voice, and we keep the taste where it belongs, with a person.
[36:37] Lexi: And there's no reseller on this one. It's us, at the only place you can get it. So if you want to hear what that would sound like for your brand, there's a link in the show notes.
[36:45] Lexi: And that's the tea. I'm Lexie Meskouris, this has been Assistant to the CMO, and we will see you next week.