AI drives real retail orders now. Reddit seeding becomes an agency product.

In the tenth episode of *Assistant to the CMO*, Lexie Meskouris speaks with Will Nash and Harjot Singh about whether AI discovery is worth real money yet, and whether you can buy your way into the evidence an AI reads.

Shopify reported that AI-driven traffic and orders to its merchants both tripled year on year, and said AI is a complement to search rather than a substitute. Both men accept the growth and distrust different halves of the report. Will questions the number, a slowdown on Shopify's own earlier figure with no denominator. Harjot questions the sentence, calling the complement framing lazy or disingenuous.

That splits them on money. Will would not move budget yet, since the structures worth optimising for are not settled and big spend now gets wasted. Harjot would move at least fifteen percent of the SEO budget into experiments now. The second half turns to agencies selling Reddit seeding, which Will refuses outright and Harjot would only build in-house. The practical move is smaller than both: write the questions you want to be the answer to, then find who owns the facts you are missing.

Highlights of this Episode Include:

• Tripled, But Decelerating: Will recalls Shopify citing near thirteen times growth in Q1, so tripling is a slowdown, and no denominator is given.
• Complement, or Just Reassurance: Harjot calls the complement-not-substitute line lazy or disingenuous, and aimed at stopping merchants panicking.
• Three Layers of Retrieval: Live search, a crawler-built vector index, and what is baked into the weights, each with very different lag.
• A Model Doesn't Care About Your Brand: Harjot argues live search flattens brand equity, so a day-one brand can compete with an incumbent.
• Fifteen Percent, or Wait: Harjot would move fifteen percent of SEO budget now. Will backs a test but not a yearly escalation.
• The Reddit Deck Nobody Would Sign: Will calls it fake reviews under UK rules, and Harjot gives the tactic six to eight months at best.
• Ten Questions and a Name: Write the questions you want to be the answer to, run them, then name who owns each fact you're missing.

Listen and subscribe: https://AssistantToTheCMO.show

Learn more about Aloudable: https://aloudable.com/

Book a demo: https://aloudable.com/get-a-demo

New episodes of *Assistant to the CMO* land every Thursday.

Creators and Guests

Host

Harjot Singh

Host

Lexie Meskouris

Host

William Nash

What is Assistant to the CMO?

A weekly podcast about AI and marketing, hosted by Lexie Meskouris with Will Nash and Harjot Singh, who build AI for a living. Each week they go through the latest developments and what they mean for marketing teams. They often disagree. About twenty minutes, every Tuesday.

[COLD OPEN - dry, no music]
LEXIE: Shopify says AI-driven orders to its merchants tripled last quarter. Tripled. And on the same call, their president said AI is turning out to be a complement to search rather than a substitute for it. Which sounds like good news, so I took it to the two people I know who actually build this stuff, expecting them to argue about the number. They didn't. One of them went after the number and the other one went after the sentence. And then they spent the rest of the hour disagreeing about the only part that costs you anything, which is whether you move money this quarter or wait. One of them says fifteen percent, starting now. The other one says you'll waste it.
[THEME MUSIC]
LEXIE: 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.
LEXIE: 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.
LEXIE: And this week starts with a number. Shopify reported its second quarter on the fifth of August. Three point six billion in revenue, a hundred and sixteen billion in gross merchandise volume, up thirty two percent. But the bit everyone pulled out was this. AI-driven traffic to their merchants' stores tripled year on year, and AI-driven orders tripled as well. Traditional search kept growing alongside it, still about a third of storefront sessions. And new buyers arriving through AI are ordering at close to twice the rate of other channels. Now, that is Shopify's own reporting and their own attribution, it's global, and it's not broken out for the UK. Will, I'm starting with you. Is that a genuine shift in how people are buying, or is that a company grading its own paper?
WILL: Yeah, so, it's genuine. I think this is uh the continuation of a real shift, and it's it's pretty in line with what we'd expect to see. It's not a surprise to me, it's not a a new thing appearing, it's the same trend carrying on.
WILL: But I think the thing that complicates it slightly is, um, and this is my recollection, so, you know, take it as that, but I'm fairly sure Shopify themselves talked in Q1 about their volume of AI-referred traffic growing almost thirteen times over. So if that's right, then tripling is actually a slowdown. It's still very, very high, I mean three-x is a a huge growth rate by any normal measure, but it's it's decelerating rather than accelerating, and I think that's worth sitting with for a second because the way it gets written up is as this kind of explosive new thing.
WILL: And then the the other bit, which is the one that always gets me, is the denominator's just missing. We're being told it tripled. Tripled from what? Nobody outside Shopify actually knows what the absolute volume is here, so you you can't really tell whether we're talking about something that's now material to a merchant's week or something that's still uh quite small and growing fast off a small base. And those are two very different businesses to be planning around.
LEXIE: Right, so it's real. It's just being sold to me as a launch when it's actually a curve that's already flattening off a bit. And when I put this same report to you, Harjot, you didn't go near the number at all.
HARJOT: Yeah, no, I mean, the number's the number, right? Like I'm not gonna argue with Will on the maths, um, and it doesn't really change anything for me, because the signal is just that AI traffic is growing and it's gonna keep growing, and I don't think anyone in the room disputes that. That's, that's fine. Um. The bit that got me, and I read that quote a few times to make sure I wasn't being unfair, is where they say it's a complement to search rather than a substitute for it. I think that's, um, to be honest with you, I think that's a bit stupid. Like, it's true today. Right now, yeah, sure, absolutely, the two things are sitting side by side. But to state that as a long term thing, that's, um, that's naive, and they're either being lazy or they're being disingenuous, and, you know, it's one of the two, I don't really know which.
HARJOT: If it's disingenuous, then it's a reassuring message, right? It's aimed at their own merchants, it's aimed at the shops, it's basically saying don't panic, your SEO still works, nothing's going away, um, and that's a very comfortable thing to say when your whole business is those merchants not panicking. And if they actually believe it, then, you know, then they're just being lazy about it.
HARJOT: And the reason I think it's wrong is, um, and we've talked about this before on the show, search isn't going away, but search is becoming a tool that the AI uses rather than a thing that a human uses. Right? So the model goes off and it runs the searches on your behalf, it's, it's fanning out queries, it's doing the looking. So of course search volume holds up. Of course it keeps growing. But you're measuring a machine using search, not a person choosing to use search, and those are, um, those are just not the same thing at all, and quoting the one to reassure people about the other is where I think it falls down. So yeah.
WILL: I'd I'd probably stop short of disingenuous, Harjot, to be fair to them. I think, you know, that's that's quite a strong word to put on a company on their own earnings call. Where I do agree with you is that as a long-term statement it's it's the weak part of it. Whether that's them being careful with their own merchants or just, you know, describing what's true this quarter, I I don't think we can really know from the outside.
LEXIE: Hold on, I want to slow down here, because you've just told me the machine is the one doing the searching now, and I've been nodding along like I know what that means. If I ask an assistant to find me a good hand cream and it comes back with three names, what has it actually read to get there? Because in my head it's gone off and looked at the websites, and I've got a feeling that's not really it.
HARJOT: Yeah, great question, and, um, no, you're right to push on it, because it's not one thing. It's not like it goes and reads your website and then decides. There's, uh, there's really sort of three layers going on and they're all quite different from each other.
HARJOT: So the first one is the one people picture, which is it actively goes out and searches. That's the stuff it doesn't know about, right? It doesn't have the answer, so it goes and looks, it fetches, it reads what comes back, and that's live, that's happening in the moment you asked. Um. Then the second layer is there are crawlers running, uh, all the time, and they're building an index of your site. And it's not an index in the old sense of, you know, a big list of keywords, it's a vector index, so it's basically correlating queries to content. It's, um, it's a way of saying this chunk of your page is close to this kind of question. And then the third layer, and this is the one nobody thinks about, is the stuff that's already baked into the model itself. That's from way before, it's from training, and it's kind of a very summarised, very compressed version of that same index. It's already in there.
HARJOT: And the thing that actually matters about that, um, is that those three have wildly different lag. Right? Like the live search is now, the index is, you know, days, weeks maybe, and the baked-in stuff is months, could be years old. So if you're a brand that's been around and you've been well covered for a long time, that just compounds. You're in there, you're sitting in the weights, you get named without anybody having to go and look you up. And if you launched last quarter, you're just, um, you're invisible in that layer. You don't exist in it. And you can't really do much about that layer either, you know, it's, it's already set.
HARJOT: And then the other half of it, which is what's it actually weighing when it does go and read. Um. It can't see your brand imagery. It can't get feelings off a page. It can't watch your video the way a human watches a video and comes away going, oh, I like these people. So what it does instead is it goes at the factual picture. It's looking at specs, it's looking at reviews, it's aggregating the social proof, it's looking at, you know, how long does this hand cream last, is it heavy, what did people actually say about it. So it's, um, yeah, it's all really spec driven.
LEXIE: So if the thing making the recommendation is reading a spec sheet, then somebody has to be publishing a spec sheet it can actually pick up. And that's where you spent a good chunk of our conversation, Will, on what a merchant can even publish into. It did not sound like a solved problem.
WILL: So, it's it's not solved, no. And I think this is the bit that that doesn't really come through when you read the coverage.
WILL: So the retrieval side, um, the actual structures you'd publish into, there are a few of them and they they keep moving. So originally ChatGPT had its own, and there was an ability to actually buy inside of ChatGPT, that was called Instant Checkout. And my understanding is that's now been taken away, that's been removed. Uh, and then Google has this thing called UCP, which was announced at the start of this year, I think, and that that allows you to present a a whole bunch of structured information about your product. And that's probably the closest thing we've got to something established, but even then it's, what, something like twenty large retailers associated with it. So that's, you know, that's not the market.
WILL: And then there's a lot of talk about schemas as well, and the the files, robots.txt, uh those were said to be the right way to do this until really quite recently, and then it it turned out that that doesn't make the difference people thought it did.
WILL: So I I suppose what I'd say is nothing here is fully established. The the specific methodology is just not settled yet, and I I don't think anybody can honestly tell you today which of these is the one that matters in eighteen months.
LEXIE: This is the bit I actually want you two to fight about. I asked you both the same question separately, which is what do I do with my money this quarter, and you did not give me the same answer. Yours is the one that sounds like a no, Will, so you go first.
WILL: It's a no. But it's it's a specific no, and I I want to be careful about that.
WILL: So my position is that this is not yet a strong enough signal for people to be drastically moving budget towards AI discovery. And, and that comes straight out of what I've just been saying about the structures, because if that isn't pinned down then my concern is that large amounts of budget pushed through this now would ultimately just be wasted. And, you know, the last two things everyone was told to do turned out not to matter, so I I don't think that's a paranoid concern, I think that's just the recent track record.
WILL: But, and this is the bit I really don't want to lose, what's not an option I think for most CMOs is to just sit on their hands. That's that's not what I'm saying at all. They need to be in a position where as this becomes more important, they are ready for it, and they're not running to catch up.
WILL: So the shape of it for me is you get ready rather than you move the money. So that's structured data, um, that's starting to run experiments, you know, you might take two similar products and really build out that structured data on one of them and and see whether it makes a difference. But that's, that's a test, it's it's not a reallocation, and I think there's a real difference between those two things.
HARJOT: So Will, I I think I'm on the other side of this to you, um, and I've said this pretty consistently, like in previous episodes as well. My worry isn't the wasted money, my worry is the cost of being late. Because what happens if you wait, right, is you don't feel it for a while, everything's fine, and then, um, you're a few years down the line and people are ahead of you, and you're looking at a decline you can't fight and you genuinely don't know what happened. Like you can't diagnose it, you can't reverse it, you're just, you know, you're left holding the bag at that point. And that's a much worse position to be in than having spent some money on something that didn't work out.
HARJOT: So, um, at the very least you need an experiment budget. And I'd say at least fifteen percent of your SEO budget is a great starting point. Not, not moving your whole strategy, right, at least fifteen percent, with a view to expanding it. And I do think it should grow year on year, because the pie itself is growing, and more and more people are going to shift onto these interfaces. You want to be riding that wave rather than, um, rather than watching it go past. So each year you're putting more in.
HARJOT: And actually, uh, let me go further than what was asked, because I think this is the bit people aren't ready for. The ratio inverts. Like right now we're all sat here talking about moving fifteen percent out of SEO into AEO, and in five years, ten years, I think we're going to be in a position where SEO is the fifteen percent line item inside AEO. It's a subset of it. It's one part of a much bigger strategy, and it's, um, and it'll be a smaller part every year.
HARJOT: But look, on your actual point, I I do agree with you, right? It is unsettled. Like nobody knows what the winning structure is, that's, that's just true. I just don't think unsettled means wait. Um, if anything it means run small things now, so you've got some idea what works when it does settle. And, um, and the other thing is I think you get diminishing returns on some of the SEO stuff anyway, and, you know, a chunk of the SEO work is turning into machine-facing work regardless of what you call it. So some of that budget's already crossing over, whether you've labelled it or not.
LEXIE: Hang on, though, because you two are doing the thing where you sound like you're disagreeing and then you both land on run some experiments. So let me find the actual gap. He just said fifteen percent, Will. Is fifteen percent of an SEO budget a large amount of budget? Because large amounts being wasted was your phrase, and I want to know whether he's inside your line or outside it.
WILL: Yeah, that's that's a fair challenge, and I I should be honest, an experiment budget is not what I was objecting to. Um, if we're talking about fifteen percent of an SEO line as a as a test, a one-off, this quarter, that that's inside my line. That's not the thing that worries me. A test is a test, and and actually I I'd probably encourage it.
WILL: Where I I think we do genuinely part company is on the escalation. Because Harjot didn't just say fifteen percent, he said fifteen percent with a view to growing it year on year. And committing today to an increasing spend on a methodology that nobody has pinned down, that that's a different proposition to running a test this quarter. That's, you know, that's a multi-year commitment made on the back of, uh, what, one quarter of platform-reported data and a set of structures that are still moving.
WILL: And then the the inversion, the the SEO becomes the fifteen percent line inside AEO thing, I mean, look, he he might well be right. I I suspect he's probably directionally right about where this ends up. So it's it's not that I think he's wrong about the destination. I just, I don't know how you'd budget for it. That's, that's my problem with it. I'll budget against what I can actually observe.
LEXIE: But here's the thing I keep snagging on. Everything you told me earlier about the baked-in layer says the big brands are already sitting in there. So if I'm a huge brand, waiting costs me nothing, because the machine already knows who I am. And if I'm a year old, I'm nowhere. So who does this actually favour?
HARJOT: So, um, I'm not walking back what I said, right? Like an established brand is in there. If you've been around, if you've been well covered for years, that stuff is baked into the model and it compounds. That's real, that's an advantage, um, and I don't want to pretend it isn't.
HARJOT: But the thing that changes the picture is that these assistants don't just answer from what they already know. They go out and look. They're running searches, they're fanning out queries. So if I go and say to one of these things, um, I need a pair of trainers, it doesn't come back and go, right, here's a pair of Nikes, because Nike are a good trainer brand. That's not what happens. It goes off and it's asking what are the best trainers in 2026, and if you're a runner then what are the best running shoes, and what size are you, and what's actually in stock and available right now. It's like five, six queries fanning out, maybe more. And then it assembles the answer from what comes back.
HARJOT: So Nike's advantage is real, it's just, um, it's narrower than it looks from the outside. The model knows Nike are reputable, yeah, fine. But it underweights the brand image that Nike have spent, you know, decades and an insane amount of money building. All that stuff about running, about athletes, about what it means to wear them. A human gets that instantly. The model doesn't really care about it, or, uh, it doesn't recognise it strongly enough for it to decide the answer. Because ultimately, um, a model doesn't care about your brand, it cares about what's best for the user who's asking the question.
HARJOT: So to actually answer you, I think it favours the challenger more than people expect. A day-one brand that makes everything super, super optimised for the machines from day one can genuinely go toe to toe with an incumbent that's spent all that time building something the machine can't really see.
WILL: I mean, the fan-out point's a good one, and I I actually buy the mechanism, Harjot. If the model's off asking what's the best trainer for this particular runner rather than just, you know, who's famous, then yes, some of that brand equity does get thinned out in the answer. I I don't have a problem with that.
WILL: The problem I've got is with the middle of your sentence, which is optimises properly for machines from day one. Because that assumes there's a settled thing to optimise for, and that's, I mean, that's the whole point I keep coming back to. Your challenger has got one shot and and not very much money. So if they put it into the structure that turns out not to be the one that matters, they haven't just wasted it, they've burned the exact advantage you're describing, and they don't get a second go at it. Whereas the the incumbent, I mean, they can be wrong a couple of times and and still be standing.
WILL: So I I'm not saying the incumbent wins, I I want to be clear about that. I'm I'm just not sure anybody can act on this reliably yet. And, and if that's right then it it hurts your challenger harder than it hurts them, doesn't it.
LEXIE: Okay, I want to bring this home, because most people listening are sitting in London, not California, and I've just been handed a global number by an American platform. So how much of this is actually landing on a UK desk this quarter, Will?
WILL: Good question. Um, and I I think it matters more than people give it credit for, actually.
WILL: So these are global numbers, they're very high level, and my read is that this is going to be led by the US, with Europe in a a close second. So if you're not in those two markets, I I think it's reasonable to assume this stuff is coming down the pipe, but you've probably got six, maybe nine months more before it really bites on your desk. And the flip side of that, if you are in the US, or or particularly if you're in a tech-savvy part of the US selling to fairly tech-savvy people, then that might be the moment to actually accelerate some of your plans.
WILL: But I I want to be careful, because six to nine months sounds like comfort and it it really isn't. That's, I I'd say that's roughly how long it takes to get ready anyway. So it's, you know, it's the runway, it's not a a period where nothing has to happen.
WILL: And then the the one thing I'd put on the positive side of my own argument, because I I don't want to only be the cautious one here, is that Finkelstein made a point on the call about where that traffic actually lands. So when you compare the AI-referred sessions with organic, the AI ones go straight onto a product page far more often. It was, I think, around half the time for the AI referrals against something like a fifth for organic. So those are approximate, but that gap is is quite striking. And what that says to me is we're talking about quite high intention activity. Um, it doesn't prove there's a long-term return there. I I wouldn't go that far. But it it points that way fairly strongly, and I I think you'd be, you'd be unwise to ignore it.
LEXIE: And what if I'm not selling hand cream? Because a good chunk of the people listening are selling a six-month consulting engagement. There's no product feed, there's no stock level, nobody's leaving a review.
WILL: I'd I'd be pretty blunt about this one, actually. If you're B2B or enterprise, this data is just not relevant to you. It's not information either way. Because what we're looking at with Shopify is overwhelmingly low to medium value goods and services, and it's mostly retail. So if you're selling a six-month consulting engagement and you read that number, you you haven't learned anything, in either direction. Which is not the same as saying AI discovery doesn't matter for you. I I just don't think this tells you whether it does.
HARJOT: I mean, I don't think I'm disagreeing with Will on the number, right? Like he's talking about what that Shopify figure tells a B2B marketer and I I think he's right, it tells you basically nothing. But if you're asking does the underlying mechanic carry, then, um, of course it does. It just carries in a different form.
HARJOT: So you're right that you can't capture reviews, you haven't got stock levels, you haven't got a feed, you haven't got all that public-facing proof sat there for a crawler to go and aggregate. Fine. So the proof just changes shape. What you're actually doing instead is proving to the model that you're an expert in this area, in this topic. Um, so, you know, if you're a content business, you're putting out material on what the different types of content are, why some of them are better than others, when you'd use one over another. And you're doing that so that when the machine goes looking for anything to do with content, it can see you know a lot about it. You keep coming up. You're all over that space.
HARJOT: So it's the same game, it's just, um, yeah, it's expertise being evidenced rather than specifications. That's the thing you're putting in front of it.
LEXIE: And I'll be honest with you both, that's the bit that's been bugging me all week. You're both describing something you can't hand your own answer to. You can only go at it sideways. Which is exactly where the second story comes in, because somebody has worked out how to go at it sideways, and there is now an industry charging for it.
[TRANSITION STING]
LEXIE: The Verge reported on the fifth of August that brands and a growing pile of agencies are seeding Reddit threads specifically so ChatGPT and Gemini will name them. Skincare brands, record labels, weight-loss apps. Researchers at Cornell Tech found as few as thirteen words in a Reddit comment can swing which product an AI recommends. And Reddit is hunting it with its own models, something like twenty-five thousand spammy posts and comments a day. And what I don't get, Harjot, is why a machine that just told us it wants specs and facts cares what a stranger says on Reddit.
HARJOT: So, um, to be honest with you I don't really think of this as a separate thing at all, right? Like this is just an extension of the picture I was describing before. The model is building a holistic view of you out of multiple data sources, and part of what it's doing is deciding how much trust to give each one, how much weight to put on it. So the question isn't really why does it care about Reddit, it's, um, it's what does Reddit look like to it compared to everything else.
HARJOT: And to the model, Reddit is a third party source. That's the key bit. Your website is you talking about you. Your product page is you talking about you. Reddit looks like a secondary source of reviews that hasn't got a misaligned incentive sitting behind it. It's, you know, it's a bunch of people who've got no reason to sell you anything, apparently, having a conversation about whether the thing is any good. So it carries weight for exactly the reason you'd expect it to carry weight, um, which is that it looks disinterested.
HARJOT: And then the other half of it, which is much less glamorous, is it's just easier to scrape right now. Like it's structured, it's accessible, there are deals in place, it's, uh, it's cheap to get at compared to a lot of the rest of the web. So it's heavily used partly because it's good and partly just because it's there. And that could change, you know, that may well change.
HARJOT: The other thing I'd say is this is particularly important for D2C brands, and to be honest with you a lot of them should have been doing this sort of thing anyway, like proper community curation, that's not new. I think the difference now is it makes sense for people in B2B to start doing it too, which it maybe didn't before.
WILL: Yeah, and I I'd add to that, from from where I sit, the assistants do seem to have an existing preference for user-generated material generally, and and particularly for Reddit, and and Wikipedia to some extent.
WILL: But I I want to be a bit careful about what kind of preference that is, because I don't think anybody sat down in a room and decided this. It it looks to me like a natural preference, or or an emergent one, let's say, rather than a a policy.
WILL: And and it turns out that this can be exploited, um, would would you believe it. Which, you know, is not a new story. We we had exactly this in the SEO world for years. It's just far, far worse now because, um, you know, before you had ten blue links to spread it across, and now there's, there's one answer. So the same trick goes an awful lot further.
LEXIE: Let's make it the thing it actually is, though, which is a purchase. There is an agency out there right now with a deck and a price on it, and it's going to land on someone's desk this quarter. Will, do you sign it?
WILL: No. No, I I wouldn't sign it, and and the first reason isn't a commercial one, it's a regulatory one.
WILL: So, I I want to be careful here because I'm I'm not a lawyer, but my understanding is that in the UK this falls foul of the fake-review rules. Because what's actually being described, if you strip the language off it, is is creating fake reviews. That's that's what it is. Someone purporting to be a real buyer of your product who isn't. And my understanding is the US is is broadly similar, though I I'd be less confident on that half of it, so, you know, take that with with a bit of caution.
WILL: And then on top of that you've got the the commercial argument, which is that it's a a brilliant way to waste a bunch of money. And I mean that fairly literally. You set the campaign up, you you pay a vendor to seed it, you wait, and then you find that the money's just gone. Because, um, the CMA may well find out about it, and and again I'd hedge that, I don't know what their appetite is here. But your competitors may well discover it. And and your competitors have got every reason to report you. And what struck me about the the research on this is that the manipulation is fairly obvious. It's fairly easy to detect. It's it's not a subtle thing you're doing.
WILL: And then, even if none of that happens, even if you get away with it entirely, you've you've invested in a a very short-term trend. Um, something with an expiry date on it, and and money you don't get back. So, no. I wouldn't sign it.
LEXIE: When I put the same thing to you, Harjot, you didn't hesitate. You said yes, a hundred percent. And Will has just called it fake reviews, so I want to be really clear here. Are you buying the thing he's refusing to sign?
HARJOT: Yeah, no, I mean, I did say yes, and I'd still say yes, um, without hesitating. But I think what I said yes to and what Will's saying no to might not actually be the same thing, so let me be really specific about it.
HARJOT: The yes is to running a Reddit programme, right? Us, on Reddit, talking to people. What it would not be allowed to do is give information that's basic and non-contextual. Like that's the disqualifier for me. It has to be providing something that's a real learning, something that either of us would be able to stand behind and go, yeah, that's ours, we found that out. Which means, um, it has to have tapped into our own knowledge and our own experience somehow. It has to have come out of our heads. And, you know, we've spent a long time now on voice cloning, on training pipelines, breaking things, finding out that the fifth regeneration is no better than the second one. That's the stuff. That's what you'd be putting into a thread, and only we've got it.
HARJOT: And that, that's actually the reason I struggle a bit with the agency version of this, because, um, an agency hasn't got our knowledge and our experience. They can't. So the thing I'm saying yes to is a thing that, that isn't really purchasable, if that makes sense. It's not that I've, um, looked at their deck and didn't like the price. It's that the bit that makes it work is the bit they haven't got. They can't have it.
HARJOT: And the framing I keep coming back to is it's got to be more about giving than getting, which, um, I know sounds a bit cheesy. But the giving part is real. You're sharing something that genuinely helps that person with the actual problem they've got, not, not something engineered so your name appears in it.
HARJOT: So the test, um, the test is very simple really. If a human wouldn't find this useful, then the value you're providing to the AI is lip service. It's paper thin. And, you know, it might work today. Fine. It's probably not going to be durable.
WILL: The distinction you're drawing is is the right one, Harjot, I I think. Although I I'd say it's not really a builder's insight, is it, it's it's fairly superficial once you actually say it out loud. If you're going onto Reddit, or sending bots onto Reddit, to purport to be real buyers of your own product, that's, you know, that's quite clearly manipulation. Nobody needs a a framework to work that out.
WILL: And on the other side there are plenty of well-known ways to engage with people on Reddit that are perfectly legitimate. So the the direct one, which I I actually think is the less powerful of the two, is the AMA. You put a senior leader up, they answer questions. Fine, that's, that's a real thing.
WILL: The one I think works better is the indirect version, which is you you go to a relevant subreddit, or you you have your own, and you post genuine updates. Changes to the product, changes to the software, things like that, and then you legitimately ask for feedback.
WILL: Um, and there's a caveat on that which I I think is the best part of it, actually, which is that you have to be reasonably confident the responses are going to be positive. And that's going to vary a great deal company by company. But, you know, hopefully that keeps you honest as well. If you can't put your product in front of a room of your actual users and ask what they think, then, uh, then you've probably got a bigger problem than than your Reddit strategy.
HARJOT: And the asking for feedback bit, um, that's the right instinct, Will, because that's the thing they genuinely hate on there, right? Straight up adverts sat in the comments. Like you'll get shut down very quickly, and, um, and more so as these models reviewing the comments get deployed across everything. So, you know, don't be naive about it.
LEXIE: Okay, but now imagine everyone listening does this. Every brand turns up in the threads being helpful. What does that look like a year in?
HARJOT: Yeah, this is the bit that worries me, um, and it reminds me of something quite specific actually. Do you remember the old tech support forums? Like the Microsoft ones especially, their own one. If you had a problem, you'd search it, and you'd land on this thread, and there'd be, you know, someone with a badge under their name, and the answer would be, um, have you tried restarting your machine. Or have you run Windows Update. Something clearly obvious that anybody could have told you and that you'd obviously already done before you went looking.
WILL: The badge under the name, yeah.
HARJOT: And, and the thing is those people weren't malicious, right? They were just incentivised to reply. That's where their incentives sat. They needed to be answering things, so they answered things, and they provided, um, no value at all. Zero. Every single one of those threads.
HARJOT: So what happens a year in, if everyone's turning up being helpful in inverted commas, is you get a trust collapse. People stop trusting you. Like they're not neutral about it, they're actively repulsed by your comment, because it's patronising, right? It's you turning up to tell them something they already knew, with your company name underneath it. And at that point it's unhelpful and it's just spam, that's, that's all it is.
HARJOT: And that is exactly the thing you don't want to be doing. It's what you get when the incentive is to appear rather than to actually help. Like the moment the goal becomes being in the thread rather than solving the thing the person came in with, um, yeah. That's, that's how you end up there.
LEXIE: So let me take the other side of it for a second. There are tools now that tell a CMO they show up in forty percent of answers for a set of questions, and that number goes in a board pack. But if thirteen words in one comment can flip an answer, that forty percent could move next week for reasons nobody in the room controls. Is that a number you'd let someone report?
HARJOT: Well, the thing is, um, that isn't the relevant metric in the first place. Like before we even get to whether it's stable, I don't think it's the thing you should be looking at. And if you focus on it, then yeah, you're going to get swung around by it, because you've picked a number that moves for reasons you don't control and then you've put it in a board pack, so now you're explaining it every quarter.
HARJOT: It's a bit like, you know, in the older days going, well, I appear in this many mentions in terms of backlinking for SEO. Um. Right? Like it's just the wrong level of granularity to be looking at. It felt like a number because you could count it.
HARJOT: What I'd actually put in front of a board is the traffic that's landing from AI sources, and then the follow-on actions off the back of it. Did they go on and buy. Did they do the thing you wanted. So really it's, um, it's the difference in intent, and it's the growth in the channel itself. Those two. That's a much smaller, much more boring number and it's the one that's actually telling you something.
HARJOT: And, uh, the second thing, and this connects back to what me and Will were going round on earlier, is that this instability is precisely where the challenger gets an opening. Because the incumbents are very, very focused on a performance-based approach, right? Everything has to be justified, everything has to be measured before it gets funded, that's how a big organisation works and, you know, fair enough. But given how immature the measurement is here, the investment is just a lot harder for them to pitch for. Like the person inside that business who wants to do this work genuinely can't build the case, because the instrument doesn't exist yet. So they wait. And, um, and by the time it's easy to prove, the window's a lot smaller.
WILL: Yeah, and mine's a slightly different kind of measurement to yours, Harjot, because what you're describing is a dashboard, something you take to a board and and show growth on. Mine's more of a precondition.
WILL: So before you spend anything on Reddit at all, I I think you go and find out what Reddit already says about you. And almost nobody does that. They they go straight to the campaign.
WILL: And the tools exist. Reddit have their own offering, I I believe they call it community intelligence, something like that, don't quote me on the name. And then there are third-party commercial services that will, for money, give you a read on on community sentiment. So it's it's not hard and it's not expensive relative to what you'd spend on the actual programme.
WILL: But if you're not already measuring that, then throwing money into any kind of interaction on there is is almost certainly dead money. You just don't know what you're walking into. You might find there's a a thread from two years ago doing all the damage, and no amount of new posting fixes that. So I'd I'd put that underneath your number rather than against it, because, um, without it you don't really know what your number's telling you.
LEXIE: Alright, last thing on this, and it's the bit I'd actually want to know if I were signing something. Reddit is now hunting this with its own models, so it's machines against machines. If somebody buys a twelve month programme today, what have they actually bought?
HARJOT: Yeah, I mean, the seeding stuff is going to get you far enough today. Um, probably for the next six to eight months. Yeah, maybe even a year actually, thinking about it, I'd probably stretch that. Long term I don't think it lasts, no. But it will get you a foot ahead, and I want to be clear about that because it's not nothing, right? Like it's a real advantage for a real window and I'm not going to pretend otherwise.
HARJOT: The thing to be aware of is it's a one and done type situation. It's not a thing you build on, um, you don't compound it, right? You do it, you get the bump, and then, um, then that's spent.
HARJOT: And there's a tail on it as well. If you overdo it you can genuinely create a bad rep for yourself, and that's, that's especially true in smaller communities, because people notice. Like in a big subreddit, maybe it washes through. In a small one they, um, they know exactly who's turned up and what they're doing. So, um, yeah, it's got to be done tastefully and properly, or don't do it.