The Economist is working on a website for the 'two-track web' (one for us, one for AI)

Amelia (00:02.381)
Hi there, I'm Amelia.

Paul Wood (00:04.648)
I'm Paul.

Russell (00:06.126)
I'm Russell.

Amelia (00:07.629)
And this is Fin the Week. How is everybody?

Paul Wood (00:12.353)
Doing very well, yeah. The end of another week — although we'll let everyone in on a secret: we're recording this on a Monday morning, not a Friday afternoon. But we can pretend it's Friday.

Amelia (00:22.861)
Feels a bit weird, doesn't it?

We'll get that Friday feeling going. But today we've got a bit of a timely follow-up to our previous episode. Last week we discussed the recently published GEO guidelines from Google, and in that discussion, Paul, you used The Economist as an example. Since then, it just so happens that we've stumbled across a real-life case study from none other than The Economist. So Paul, what is this story?

Paul Wood (00:55.861)
Yeah, 'cause last week I went on a bit of a rant about Google's GEO guidelines and said that I didn't think it was really that useful. One of the things I said it missed out on was the idea that some website owners might want AI search bots to find and surface their content, but they might not want AI training bots to come and hoover up their content and use it for their own goods.

The example I used was, as you say, The Economist. Being a publisher, they rely on people coming to them and reading their content and becoming subscribers. So they want to remain findable, but they don't necessarily want all of their information to be repurposed by an AI chatbot. After we recorded that, I was reading some tech news online and stumbled across an article on the Digiday website.

It was about The Economist talking about working on a concept for what they've called a two-track internet — one for humans and one for AI agents. They're working on this concept that they're going to surface select bits of content that are suitable for AI agents and keep the important stuff behind the scenes for the humans.

They're testing out different methods of doing this, seeing how it works. So it was a timely bit of news and something I think is worth us digging into today.

Amelia (02:36.201)
Absolutely. This is going to be a bit of an exploration into the concept of a two-track internet, which we've kind of mentioned before, but shall we start with what the article itself actually has to say?

Paul Wood (02:49.214)
Yeah, so there are quite a few things covered in what's a relatively short article, but the crux of it is that they've described the concept of a two-track internet as a strategic shift. They're saying that a lot of the challenges that publishers, and website owners in general, face with how to deal with increasing levels of AI-driven bot traffic — some of it good, some of it not so good — could be solved by thinking of your website as having two tracks.

It covers topics like content gating and paywalls. These are terms people use to talk about when you might have high-value bits of content hidden behind a paywall or gated behind a form. The Economist again is a great example of this because, as a publisher, their product is the information that they publish and the journalism that they produce. To have access to it, you need to be a subscriber. So they have an actual challenge to figure out: how do we keep our product behind a paywall but at the same time still get referenced in search results and in AI chats when somebody asks about something that we know about?

They have to overcome this challenge, and it's becoming more of a challenge now because AI chatbots can reliably — well, it depends on who you ask, but I'd say reliably — reproduce a lot of the information that publishers promote. But this subject is also relevant for financial services firms in general, because it's not uncommon for finance firms to have content that is gated for one reason or another. It might be that they'll have a piece of proprietary research that is used as a way to attract new potential clients, and so you might hide that information behind a form. But at the same time you're going to want the wider world to know it exists and to know a lot of the takeaways. And so there's a real challenge there.

Paul Wood (05:13.546)
How do we publish this stuff, keep it for our own benefit, but make sure an AI chatbot can still understand it and surface the key takeaways — without keeping it completely open so that the AI chatbot just hoovers it up and repurposes it all? The article is talking about that, and about how The Economist is trying to tackle it with this concept of a two-track web.

Russell (05:48.736)
Yeah, it's interesting, isn't it?

Amelia (05:49.069)
So this article — as I say, last week we were talking about the Google guidelines, and this was the big question, wasn't it, Paul? The Economist was the example that you used. On first glance at this article, do you think anyone's any clearer?

Paul Wood (06:12.854)
No — the tone of this is very experimental, I would say, reading it. But they're definitely pitching it as a real thing they're working on — again, the two-track web is what they've called it. There are some quotes from Josh Muncke, who is the VP of Generative AI at the Economist Group. He has said a growing share of B2B buyers now start their journey with an AI chatbot — so ChatGPT, Gemini, Claude — and he has said agents want clear structure, questions and answers, ideally text. So not carousels, not art features — all of the stuff that makes a web page look interesting and engaging, I suppose.

Simply by saying that, he's describing a web where you have something that's engaging to look at and view as a human, with the needs of an AI bot being slightly different — they just want the plain facts, easily crawlable. But it is experimental, 'cause I don't think any of this stuff is proven. Something we'll get into in a bit is how they're actually tackling this at The Economist, by building what you could call a sandbox — a test scenario where they've got their own AI chatbot, their own AI solution in the background, where they can test their publication's approach. How does our AI chatbot use what we're doing? How does it view the approach we're taking? All before they let it out into the world.

And I know, Russ, you've put some thought into the concept of: does a human user have the same needs as an AI bot user, if you can put it that way? You'd written something about this concept in the past, haven't you?

Russell (08:21.27)
Yeah, so I was calling this AIX — essentially, instead of user experience, it's AI experience. As a designer, do we need to be thinking about AIX and UX as two separate concepts? An AI is essentially a persona visiting your website with a completely unemotional response to everything it sees. It isn't going to be distracted by banner ads. It won't be interested in other services. It will have a kind of sniper-like A-to-B goal, and it will find the quickest way to achieve that and carry it out on your website. It won't be influenced by colours or by language which might be persuasive to humans.

This seems slightly different because it's going a layer beneath that, isn't it? It isn't going to be looking at the front end of the website like some of these earlier projects were. OpenAI had Operator — you'd give it a task and then it would go to a website and carry out that task. It would use forms, it would find information. If you wanted to, say, compare savings products, you could ask it to do that and it would go to the website and look at the site like a human, actually scroll down the page and use it. And you can actually see it. I think it was…

Russell (10:14.112)
Google DeepMind had Project Mariner, and you could see its thought process. It was saying, "I'd scroll down to the bottom of the page. I can't find what I'm looking for. There's a pop-up, for example, obscuring part of the page so I can't find it." It would actually be thinking like a human, exploring the website, which is very different to what we're saying here. The way The Economist is saying it's going to work is almost like an RSS feed — RSS feeds use XML, structured XML — and it's almost a similar way of looking at it, where you present structured data for the website that can be read by AI, so it doesn't have to go through all of those interfaces and can just pull the data. It can find quickly what it needs. And there's definitely a speed aspect here as well, because if you've got an AI agent going through the front-end site, that's quite a clunky, slow experience trying to find what it's looking for — booking restaurants, comparing products and all those kind of day-to-day activities that people would be using it for.

I think when it's just finding information, that's where it works really well, like that RSS example, because it could just pull content — like article content — back into what it's surfacing for the user. When it's going through journeys where various information needs to be input — people are calling this "human in the loop" — let's say you needed to make a payment, for example: would you trust the AI to do that on your behalf? I think you'd probably want a pop-up to make sure it is you making the payment, using the right card.

Russell (12:15.682)
And that you want to double-check the amounts. If you're making transactions between accounts, for example, would you want to do that through your AI agent? Or would you feel more comfortable doing that through the company's app? So is there going to be a transition where, if there are two internets, that becomes everyone's preferred way of interacting with all these different apps and all these different accounts that you have? Would people prefer still to go onto their apps and trust that the brand is surfacing exactly the right content, where you can validate everything via that app? Or will we get to a point in the future where that kind of world of the internet would be phased out and people would just start managing their lives through one source?

Russell (13:18.882)
We might be moving towards a place where everyone just manages everything through one source. Which is quite interesting, because if everyone could consolidate all of their apps — because all of that data is being poured into one central point — then that's quite interesting from a management perspective for people's lives, which you can't really do with the internet and the way that all the apps are structured at the moment. I think AI opens up that possibility: to have your mortgage, your banks, your insurance or your utilities all drawn into this one central management tool, using this underlying layer of data, with everything being pulled into there.

I think that's perhaps where things are going, and starting off with this second world of just pure data is probably a step in that direction.

Amelia (14:26.785)
That's maybe where it's heading in the long term. Thinking more immediately about this two-track internet from a design point of view — because you think about AI streamlining everything, making everything quicker, faster, but to me this sounds like double the work. Is that double the work for you, Russ?

Russell (14:46.126)
Well, I have given this a lot of thought, and I think an AI agent visiting your website is essentially going to be going through tasks in a similar way that a human would — perhaps being less taken off track. So is it a different persona? I think structurally there are more considerations to make, especially with anything obscuring the page, because a human might look at a pop-up and close it; an AI agent might really struggle with that. But I think it just comes down to creating a really accessible experience, whether it's for an agent or whether it's for a human. Making sure that you use the correct headers, for example, so an agent can see that this is the title of the page, then this is a secondary header, a third header, a fourth header. You need to be able to easily go backwards and forwards between steps for different flows, like registration flows or onboarding flows. I think it's the same kind of mechanics, but in my head I'm thinking an AI agent would be better not using the front end, because that feels like a very human thing to do — essentially it isn't interested in all the things that a human would be.

And the business needs to look at it in a slightly different way as well. What's the point of producing all these beautifully styled designs and interfaces for an AI agent who only really cares about that end result and getting that information at the end? So it makes perfect sense for it to have a completely different experience to a human. The only reason it would need that is if you trained it to.

Russell (16:57.538)
Because if you trained it to want an experience like a human, it would probably seek that out and explore it. But all you need to do is train the AI to not be interested in that, and it wouldn't be. It's completely different — it just thinks in a more binary kind of way.

Paul Wood (17:15.624)
Yeah, 'cause the stuff you were just discussing there gets to the nub of it. What The Economist seems to be pointing at here is that two-track web where you have a front end which is designed for and used by humans, and then a sort of alternative bot version, which I think they're saying would be based on simplified journeys — probably using technical solutions like what we discussed last time: llms.txt, a file that guides bots around a site and provides markdown versions of content, which in layman's terms is just simple, text-based versions of the content and the journeys that you're preparing. So that's the true two-track approach that a lot of people are talking about online at the moment.

But then the alternative, which you alluded to a bit there, Russ, is what I suppose you'd call a one-track web — which is what we know and use now — where you have normal websites and you build agents that can work with those normal websites and use them. So the question for a website owner at the moment is: do you build for the two-track system, where you're basically reproducing key bits of content in markdown format, let's say for the sake of argument, so that you end up with a human layer and a text-based layer that sits underneath it that only a bot would access? Or do you work on your front-end site to make it as clean and usable and accessible as possible, so that either a human or a bot could effectively use it? I don't think there's an actual answer to that, but I would imagine the two-track approach has a lot of benefits simply because it's more efficient. Because AI systems use a lot of resource, energy and — well, yeah…

Paul Wood (19:38.408)
It basically all comes down to energy, which equals cost. Processing a markdown document is cheaper than processing a complicated mixture of HTML and JavaScript and whatever else happens to be on the page. I think simply for that reason the markdown bot-version approach makes an awful lot of sense, and I think that's what The Economist is getting at here as well.

Russell (20:07.982)
Yeah, it's interesting, and I guess you could use AI to turn your website into that fairly easily these days, because AI is just really, really good at crunching information and turning it into something structured if you give it that initial structure.

Russell (20:33.676)
So if you outlined what the structure of that second version of your website needs to be, you could either get it to crawl your website and create that, or, if you had access to the repository of your website, it could crawl the files in that repository and probably produce, fairly quickly, a clean version of your website for that second version. And in a lot of these platforms, like Bing, like Search Console — in the same way you upload an XML file — perhaps you could even upload that second version of your site to make sure it's being crawled properly and being seen.

Amelia (21:18.656)
That's an interesting point, because I was going to ask this from the perspective of somebody who has no clue about this. For somebody who's listening now, who owns a business and is thinking, "Oh my God, I've seen the Google guidelines, I've read this article — what am I meant to do with my website now?" Because to me, this sounds like a huge job, a massive project, to make sure that your website is acting accordingly for AI. Is that the case? You say you can use AI to help do that.

Russell (21:45.551)
It depends what your website does as well. I think where it's going to struggle — and this is different to articles — is when there are complicated processes, like a savings account application form, something like that, which could be a number of different steps. Are we just saying that this second version of the web is more information-based than, say, transactional? Because I think when it comes to forms and those human-in-the-loop processes I was talking about, it'd be interesting to see how it copes with really complicated processes on the website, and how it surfaces that back to the human. Is it pulling all those forms and surfacing them back to the human for that human-in-the-loop validation within Claude, within ChatGPT, within Gemini or wherever that initial request came from, and recreating the website so the human can use it?

I don't really know how it's going to do that, because it isn't going to the website and carrying out those actions on your behalf like I've seen some of the other agents doing. So I get it for articles and content and research — that absolutely 100% makes sense. But it'd be interesting to see how we move towards that second version of the web for those more complicated forms and processes. That area needs more thought.

Paul Wood (23:22.289)
I think the answer is quite similar to when you think about the world of accessibility — optimising a system for accessibility guidelines. There are quite a few clear guidelines about how something should work, and it's usually just a process of simplification in many ways. Colour contrast, for instance, is often about keeping stuff clean, clear, uncluttered — I say colour contrast, but that applies to everything. I wonder whether the way to optimise for an agent browsing a system, going through a form or something, is really about simplifying the process. So you almost converge on something, because humans respond well to simplified, uncluttered processes, and bots will respond in the same way. You're moving in the same direction, but perhaps from two different angles.

But also, as you were talking, it reminded me that in a previous episode we talked about the concept of design, and Russ, you mentioned that really fundamentally design is about solving problems. Design isn't necessarily about how something looks; it's about how a solution is pulled together. So if you think about this challenge that The Economist is facing — and that everyone really will eventually face — it's: what are the problems that we need to solve? And they're looking at it from the perspective of two users. You've got the humans who use their site and their services, and then you've got the agents and the bots who use it.

And I suppose when you look at it from that perspective, the problems are slightly different between the two groups. How you communicate a news story, say with lots of data in it, to a human is probably different to how you communicate it to a bot, because a human will need that data to be visualised in a smart way. Like, for example…

Paul Wood (25:44.476)
Russ, you've got a diagram behind you there — I think it's music genres or something? That's a smart way to show a lot of data to a human. Whereas with a bot, you could just give it the spreadsheet and say, "There you go," and the bot would get it straight away. And the AI chatbot would then produce that output for the human that's using it. So from that perspective, you look at what The Economist is doing with this two-track system.

Paul Wood (26:13.991)
That makes a lot of sense, because the human side of the site has a lot of the visual clutter, I suppose, to make information understandable. The bot version is plain text: it's clear, it's just usable by a bot. I think that approach makes a lot of sense, and I think companies in general will probably have to start thinking along those lines.

Paul Wood (26:42.663)
Especially if you're dealing with a lot of data or a complicated journey: how do you filter it right down to its core elements for a bot, and what are the add-ons that a human needs to make it easy for them to consume? And I suppose you want to work towards a happy medium where, to avoid too much work, you're getting the core elements that work for both.

Amelia (27:14.541)
So shall we talk now a bit about trust? Because we know that AI agents can drive that discovery, but can they drive trust? And of course, if you look at something like The Economist, they rely so much on subscriptions, which in turn rely on that trust. How does that work, do you think?

Paul Wood (27:35.473)
Yeah, I think this is the big challenge. So one of the other things that comes up in the Economist article — and I don't know if this is straight from The Economist or whether this was the writer giving their view on it — is that part of the challenge The Economist has is that they have a really clear tone of voice. They're kind of famous for it.

And so they will have a fear that even if the content they want an agent to understand gets referenced — say an AI chatbot says "The Economist says XYZ" — they want to make sure it is in their tone of voice. They don't want it muddled in a different tone of voice. And the reason the tone of voice is important is really because it's a trust signal. They put a lot of effort into having a tone of voice, a way of communicating, so that their readers know that tone of voice and trust that tone of voice. I can't remember what the original question was, but essentially — yes, an AI agent can definitely get you to a place where you see the information and you understand it. Can it…

Paul Wood (29:02.194)
Can it then help you start forming a relationship with a brand? I'm not sure it can, no. And actually, I ran a few tests. I searched for a few things — one was that I mentioned a particular firm and said, "Give me this firm's take on US inflation so far this year." And the chatbot — in this instance, I think it was Gemini — gave me a good analysis, but it didn't provide any citations. It didn't mention any brand. So even though I specified which brand I wanted, it didn't cite anything from that brand; it just said this is what that brand thinks. And then I pushed it with a further question and it still didn't highlight the source of what it was saying. So that's a good example of the chatbot putting words in the mouth of a brand. It might have been accurate to the voice of that brand and what that brand actually wanted to say. But equally, it might not have been. It might have just been a random string of words that made sense to me as a reader, but didn't give me any discernible sense of the voice of that brand.

And that's a big issue, definitely for brands like The Economist and publishers whose tone of voice is a key part of their value proposition. But it's also an issue for any finance firm who needs to be credible, authoritative and recognisable in a crowded market.

Amelia (30:40.269)
Sorry, I was going to say it's quite interesting — talking about The Economist, in the article they were talking about these little pods, these teams they've put together consisting of a designer, an engineer, somebody who works in production. But they're also putting editorial staff directly within those teams, specifically to ensure that they've still got that look and feel and their tone of voice. And when you look at a publisher like The Economist, you think, well, yeah, they've got the resources to do that — but does that mean we're going to get so much less from smaller publishers, or independent writers who want to have their say?

Paul Wood (31:21.094)
Yeah — and I think even for companies, because one of the modern tropes online for marketing is that everyone has to be a publisher. So if you are, I don't know, a fund management company — or you've got fund managers who produce commentaries on the funds that they're managing — you have to think of yourself as a publisher who's providing that commentary.

And it's really important that what they say is in a particular tone of voice and does have the nuance that's needed to clearly communicate something. One of the fears with the way AI chatbots and AI systems work is that they strip out a lot of that colour that comes from someone's particular approach to writing. They're quite formulaic. And everyone will have seen this in practice when they've asked AI to write them something and it comes out factually fine, but very dry and lacking in any sign of emotion. So it's a fear that I think brands have.

And The Economist, like you say, have pulled together these teams including editorial staff, and part of the reason for that will be that they build their own internal AI chatbot to test what happens. So they'll query their own sort of database, if you like — their website basically, which is their archive — and say, "Find me everything. What did The Economist say about XYZ news story?" And their AI chatbot will come back with what it can get from their own setup. They're testing that to, first of all, figure out: is it accessing the stuff we want it to? But secondly: is it pushing that out in a format that matches our tone of voice, or is it just reducing it down to dry, basic language that doesn't suit us? Having read the article, it sounds like that's the stage they're at — they're testing this internally before they roll out a sort of two-track system approach.

Amelia (33:42.933)
I mean, you said it yourself, Paul — this wasn't a super long article, but there is so much in it, so much to think about. Looking at it overall, and thinking about the conversation we had last week about the Google guidelines, what should brands, firms, companies be doing now, moving forward, in the short to medium term?

Paul Wood (34:04.242)
It's a tough one, because even in the last week I've seen multiple competing posts on LinkedIn, with a group of people saying you should read Google's guidelines — they say you don't need things like an llms.txt, you don't need a markdown version of your website, a kind of plain-text version of your website. And then on the other hand, I've read other work where people have done a bit of research and they found that markdown versions of content are surfaced much more frequently in AI chatbots — so that completely contradicts what those other people have said. And even in the case of the Google document we discussed last week: that document said you don't need to worry about that stuff, but the Google Chrome team have again updated their auditing tool — it's called Lighthouse — and that specifically mentions llms.txt, gives you guidelines and points you to the organisation that oversees it. So it's just very confusing.

And so for a brand, for a company that's running a website — and thinking of it from our perspective as an agency, what do we advise clients? — what I would say right now is you have to just be pragmatic with it. Come back to the design thinking that we touched upon, which is: what's the problem that needs solving? Then go and solve that problem. I think the problem for most content-heavy brands is you do have a set of human users and you do have a set of AI agents that are currently accessing your content — and the AI agent side is only going to increase. That's not going anywhere. So I think you do have to start thinking about how humans and agents interact with your content, and how you can make that as clean and usable as possible. And for some organisations that means: actually, you've got a really well-structured website, it's not complicated — so just focus on keeping that clean and efficient.

Paul Wood (36:29.029)
But for others, it might be that you do have quite a lot of complicated content and a complicated journey, and therefore thinking about an AI version of your site — like a sort of markdown version of your site with a specific journey for AI agents — is probably not a bad thing to start thinking about. I wouldn't bet the house on it, but I'd probably start doing some testing — maybe building your own internal AI chatbot with a caged-off version of your site that you can tweak, play around with and test, and say: if I use this AI agent to query our database of historic content, what does it come back with? What does it find? Is it what we want it to see? Because it's definitely not a clear picture at the moment.

One other thought that occurred to me while I was reading through all of this was that a lot of the Google guidance was talking about search in the model that it currently works. So currently a search engine crawls the web, builds an index of content, and then uses that index to surface search results when people request them. But the way that AI agents are moving, real-time search is going to become a thing in a way that it's not currently. So rather than querying an existing index, I can see a day when you would go to your personal AI agent and say, "I want to search for this — go and look at these things," and it would go and look at stuff in real time. It wouldn't go to an existing index; I think it would visit sites and go direct to the source. And in a world like that, efficiency is more and more important, and therefore having a simplified AI version of your site again is a huge benefit.

I'm not saying any of this is happening right now, but it seems to be the direction of travel. And again, it points to Google's document last week being not that useful, because they're talking about what I think is a dated model. But ultimately, design thinking is the direction I'd go in — which is a good segue into Russell…

Russell (38:47.832)
Yeah, I think a lot of brands are struggling with the concept that they're going to be represented — perhaps as they may or may not want to be represented — as articles and as their websites. Not only the information surfaced in these chat tools, but also potentially the browsers changing the way that the interface looks as well — as we talked about with the Google patent for generative UI. It feels like a lot of people are trying, as fast as possible, to control how their brand and how their voice is being received in this new way of people trying to find them, really.

It's a completely new layer. If you look at the user journey, previously it was a search engine, and that would generally pull the information from the website without rewording it — back in the day, you'd have your metadata, really; that would be what it would bring into the search results. Now, in terms of the research journey or the user journey for a website, there's this unpredictable layer of AI, which can hallucinate, can reword articles, can represent your brand in a different way. So, like you say, these businesses are putting dedicated teams together and trying to recreate what these AI chat tools are going to be surfacing to the user, desperately trying to retain control of their brand. It's a really interesting transition. I still believe that personalisation is one of the most exciting things about…

Russell (40:58.414)
…AI for everybody, especially for those with access needs or other needs. Talking about forms and really complex user journeys — if someone's using Gemini and it can recreate all the forms in the way that suits that user, then why would that not just happen within Gemini, instead of an augmented experience where Gemini would essentially rebuild the website? It's going to be one of the two; I don't know which way it's going to go. It will either be recreated in that one central place where you've searched — making sure, for that user, it has a bigger font size, let's say, so they can read the form labels and they can read the copy better. Perhaps it distils everything down to just black and white, for example, because that's their preference, and it almost just lists everything out and the user goes through it in that way. Or, for some journeys, it would take the user to the website and it would recreate it and create an augmented experience which would be suited to that user. Perhaps it's both.

It's creating those possibilities, which is really exciting. But I see all the big brands fighting to stay in control of their brand, and that's why we're talking about this two-track internet thing.

Paul Wood (42:34.019)
Sorry, I was just going to say — I ran a couple of other very basic tests. This isn't scientific at all, but one thing I searched for was particular fund performance, and an update on the approach that the fund manager had taken — again using Gemini. And again, it gave a good overview, but it didn't once give a citation for the brand that actually runs that fund, or mention the fund manager.

What it did do is pull from commentaries that came from aggregator sites and news sources, which would suggest to me that the AI chatbot is relying on sources that it finds easier to draw up-to-date information from. And then I used Gemini to search for a key information document — so a KID — for an investment product. Again, Gemini correctly explained how it all worked, but it didn't give a citation from the brand that it comes from. What it did do is offer up a landing page for an app on the Google Play Store. So whilst on the one hand Google, in their document, really went down the line of "search in the world of AI isn't different to search how it used to be", on the other hand Gemini is — in the tests I ran — biasing towards sources that are clearly easier for it to draw information from. I would argue that aggregator sites and news websites are probably structured in a way that just works well for AI agents, and the Google Play Store is an internal product, so they presumably plug into that database in a much deeper way than they do a typical fund management website.

So yeah, I think the conversation we've been having today about this concept of a two-track web is kind of at odds with the document Google published that we spoke about last week. But I would argue that everything we've seen, and the direction of travel, would suggest that a two-track web is the way things are moving, because it just makes sense for an AI agent.

Paul Wood (45:00.046)
Having a dedicated, structured setup makes perfect sense, and I can't see it not going that way. And I think brands need to start thinking about that.

Amelia (45:14.925)
Are you on the same page there, Russ?

Russell (45:19.008)
Yeah — one thing I noticed on The Economist as well is just the amount of advertising on there. It's going to be really interesting to see how that's managed. I'm sure it's quite heavily monetised, and almost certainly…

Russell (45:36.195)
…those adverts aren't going to be displaying in Gemini, even if there is a citation for The Economist. So how are these websites going to navigate that and survive? Some of them, perhaps. It must be a really big part of their business model — a massive part of their business model — and essentially they're just going to lose all of that if people aren't visiting their website.

Paul Wood (46:07.374)
Yeah, it's the same, isn't it? It's relevant for publishers because they need people to land on their site to see the adverts that pay for the product, or to subscribe and then pay for their content in a different way. And then it's the same for commercial brands — they need people to arrive there eventually and to trust them as a brand.

Paul Wood (46:35.918)
Because otherwise everything becomes a commodity, and price becomes the only differentiating factor. But as we all know, you get what you pay for a lot of the time, price isn't the only thing that matters, and we can't have a world where agents are just reducing everything down to the basics you need. So again, the two-track web seems to be a solution towards that, because it means that you can start to much more carefully manage what is surfaced. And that's important.

Amelia (47:15.689)
Big conversation. We probably asked more questions than we answered, but hopefully we've made it a bit clearer for people. Shall we play some Jargon Busters? Last week, you guys didn't do very well. I feel like up until now you've always been along the right lines, but last week with "churning", not so much. So let's see how you get on this week.

Paul Wood (47:27.736)
Yeah, let's see how we do.

Amelia (47:41.357)
If you've not caught this before, this is where I put the guys to the test with some industry terms, some jargon, to see how well they know them. So this week's term is: haircut.

Paul Wood (47:57.339)
I mean…

Russell (47:59.567)
I don't know — I'm going for one of those tomorrow, but I don't think it's related to that. I can have a go first, because I always seem to hold back and go second or third… although I wish I hadn't said that now. It must be taking a bit of something off the top.

Russell (48:29.066)
It's difficult to say, but trimming springs to mind as well. It must be in the context of… we seem to be on this investment track with a lot of these terms, so I think I'm going to stick with that. Perhaps it's trimming down your investments — the ones that aren't working — or taking some money off the top if you've made some money, and moving that to somewhere else? It feels like maybe one of those two, but I'm not sure which one.

Amelia (49:04.909)
What do you reckon, Paul?

Paul Wood (49:05.496)
Yeah, I agree. We always move towards the world of investments, and I can't escape that. But I feel like it's in that world, and a haircut is about removing excess, I suppose. So it's trimming off some of your investments — but it would be to make the portfolio look better afterwards. Which is like a haircut: you take off stuff you don't need, to look better.

Amelia (49:44.301)
Well, hopefully we'll see tomorrow, Russ, if it's better. Shall I tell you? So it's the percentage reduction in the value of an asset when used as collateral.

Amelia (49:59.297)
So sort of — I mean, sort of, we were talking about the trim — but yeah, specifically the percentage reduction in that value.

Paul Wood (50:06.348)
Okay. Yeah, so it's a result, rather than an active thing someone's doing. Yeah.

Amelia (50:12.361)
Exactly, exactly.

Russell (50:14.35)
I see, yeah. Well, maybe… have we got it half right, maybe?

Paul Wood (50:16.098)
Well, we'll find out at the end. Yeah.

Amelia (50:20.301)
Yeah, we'll give you half a point, but maybe you need to go and do a bit of revision before next time.

Russell (50:26.35)
Yeah, it'd be interesting to see, maybe after 10 or 20 episodes, how many we've actually got right.

Amelia (50:33.771)
Yeah, exactly — we might keep a little tally and we'll see. But thank you so much for listening; that is about it for this week. But we will be back next week. Russ, enjoy your haircut. See you next week!

Russell (50:46.936)
Thank you.

Paul Wood (50:47.217)
Cheers.

Russell (50:49.647)
Cheers, see ya. Bye.